UCLA have been publishing annual reports on diversity in Hollywood since 2014 (covering films from 2011 onwards), each with the core finding that diverse movies make more money.

Gadfly’s contention was that nobody ever seems to check the maths behind these claims. I realised I too had been reading versions of that headline for years but had never once checked it myself.

So I decided to take up the challenge and address the question: Do movies with diverse cast members make more money?

What I ended up uncovering was far more interesting than I thought it would be at first, and I learned a lot along the way about measuring subjective topics.

This is a long, complex piece, so I thought I’d help everyone with a TL;DR before we get into the detail:

1. Diversity is not one thing. “A diverse cast” can be measured in at least six defensible ways, with each method producing different findings for the same film.

2. The current headlines come from a shaky process. UCLA’s method is questionable and produces volatile results. The findings are more likely a result of the choice of method and the effect of small numbers than tracking a real connection.

3. Once you account for things like budget and release size, there is no academic consensus that increased cast diversity will lead to higher financial returns.

4. I tested 108 defensible methods across 2,600 films and 26 years. I found that almost all the methods produce no meaningful connection, and the median answer is a 2% dip. No method shows cast diversity meaningfully lifting box office. None shows audiences punishing it either.

5. The one pattern that survives everything tracks budgets, not audiences, with the most diverse casts being concentrated in the cheapest films. This is the problem the industry should be focusing on, rather than putting stock in the idea that cast diversity drives revenue.

Ok, let’s start at the top and work our way into the findings.

What does a ‘diverse cast’ mean?

“Diverse casts earn more” sounds like a single claim, but in fact it can mean two different things.

It can mean how much of the cast comes from minority backgrounds. Or it can mean how varied the cast is, i.e. how many different backgrounds are represented and how evenly.

I know they sound similar, but they are not, and the difference is more than pure pedantry. Look at the two fictional films below:

1. Movie one has eight actors, all from the same minority background, meaning that it is “100% minority actors” but has zero variety.

2. Movie two is mostly White but has lots of variety.

So which is the “more diverse” cast? Your answer depends entirely on which definition you use, and most headlines never say which one they mean.

Not to mention that “diversity” can cover background, ethnicity, gender, sexual orientation, disabilities, socio-economic background, etc, etc. I’m going to focus on ethnicity and gender for this article, but only to keep it focused, not because they are more important than others.

How do you turn eight people into one number?

I have laid out my methodology in detail at the end of this article, so I’m just going to gloss over it here for readability's sake. I looked at the top eight billed actors, plus the credited directors and writers, of the 100 highest-grossing films at the US box office 2000-25. For each person, I researched their publicly documented background from their own public statements, interviews, official biographies and reputable coverage, ending up with classifications for almost all.

We now need to turn that biographical data into statistics around diversity. This is much more subjective than you might think.

Here are six reasonable approaches one might take (names are my own):

• Minority Share. The share of the top eight cast from minority backgrounds. The simplest and most common yardstick.

• Weighted Share. The same share, but the top of the bill counts for more, on the logic that the lead sells more tickets than the 8th name.

• Minority Lead. Forget the ensemble: is the number one name on the poster a minority actor, yes or no?

• Mix Score. The chance that two randomly picked cast members come from different backgrounds. This measures variety rather than quantity.

• Group Count. How many different backgrounds the cast spans, and how evenly. A cast split across four groups scores higher than a cast drawn from one.

• Female Share. The share of the top eight cast who are women, kept separate from the race measures throughout.

Let’s take the example of Fast Five, the 2011 edition of the Fast and Furious franchise in which they drag a bank vault through the streets of Rio. Each method ranks it differently in my dataset, from being more diverse than 99% of movies down to more diverse than just 10% of them

So when someone tells you “diverse films earn more money”, your first response should be to ask “What do you mean by diverse?”

Now, let’s tackle the second half of the claim by considering what we mean by “make more money”.

How do you test the money question?

You cannot just compare the earnings of diverse and non-diverse films, because films differ in ways that swamp everything else. A $200m franchise sequel will outgross a $5m horror whatever their casts look like. So we should compare like with like - same genre, same year, franchise against franchise, animation against animation, and so on.

But sadly the complexities of the real world mean that it’s not that simple. Some things that look like fair controls are actually part of the subjectivity and might mean the person who's doing the counting has their hand on the scales.

The number of cinemas a film opens in is decided by the distributor, after seeing the finished film and deciding how hard to push it. If films with diverse casts were systematically given smaller releases, then “controlling” for release size would remove the very thing we are trying to measure. The same logic applies, more arguably, to budget, where the final budget is heavily linked to casting.

If at this point you’re getting exasperated at the futility of something that felt like a simple question, don’t worry! Because we cannot fairly pick just one methodology without adding subjectivity, I tested many.

But before we look at all the tests I ran, let's look at the study that prompted Hollywood Gadfly to reach out to me, the UCLA Hollywood Diversity Report.

What does the best-known study say?

The Hollywood Diversity Report has been published annually since 2014 by researchers at UCLA. It is a serious, long-running series, with its authors having studied this field for over a decade.

They use the "Minority Share" method I mentioned above, sorting films into six bands (e.g. under 11%, 11 to 20%, and so on up to over 50%), then comparing the median box office of each band.

The latest edition reports that films in the 41 to 50% band earned the highest median global box office. They note that:

This cast diversity interval happens to align with the BIPOC share of the U.S. population (45.2%).

And the report concludes:

As found throughout this report series, new evidence from 2025 further demonstrates that America’s increasingly diverse audiences prefer diverse content in theatrical film releases.

This tidy, optimistic, quotable finding has been the source of a decade of headlines on the topic.

There are two rather big problems with the method, and consequently with the conclusion.

Problem one: The banding method is unreliable

Their scale sorts films into bands 10 points wide. But a cast here is eight people. This means that one person represents 12.5 points. The smallest possible change in a cast is bigger than the width of the boxes the films are being sorted into.

Let's see what this means in practice. If we imagine a film with zero minority actors, then it will be in the under-11% band. If the 7th- and 8th-billed actors are recast to be minority actors, then the film jumps up to the 21%-30% band straight away.

This is the kind of casting I doubt most audience members notice, and yet the film has crossed two full bands.

It’s true that representation of all types matters, and that it's important for people to see characters like them on-screen. However, it feels a stretch to then extrapolate that to the point where you’re attributing the film’s overall performance to the two minor casting changes.

The banding method they use massively amplifies the effect of minor casting choices and creates data which implies there are big differences between films which are, for all practical purposes, not different.

Problem two: the results are very inconsistent.

If the sweet spot were a real feature of audience behaviour, you would expect it to show up year after year. The chart below tracks the winning band in each year of the report series’ own published tables, going back 15 film years

The winner has bounced across three different bands. The 21 to 30% band has won six times, the 31 to 40% band four times, and the 41 to 50% band five times.

The recent run of 41 to 50% wins, the one that aligns so pleasingly with the population share, describes the last couple of years of a series that spent a decade contradicting it.

And this is where I can add my new data, to go deeper than the UCLA published tables. I rebuilt my own 26-year version of their exercise using their method. I believe I have almost all the same classifications of individuals, but, as we have seen above, even one different classification can have outsized effects on the final data. So my charts are not an exact copy of theirs.

The chart below is entirely my data, and shows (a) how volatile their method is; and (b) that there hasn’t been a sweet spot for diversity this century.

So the UCLA data does not provide a robust, clearer answer than the headlines would have you believe. Instead, it reflects the often changing output of a particular method, heavily influenced by the property of small numbers.

What have other studies found?

I had built up a cache of 29 other studies and reports on the topic, to see what others have said on the issue (details and links in the Notes section).

16 found a positive link between diverse cast and financial success, seven found no effect, and six found negatives abroad or results that split by market and method.

But yet again, we need to look under the surface. Nine of those positive studies did not control for other factors, and the seven that did mostly used only the US box office.

Every study that compares raw grosses concludes that diversity pays. These are also typically the studies behind the headlines, such as the annual diversity reports, the talent agency indices, the consultancy studies with the billion-dollar opportunity numbers. But because they do not control for key factors, we cannot count them as rigorous tests of the claim.

Of the studies that do compare like with like:

• Several found genuine positives for Black-led casts in the US market.

• Seven found nothing at all.

• Six found negatives abroad, or answers that flip depending on the market, the measure or the statistical correction.

One finding was reported as a 149% revenue boost for films with multiple Black leads when its working paper made headlines. By the time peer review had finished, the published figure was revised to 47%. It is a genuinely good piece of work, and its published finding still stands as one of the strongest positives in the field. But I wanted to highlight it here to show how far a headline number can travel from the figure that survives review.

So we are STILL left with a murky picture. I’m sure by now you are screaming at the screen - Stephen, please just answer the damn question! Do diverse casts make more money at the box office, or not?!

Ok, ok, I’ll have a go.

Asking the question every possible way

Rather than just picking a single subjective methodology, I decided to test 108 of them. That’s a combination of:

• The six measures of cast composition we saw above.

• With and without budget as a control.

• Four different statistical models, including ones designed for the fact that box office is dominated by a few giant hits.

• Worldwide and domestic takings.

• Plus a version that only compares films from the same studio.

In total, 108 defensible ways of asking “is cast composition related to box office?”, every combination decided in advance so I could not fish for a result I liked or didn’t.

Each test produced a figure for the relationship between cast diversity and box office (worldwide and domestic were tested separately). A positive figure means the more diverse the cast, the more money it brought in, and a negative figure means the opposite.

I then plotted each result on the chart below.

86% of the answers land within 5% either way of zero, and the median answer is a 2% dip. It’s fair to say that there are few ways we can reasonably read the data to produce the finding that “more diverse casts earn more money”.

To give you a sense of how the tests work, here are five of the 108 answers, spanning the whole range:

• The most negative answer anywhere was an 11% dip, which came from Minority Share against worldwide takings, using median regression (which asks about the typical film rather than the average) and no budget control.

• A 4% dip was found for Minority Share against domestic takings, standard model, no budget control.

• A 2% dip, indistinguishable from zero, showed up for Female Share against domestic takings, median regression, no budget control.

• Almost exactly zero (+0.5%) was the outcome of Minority Share against domestic takings, using the skew-robust model with controlling for budget.

• The most positive answer was a 7% lift and came from Group Count against domestic takings, standard model, no budget control.

A motivated researcher could truthfully report “diverse casts are associated with lower box office” or “more varied casts are associated with higher box office” from this same dataset, just by choosing the measurement criteria.

Both would be technically true, and if they didn’t reveal the 107 other ways to crunch the same data, we might never question it.

What this all adds up to

So my answer to the question "do films with diverse casts make more money than those with less diverse casts?" is no. I ran every defensible method on 26 years of hit films, and couldn't find a version of the analysis where cast diversity meaningfully lifts box office.

And the most rigorous academic work in the field agrees with me on this. Once budget, marketing and release width are taken into account, they dominate everything, and the cast’s own effect shrinks towards nothing.

The diversity of the cast is just not a meaningful factor when assessing the financial performance of films en masse.

By the way, there is no honest reading of this data in which audiences punish diverse casts either. What this exercise really measures, in the end, is the measurers. The choice of yardstick moves the answer more than the casts move the money.

(Regular readers may remember that last year I tested the “Go Woke, Go Broke” claim across 10,524 films and found no general rule there either. This all reflects the same idea that movies are very complex products and that it should not be surprising that their financial outcomes are unlikely to be reducible to one factor).

The one effect that is there and needs more study

A McKinsey study reported that Black-led films receive budgets roughly 25% lower than comparable projects. And this is a finding I replicated in my study too.

So something is happening in Hollywood, but it’s taking place in studio greenlight meetings, not by audiences in line at the cinemas.

There is a genuine question about why the most diverse casts are concentrated in the cheapest films, which deserves better study than a diversity band.

Advocacy versus academia

I want to end on a question that kept coming up when I was researching this topic: what are reports like this actually for?

There are two very compelling reasons to take the time to do what UCLA have been doing so consistently over more than a decade:

1. Advocacy. I.e. making the case for a fairer industry, the way a guild, a trade body or an agent would. There’s no doubt there is a lot wrong with how the industry currently functions, and we need dedicated, passionate people to take the time to make us all better.

2. Science. I.e. finding out what is true, including when it is inconvenient. It’s important we know how things operate and, ideally, why. This allows everyone to have the same view of the world and to make realistic plans and accommodations in their own lives.

Both jobs are valid and needed. But they are different jobs.

Given their language and press releases, I doubt many would disagree that the Hollywood Diversity reports are doing the former. They are passionately advocating what they believe needs to change and why. But they do so under the guise of the latter, an independent study from a university, a research institute, endnotes and all.

The series describes its own purpose as exploring “relationships between diversity and the bottom line in the Hollywood entertainment industry”. Their conclusion is that the binned medians “demonstrate” that audiences “prefer diverse content”. The latest edition puts it like this:

During a tumultuous era in the business of Hollywood, one thing remains true, films with relatively diverse casts tend to do better at the global box office.

Nothing in their tables is false, and I have no reason to doubt their commitment to gather data or computational rigour.

But the method behind the headline, raw medians of 10-point bands on a single year of top films, with no like-for-like controls, is one of the very few ways of analysing this question that produces a result that big, in that direction.

The UCLA answer is not THE answer. It is one answer from the far end of a distribution of many. It is the answer that sparks the headlines they feel are needed to change the industry in the way they believe it must change.

Notes

I studied the 100 highest-grossing films at the US box office in each year from 2000 to 2025, using Domestic gross to define each year’s list. Box office and budget figures come from OpusData, Box Office Mojo and IMDb, converted to 2025 dollars throughout. This is deliberately a hits-only sample, because I was asked to respond to UCLA’s chosen focus. This means that any conclusions should not be automatically applied to the thousands of smaller releases each year - indie films operate very differently to the mainstream.

For each of the 8,854 individual people, I researched their background via public documentation, prioritising their own public statements, then interviews, official biographies and reputable coverage. Categories follow the convention used in the research this article engages with: White, Black, Asian, Latino, Middle Eastern and North African, Native American, and Multiracial, alongside gender. Each classification carries a confidence level and a note of its source.

When these sources could not reliably define the correct categories (i.e. 16.4% of people), I assigned working categories in two logged tiers. I reviewed photographs for 723 people alongside a second automated opinion and individually logged best guesses for the remainder. The blind double-coding, the 100 redraws and the two extreme scenarios described are how that guessing was priced; only findings that survive all of it are reported.

The UCLA team does not seem to have this problem (or at least they do not report it), as their tables carry a value for every person. Their methodology cites commercial casting databases (Gracenote Studio System, Luminate and IMDbPro) alongside their own research, and with a decade of accumulated records, institutional resources and the ability to contact people directly, they may genuinely have closed the gap that public records cannot close. The first edition of the report in 2014 allowed a person’s background to be “left blank if unascertainable”, but after that I can’t tell what happened to the unknowns.

I don’t like that I have any data gaps, so I dug a little deeper to understand if this could be skewing the results. I ran two sets of tests:

1. I re-researched a random sample of 242 people blind, twice over, to measure how often two honest attempts at the same person agree. For people with strong documentation, the answers agreed 94% of the time. For the middling cases, 74%. For the weakest, 57%. Those rates became probabilities, and I rebuilt the entire dataset 100 times, letting every uncertain person vary, and re-ran the headline models each time.

2. I ran the two most extreme scenarios possible, with every undocumented person assigned White, then every undocumented person assigned minority status. Whatever the truth is, it lives between those two brackets.

The result was that the error bars widen, the "more varied casts earn a bit more" signals dissolve into noise, and the small financing-shaped negative survives even the worst case. Most importantly, no assumption about the undocumented people, however extreme, produces a world in which cast diversity meaningfully lifts box office. So my handling of these 'unknowns' bothers me, but they do not change the conclusions.

For readability, I have been rather colloquial in the writing of the main article. For those who want the technical terms:

• The Mix Score is a bias-corrected Blau (Gini-Simpson) index

• Group Count is the exponential of Shannon entropy

• The models are ordinary least squares on logged revenue with standard errors clustered by distributor, Poisson pseudo-maximum likelihood on real dollars, and quantile regressions at the median and 90th percentile

• Controls are genre, creative type, franchise and sequel status, animation, release year and month, with production budget included in half the specifications by design

• The uncertainty pass pools the redraws using Rubin’s rules

• The extreme scenarios are Manski-style bounds. The 108 specifications were fixed before any results were computed.

Quotes and figures attributed to the Hollywood Diversity Report series are taken from the published PDF editions, 2014 to 2026, all of which I read for this project; the winning-band chart is built from their own published tables. I want to repeat the point that their team has studied this field far longer than I have, with resources and data access I do not have, and where our approaches differ it does not follow that mine is right.

This is a summary of the overall claims they have centred their research around each year:

[2014] Films with Relatively Diverse Casts Excel at the Box Office

[2015] Evidence from this report (and its predecessor in the series) shows clearly that America’s increasingly diverse audiences prefer diverse content created with the input of diverse talent. Diversity sells.

[2016] Consistent with the findings of earlier reports in this series, new evidence from 2013-14 suggests that America’s increasingly diverse audiences prefer diverse film and television content.

[2022] films with casts that were from 21% to 30% minority enjoyed the highest median global box office receipts, while films with casts that were less than 11% minority were the poorest performers.

[2023] New evidence from 2022 continues to support findings from earlier reports in this series suggesting that America’s increasingly diverse audiences prefer diverse film content in both theatrical and streaming releases.

[2024] New evidence from 2023 continues to support findings from earlier reports in this series suggesting that America’s increasingly diverse audiences prefer diverse content in theatrical film releases.

[2026] As found throughout this report series, new evidence from 2025 further demonstrates that America’s increasingly diverse audiences prefer diverse content in theatrical film releases.

I contacted all the authors and the appropriate UCLA Press liaison. I asked them to share their data and/or to have a chat with any of them (on or off the record) to better understand their data and methods, and to look for weaknesses in my work. Both were refused (albeit politely!).

The 29 studies in the field chart, classified by whether the analysis controlled for budget, genre and release factors.

• Raw comparisons (9, all concluding diversity pays): the UCLA Hollywood Diversity Report series (counted once), CAA's Motion Picture Diversity Index (2017), the CAA/shift7 female-led films study (2018), the Geena Davis Institute reports (2016 to 2019, counted once), ReFrame (2025), McKinsey (2021), Vocativ (2014), Slated (2016) and the BBC's return-on-investment analysis (2018).

• Controlled, positive (7): McKenzie (2010), Kuppuswamy and Younkin (2020), Zhong, Crema and Paserman (2024), Murray (2022), Nguyen (2023), Evans and colleagues (2024) and Valentowitsch (2023).

• Controlled, no effect (7): Lindner and colleagues (2015), FiveThirtyEight's controlled result (2014), Yang and colleagues (2020), USC Annenberg (2023), Roth and McAndrew (2026), Hesslevik and Ramm-Ericson (2024) and Ravi (2025).

• Controlled, negative abroad or split (6): USC Annenberg and ReFrame (2020), Pedace and colleagues (2019), Pattnaik and colleagues (2025), Giannetti and Chen (2023), Stroube and Waguespack (2024) and Fan (2021).

These classifications are mine, from each study's published method, with composite series counted once to avoid double-weighting.

As a side note, I do believe that advocacy research is perfectly capable of rigour. The most thorough controlled study in this field was commissioned by an advocacy group, and when its controls erased the famous claim that female-led films outperform, it published the null anyway. That study is "The Ticket to Inclusion" (2020), run by USC's Annenberg Inclusion Initiative and commissioned by ReFrame, the gender-equity coalition founded by the Sundance Institute and Women in Film. It modelled 1,200 top films with controls for production cost, marketing, distribution and genre, and found the gender of the lead made no significant difference to revenue in either direction, while also showing that female-led and underrepresented-led films were given smaller budgets and narrower releases. Passionate advocates for change are the main reasons we have anything nice in this world, and in many cases it's the ONLY thing the oppressed have on their side. My issue here is method, not mission.

Everything here is association, not causation. Casts are chosen, not randomly assigned, and no statistical method can fully separate a cast from all the other decisions a studio makes alongside it. The budget pattern is a description, not a claim about the 'true value' of a movie. Cheaper films may cast differently rather than diverse projects being given less money.

No work like this is truly objective, and what we count is shaped by who we are. I am a middle-aged CIS white man living in a Western capital city, classifying the public identities of thousands of people I have never met. I have tried to let people’s own public statements lead wherever they exist, to log every judgement call, and to report what breaks my conclusions as prominently as what supports them. Despite trying to be aware and sensitive to all points and lived experience, it’s more than possible I have made mistakes or assumptions. I would be very keen to learn of any, and will be delighted to correct them. Please do reach out to help me in this regard.

Finally, I would love for this to be part of a big conversation. I don’t seek to police what people think or say, but I would ask two things:

1. Be nice. We’re all humans trying to make sense of the same confusing world. Treat others with respect and grace, and I’ll back you if others fail to afford you the same courtesies.

2. Focus. This article’s comments section is not where we are going to settle big questions like “What is a woman?” or “Is [x] racism?” The internet already has myriad places for everyone to shout their views. Let’s keep this one about the facts and methods listed above.

Epilogue - the world of AI-powered research

This project would have been impossible for one person until very recently. The person-by-person research ran on AI models, with research passes with live web search to find and read interviews and biographies (with every claim carrying its source), a separate auditing pass by a second model checking that each classification matched its own evidence, batch photo checks against my own manual review of around 650 photographs, and in-session research agents for the hardest cases.

The classification machinery ran over a handful of days and cost roughly $480. So a HUGE thank you to my paying subscribers who make this kind of research possible.

A decade ago this would have been a research team working for months. The Hollywood Diversity Report is what that fuller version looks like. It is a standing university research initiative with dedicated staff, commercial data subscriptions and, per its own acknowledgement pages, financial support from UCLA's Division of Social Sciences alongside sponsors including Netflix, the Golden Globe Foundation, the Latino Film Institute, Disney and Hulu. The report receives public money, too, in the form of a $250,000 allocation in California's 2021-22 state budget, and a $1.5 million state investment announced in 2022.

That money buys real things my $480 cannot, such as institutional memory, professional researchers, direct industry access and the ability to do this every single year.

When I started my journey into film data 15 years ago, I was 100% hand-coding and using Excel spreadsheets. As the tools have expanded, I’ve been able to operate at larger scales and in ever-increasing detail. That said, if you would have told me even a few years ago that a project of this depth could be done in just a few days by myself, I would have thought it fanciful.

While this is genuinely thrilling, it comes with limits worth naming:

1. An AI pipeline applies my judgement calls at scale, including any I got wrong.

2. It lacks the domain memory of researchers who have lived in this field for years.

3. Its consistency is only as good as the auditing conducted on it.

So I’m not saying ‘one person with AI beats an established research team’, but just that it is amazing how close one can get alone.

Put another way, I looked at over a million frames of Matt Damon’s on-screen performances to understand the power of his acting and his enduring charm.

Along the way, I assessed how much it’s cost for humanity to keep trusting Matt Damon (Hint: it's quadrillions of dollars).

I’ll be uploading new video essays every couple of weeks, so make sure you’re subscribed on YouTube to get notified when another drops (also, subscribing will really help me out with the Algorithm Gods, so it would mean a lot to me - thank you!).

https://www.youtube.com/@TheFilmDataScientist

And to help explain the channel, I also created a short channel trailer:

Its hand-researched data is the highest quality, tracking box office grosses, budgets, home video and streaming across more than 78,000 films and 236,000 people. It gets north of eight million visitors a year, and is treated as THE definitive authority by journalists, academics, filmmakers, prediction markets, and even Guinness World Records.

And it was this GOAT status which caused the catastrophic events of March this year.

On the 5th March 2026, TheNumbers.com website vanished.

The site was down for over a week, without explanation. A week later, it resurfaced at a fraction of its former size. Gone were the historical charts, the individual movie pages, and even the much-loved Report Builder.

With only a generic “we’re rebuilding, please bear with us” message to go on, the internet responded as it always does - with confusion, anger, and conspiracy theories.
One Reddit theory even suggested it was a deliberate rug pull designed to cripple the free site to push people towards paid products.

Three months on, I spoke at length with Bruce Nash, founder and CEO of The Numbers, about what happened. He describes quite an unpleasant and eventful experience:

We got a lot of angry emails from people who are like, 'Where's this page that you used to have and you don't have anymore?'

Within his tale are a number of things that should worry anyone who runs, relies on, or simply appreciates the internet.

First, some background

On Friday 17 October 1997, mathematician and former IBM software developer Bruce Nash launched a Geocities site that tracked 300 films.

Bruce described the launch in a 20th anniversary essay (which now survives only in the Internet Archive, for reasons that will become clear):

I hit a button in an Access database, uploaded some HTML pages to Geocities, and made a brief announcement on the Hollywood Stock Exchange message boards to let people know that I was starting to analyze box office for films to help them pick MovieStocks to trade on HSX.

From those humble beginnings, Bruce and the team he built around the site turned The Numbers into the film industry's most reliable financial source.

At the start of 2026, the database tracked 78,396 movies, 178,375 theatrical release records, and 236,176 people.

The robots arrive

During its lifetime, the challenges The Numbers has faced have changed immensely. For its first quarter century or so, the traffic was manageable and mostly polite. As Bruce puts it:

Pre-AI, we got human traffic, mostly well-behaved search engine crawlers, and a few people crawling the site for personal projects. If someone got too greedy, we could spot them and block them.

Over the past couple of years, website owners the world over have seen their web traffic change. What was initially only people browsing gave way to an ever-increasing number of bots. By 2024, automated traffic had surpassed human traffic, and just last month, Cloudflare announced that bots had reached 57.5% of web page requests.

The Numbers felt this shift in two distinct waves. The first started around 2024:

We saw a big increase in crawls as AI training joined the search engine crawlers. The AI crawlers are generally less well-behaved than the search engines, which increased the management tasks for us to keep the site running smoothly.

And the second wave was stronger and more damaging:

Around December 2025, we saw another big spike in traffic which I attribute to agentic AI: a combination of AI agents that scrape sites in response to prompts, and people being able to write agents that scrape sites.

Like every data-rich site, by early 2026 The Numbers was being hammered hard by AI bots scraping its pages over and over at an industrial scale. Bruce says that only 10% of their traffic is from humans browsing the site, with the rest coming from AI bots and automated traffic.

Websites try to adapt to the new robots

This put enormous strain on the site, but Bruce and his team were able to take measures to mitigate the worst of it. One of the cleverest was talking to the robots in their own language:

There’s stuff on the site which is designed for an LLM to read, so that it can tell somebody ‘here’s how you licence the data’ rather than ‘here’s how you scrape the website’. It’s had a huge effect. We’re now getting probably ten times the volume of licensing enquiries.

But mitigation is not the same as escape. From December through early March, the team struggled to keep the site alive under the load. Bruce estimates that:

Around 90% of our time was spent keeping the existing site running while we spent our spare moments working on a new and improved system.

The problem was compounded by the site’s age: thirty years old, with approximately 160,000 source files serving around 2 million pages.

Then, in the early hours of Thursday 5 March, the servers collapsed.

The team scrambled to understand what had happened, initially assuming it was the sheer weight of AI traffic. It seems AI was to blame... but possibly not only in the way they first thought.

Buried in the flood of agentic traffic, the site’s logs showed something more pointed than scraping. As Bruce describes it:

Some of these used the site using legitimate URLs, others were looking for back doors, most likely so they could get to the data before it appeared on the site, or to manipulate the data presented to users.

On the advice of a friend who works in cybersecurity, the old server stayed off. For good. Restoring the backups and nursing the thirty-year-old site back online would have meant defending 160,000 legacy files against attackers who had spent months probing them.

The team rushed up a skeleton version of the website on new infrastructure, which could at least keep delivering the latest box office figures while they took stock of what had happened and what to do next. It went live on Friday 13 March.

Who would want private access to a box office website?

At first glance, The Numbers may not seem like an obvious target. It doesn’t collect credit card information, and there is no juicy customer data to flip on the dark web. It is a small, independent company that publishes how much money movies make.

How could someone expect to make money purely from having private access to their site?

In case you haven’t guessed it yet, it’s linked to prediction markets.

Polymarket runs weekly markets on opening weekends, and names The Numbers as the ultimate source of truth:

The ‘Daily Box Office Performance’ figures found on the ‘Box Office’ tab on this movie’s The Numbers page will be used to resolve this market once the values for the 3-day opening weekend are final.

The sums on any single weekend market are modest by financial-market standards, typically in the tens to hundreds of thousands of dollars, with a couple of million dollars across live box office markets at any given time.

If you could see The Numbers data before everyone else, every single week, you would have a significant edge over all the other traders - learning the answers slightly ahead of publication would allow you to front-run the trades.

In a situation like this, it is hard to know for certain what happened. We know that the logs showed months of automated probing and scraping of the site, but what finally brought the site down, and who did it, remains an open question.

But the theory that someone used AI to develop an advantage in a prediction market is entirely plausible. The Numbers experience shows us that:

1. We now live in a world where a movie statistics website is worth hacking because prediction markets empower anyone to turn almost any data into money.

2. Hacking websites is now something anyone can do with a cheap AI subscription.

3. The web, as we have it, is incredibly fragile in the face of large-scale swarms of agentic AI bots.

How hard is hacking these days, anyway?

In November 2025, Anthropic (the AI lab behind Claude) published a report on what it called the first documented AI-orchestrated cyber espionage campaign. A state-sponsored group had used its coding tool to attack roughly 30 organisations, with the AI performing 80% to 90% of the work and humans stepping in at only 4 to 6 decision points per campaign.

Anthropic’s own conclusion was:

The barriers to performing sophisticated cyberattacks have dropped substantially, and we predict that they’ll continue to do so.

In an earlier threat report, Anthropic were even clearer:

Criminals with few technical skills are using AI to conduct complex operations, such as developing ransomware, that would previously have required years of training.

Meanwhile, an autonomous AI penetration tester called XBOW reached number one on HackerOne’s US leaderboard, the ranking of the people (formerly all people) who find security holes in real companies for bounties, submitting nearly 1,060 vulnerabilities along the way.

Getting access to a thirty-year-old website with 160,000 legacy files is exactly the kind of known-flaw surface that AI tools have made cheap to probe. The expertise barrier that once protected small sites from all but the most determined attackers has largely evaporated.

What now for The Numbers?

Bruce and his team were relatively lucky. Despite having their entire site knocked out overnight, they were able to keep going. The Numbers has always been free to use, and the site hasn’t relied heavily on advertising for the past few years, so the outage didn’t destroy an income stream they depended on.

Their core business is tied to selling bulk data through the OpusData service, producing comp analysis reports for filmmakers and investors, and publishing the Business Report - all of which were unaffected by the public site going down.

But they do need to build an entirely new website, from scratch, to host those 78,396 movies, 178,375 release records and 236,176 people. Restoring the site from a backup wasn’t an option, as Bruce points out:

It was really clear that we couldn’t just put that server up again, because it would inevitably be brought down again, possibly within minutes.

That is why the site came back bare-bones in mid-March, and why features are returning gradually rather than all at once.

Right now, the team is having to reconsider what a public website even means in 2026. Bruce’s analysis is that The Numbers used to serve two audiences (human beings and search engines) and now serves roughly six: humans, search engines, LLM training runs, prompt-based AI traffic, agentic AI, and prediction market punters. Each has different needs and a different traffic profile. As he puts it:

We’ve gone from a world where running a web site meant focusing on three things (content, ads, and SEO) to about eight to ten different factors that go into every design decision.

The goal, he says, is to support all six audiences, with new OpusData services and online features for Business Report subscribers, and, importantly, to help regular human users of the site regain the data it has always provided, some of it in new and improved form.

How bad could bot scraping really be?

Pretty bad, tbh. Enough that site owners such as Bruce have to question the value of something that will take so much time and money to build and defend.

Cloudflare, which protects a huge share of the world’s websites, publishes data on how many pages each AI platform crawls for every one visitor it sends back to the websites it crawled.

Google crawls about five pages for every visitor it sends you. OpenAI crawls over 1,000. Anthropic crawls over 38,000 pages for every single visitor it refers.

Note that the scale is logarithmic, i.e. each step along the bottom is ten times bigger than the last, because otherwise the differences are quite literally too large for me to include on one chart.

For the history of the internet to date, the principle of the open web was that, in return for letting the search engine robots read your site, they would send you readers. But now, that trade no longer applies. The number of robots has exploded, and they no longer send anyone back.

When this firehose is aimed at a small site, it can inflate the bandwidth bill and possibly even take down an entire site. Sites which can relate to Bruce’s experience include:

• Read the Docs, a non-profit that hosts documentation for open-source software, who watched a single crawler download 73 terabytes of zipped HTML in one month, costing it over $5,000 in bandwidth.

• iFixit, the repair-guide database, logged a million hits from Anthropic’s crawler in a single day.

• Triplegangers, a seven-person company selling 3D scans, was knocked offline during business hours by OpenAI’s bot, in what its CEO described as “basically a DDoS attack”. The founder of code-hosting service SourceHut reported spending “anywhere from 20-100% of my time in any given week” fighting AI crawlers, with “dozens of brief outages per week”.

• The editor of Linux news site LWN described crawler traffic from “literally millions of IP addresses” and concluded: “it is a distributed denial-of-service attack”.

• When the GNOME open-source project measured its traffic, roughly 97% turned out to be bots.

• A university library banned 16,000 IP addresses in 48 hours to keep its catalogue online.

The Wikimedia Foundation, which runs Wikipedia, reported in April 2025 that bots account for about 35% of its pageviews but at least 65% of its most expensive traffic, because crawlers bulk-read obscure pages that human readers rarely touch.

Six months later came the other half of the squeeze, when Wikipedia’s human pageviews fell roughly 8% year on year, as people increasingly get Wikipedia’s knowledge from AI summaries without ever visiting Wikipedia. The machines are taking both the content and the readers at an industrial scale, too.

Testing it in public

AI tools are some of the most powerful and destructive things humans have ever created. And they are being effectively tested by the public in real time in the real world. When the Manhattan Project was trying to work out the power of their atomic tech, they did not do so by sending everyone the specs each morning and seeing which houses blew up.

The world we have built thus far is so incredibly ill-prepared for the power and scale of the AI models we all have access to.

I don’t wish for this to sound like a one-sided anti-AI fear campaign. There is a lot to like about AI and what it can do for the human race. But we do need to consider the world we’re currently stepping into.

What breaks first are the things built for the old internet. The open web was built on assumptions such as that visitors are mostly human, that traffic roughly tracks readership, and that the cost of serving your site is related to the value you get from serving it. Every one of those assumptions is now out of date.

A year ago, Cloudflare launched pay-per-crawl, letting sites charge AI crawlers per page. Last week, it went further, announcing a pay-per-use model in which publishers get paid when their content actually appears in an AI answer, and declaring that, from 15 September, its customers’ ad-supported pages will block unpaid “mixed-use” crawlers by default.

Whether any of this works depends on whether the AI companies play along rather than route around it. But as Bruce put it to me, somebody has to try.

Epilogue

Let’s look beyond the specifics for a moment and consider what happened here.

A beloved, useful, free website, run carefully by a competent, honest person for nearly thirty years, was crushed between two features of the new AI economy.

Unsustainable machine traffic hammered it from above, and in all likelihood a financially motivated intruder, operating in a world where breaking into websites has never been easier, took it down.

Bruce’s business and livelihood survived only because the website was not the whole business.

Others have not been so fortunate. Just last week, ZEGO, a German textile firm that had been in business for 37 years, filed for insolvency after a single cyberattack in March shut down its production for six weeks. Unlike The Numbers, they had no other business to fall back on.

The web is full of independent archives, hobby databases, local news sites, forums, reference works. Decades of accumulated human effort, running on old code, maintained by small teams or single individuals, quietly holding up far more of our shared knowledge than anyone acknowledges.

The Numbers is coming back, better built than before. I would encourage you to keep using it, keep supporting it, and, if you are one of the many people who emailed Bruce in fury about a missing page, perhaps send a kinder one now you know why it was missing.

Last month I was asked to return, this time to look into the future and consider what the German cinema industry might want to do.

(I am aware that an Englishman going over to Germany and saying “This is what you must do” is a tough gig! But the organisers were very supportive, the audience highly engaged, and overall I feel my suggestions were taken in the spirit they were intended. Zum Glück!)

Today’s article is a précis of the ideas, partly as a record and because many of the lessons will apply to countries worldwide.

A quick recap: the two German cinemas

Last year I showed that German cinema has two distinct identities depending on who is watching. At home, the box office belongs to comedies and family films. They are warm, silly, hugely popular and almost never seen outside German-speaking countries. Abroad, the German films that succeed are serious dramas about moral conflict, psychological weight and difficult history.

Comedies and family films have taken 73% of the domestic box office for German-language films since 2000. In North America, those same kinds of films account for 0.9% of German-language cinema's earnings.

Two cinemas, in other words - one that stays home and one that travels.

How the world talks about German film

To add to last year’s findings, I also gathered 25 years of articles from Variety, The Hollywood Reporter, Screen and Deadline, around 1.3 million pieces in total, and identified roughly 49,000 that concern German film.

Then I counted the language used about seven major film nations: how often each country’s cinema is described in terms of success and breakouts, versus crisis and decline.

Every country gets more crisis talk than success talk. That is simply how trade journalism works. But the size of the gap varies a lot, and Germany’s is the widest of any nation I studied.

To be clear, this doesn’t show that German cinema is in decline. Instead, it shows how German cinema is framed in the press, which shapes whether an audience, a commissioner, or a distributor gives a film a chance.

So German cinema carries a reputation problem it hasn’t earned. Which turns out to be less of a burden than you’d think, because reputations like this can flip.

The world falls for a country’s cinema in waves

International attention doesn’t spread evenly across the world’s film industries. It moves in waves.

• France had the New Wave, which reset the country’s cinematic reputation for a generation.

• City of God put Brazilian film on the world map.

• Britain has had at least two “British cinema is back” moments, one around Chariots of Fire and one around Four Weddings and a Funeral.

• Amores Perros and Y Tu Mamá También launched a new generation of Mexican cinema.

• Dogme 95 made Denmark, a country of six million people, a global brand.

• Crouching Tiger, Hidden Dragon opened Western audiences to Chinese-language cinema.

What makes these waves so valuable is that a single breakout film gets audiences fired up to ask what else there is, and when there’s a shelf of films waiting, many films benefit from this new halo.

The Parasite effect

Earlier this year I published a study of what happens to a country’s films after it wins the Best International Feature Oscar. I tracked the entire back catalogue of each winning nation, 12,267 films across nine countries, in the year before and after the ceremony, using Wikipedia attention and almost ten million written audience reviews.

After Parasite won, attention to Korea’s entire back catalogue rose by around a quarter relative to its peers, and the review data showed an even bigger jump. The lift lasted for years. And it wasn’t just people discovering Bong Joon-ho: the whole of Korean cinema rose by about the same amount.

Notice that Germany is on that chart too. All Quiet on the Western Front won four Oscars and moved the rest of German cinema by a modest 8%, with some of that due to Netflix’s global release rather than the award.

It could be argued that had that film received a proper cinematic release (as opposed to being a streaming title), the effect would have been larger and more profound.

The broader point is that a single film can really lift an entire national cinema. It has happened, measurably, within the last six years. It’s not something we can conjure with certainty, but we can describe the conditions which make it most likely to happen.

Three conditions at once

Parasite did three things simultaneously.

1. It reached the world.

2. It was genuinely popular, a tense, funny thriller rather than a critics-only piece.

3. And it was completely, unmistakably Korean, with no attempt to sand that off for export.

Germany’s version of that film would need the same three things: international, mainstream and deeply German. Here’s what the data says about each.

1. It needs to reach the world

The reflex here is that German films don’t travel. That used to be a reasonable bet, but streaming has changed the landscape.

Dear Child, a German-language thriller, reached number one on Netflix in 92 countries and was the second most-watched title on the entire platform in the second half of 2023, behind only Korea’s The Glory. Around 90% of the audience for Dark was outside Germany, in German, with subtitles. All Quiet on the Western Front surpassed 150 million hours viewed, and its viewership tripled when the Oscar nominations were announced.

And when the films themselves don’t travel, the stories do. Wikipedia maintains a category of remakes of German films that runs to over 150 titles, with no comparable list flowing the other way. Fack ju Göhte, a German school comedy that never stood a chance in English-language cinemas, became No Manches Frida, a Mexican franchise worth roughly $50m. Paramount has held the remake rights to Toni Erdmann since 2017, at one point with Jack Nicholson attached. The BBC remade the cult NDR comedy Der Tatortreiniger as The Cleaner, written by and starring Greg Davies.

2. It needs to be properly mainstream

While the international community of cinephiles might think of prestige, dour dramas when they think of German cinema, within Germany it’s another matter. Multiplexes are regularly packed with home-grown, mainstream hits.

Der Schuh des Manitu sold 11.7 million tickets; six German films have surpassed seven million admissions; and only last year, Das Kanu des Manitu sold around five million tickets.

In a symbiotic loop of film and filmmakers, Germany has a wealth of experienced mainstream talent to draw from.

• Michael “Bully” Herbig holds the all-time number-one and number-two spots.

• Bora Dağtekin has released six films, and each has surpassed 2 million admissions.

• Elyas M’Barek anchored a trilogy that sold over 21 million tickets.

• Til Schweiger topped the annual box office repeatedly as both director and star.

While these will be household names to most Germans, they barely register outside of Germany. In my trade press dataset, the most-covered German film-makers are Wenders, Tykwer, Akin, Herzog and Petzold. Wonderful film-makers all, but they are the festival names.

The directors who actually pack German cinemas are pretty much invisible abroad. The world’s press covers the German cinema of prizes and misses the German cinema of crowds. Hence, why one film alone could awaken international audiences to Germany’s goldmine.

3. It needs to be undeniably German

The final part of the puzzle is to ensure that this breakout film is undeniably German.

Parasite was full of Korean specifics, and that specificity is what made the world curious. The equivalent German move is to lean into what the world already admires about Germany, and the polling on that is remarkably strong:

• Germany holds the number-one “Made in” label on Earth, ranking first among 49 countries in the Made-in-Country-Index.

• 48% of global consumers trust German products “a lot”, the highest of any country YouGov tested.

• It is the second-most admired nation brand in the world and the first in Europe, having held the global top spot from 2017 to 2022.

• 73% of people say German cars are the best in the world.

• Gallup has rated Germany’s leadership the most approved of any major power for six years running.

• The Brothers Grimm tales are inscribed in UNESCO’s Memory of the World and translated into more than 160 languages. The world’s most famous fairy-tale castle is Bavarian, and Disney’s logo owes it a debt.

Engineering, precision, order, fairy tales, music, forests, castles. They may, at first blush, feel like clichés, but they are, in fact, the best route to promoting Germany and German culture. They are pre-sold story worlds the global audience already trusts, the same way The King’s Speech and Downton Abbey sold the world exactly what it expected of Britain.

Two out of three, every time

As you can see, Germany already excels at all the other things our mythical new breakthrough needs to do. Many films have hit two out of the three requirements.

But I believe that all three are needed. With just two, you get the following:

• Anonymous. International and mainstream, but not German. Films built to travel with the Germany taken out: shot in English, or a minority stake in someone else’s production. They reach people, but nothing of Germany travels with them.

• Prestige festival dramas. International and German, but not mainstream. Toni Erdmann, The Lives of Others, All Quiet. Unmistakably German and they travel, but to festivals and critics rather than a mass audience.

• Home hits. Mainstream and German, but not international. The comedies and family films German audiences adore, which the world never sees.

Three overlaps, all of them real, familiar, repeatedly successful kinds of German film. And a centre, where all three meet, that is empty.

Germany has never made a film that is international, mainstream, and unmistakably German at the same time.

What the film might look like

Here is a checklist for factors which are likely to improve the film’s path to success:

• Fun first. Parasite was a thriller before it was anything else. Films break out by being entertaining, not by being important.

• Unmistakably German. The specificity is the point, not something to hide or overcome.

• A universal feeling underneath. A local layer on the surface, and something everyone knows beneath it. War, family, survival, love, grief, betrayal, revenge, justice, ambition, freedom, guilt, sacrifice, belonging, growing up, growing old. Or the brighter register: joy, friendship, first love, laughter, hope, wonder, homecoming, a dream coming true.

• A genre to carry it. Genre is the passport that gets a local story across a border. Thrillers and science fiction travel best; comedy travels worst, which is precisely why Germany’s biggest domestic genre has stayed home.

• A bench behind it. When curious viewers come looking for more, there has to be somewhere to go. Korea had Oldboy, Memories of Murder and Train to Busan waiting. Germany’s shelf runs from Metropolis and Nosferatu to Fassbinder, Wenders and Dark. Few countries have a deeper one.

• Built for streaming. The modern breakout may be a series, and may never see a cinema. That is where the world now discovers non-English stories.

The empty middle

Germany makes films that travel, films that fill cinemas, and films that are gloriously, specifically German. It's time they did all three and aimed for what Parasite did, hopefully lifting the whole national cinema.

Notes

The trade press analysis covers around 1.3 million articles from Variety, The Hollywood Reporter, Screen and Deadline, 2000 to 2025, of which roughly 49,000 concern German film. The success/crisis comparison counts articles using breakout and success vocabulary versus crisis and decline vocabulary, applied identically across all seven countries. It measures the language of coverage, not the performance of the industries, and word-counting of this kind is a blunt instrument at the level of any single article. The cross-country comparison is more robust than the absolute percentages.

The Oscar halo analysis is described in full, with its own methodology notes, in my April piece on Parasite. In brief: winners of Best International Feature; each country’s back catalogue (excluding the winner) tracked for 12 months either side of the ceremony; controlled against the average of the other tracked countries, using Wikipedia pageviews and around 10 million dated audience reviews.

German admissions figures come from the FFA (the German Federal Film Board) and InsideKino and are measured by tickets sold rather than revenue, so that films from different decades can be compared fairly. The figure for Otto – Der Film excludes East German admissions; including them would lift it to roughly 14.5 million, putting it at the top of the list.

Netflix reach figures are drawn from Netflix’s published Top 10 and engagement reports, as well as third-party trackers (FlixPatrol, What’s on Netflix). Netflix self-reports these numbers and changed its headline metric in late 2021, so treat them as directional rather than audited.

I’ll be uploading new video essays every couple of weeks throughout 2026, so make sure you’re subscribed on YouTube to get notified when another drops (also, subscribing will really help me out with the Algorithm Gods, so it would mean a lot to me - thank you!).

https://www.youtube.com/@TheFilmDataScientist

And to help explain the channel, I also created a short channel trailer:

So I went looking for real-world examples of indie movies where the people behind them had published how much they spent on marketing, and why.

I ended up with 57 case files, built from interviews, Substack post-mortems, podcast transcripts and the small number of organisations that publish full release ledgers.

The four big lessons

Across the 57 cases I looked at for this piece, four key lessons kept occurring to me.

1. You need to seriously budget for marketing. The filmmakers who fared best (Hundreds of Beavers, Columbus, Thunder Road) had already raised money before they needed it. The ones who suffered most ran out mid-release, such as the team behind the Oscar-shortlisted documentary UNION (2024), who published their campaign sums ($800,000 planned for an 18-month campaign, $400,000 actually raised against a $1.2m production budget), wrote that their theatrical “could have kept going to even more success, but ran out of money to fund our modest social media ad budget”.

   

2. Theatrical almost never recoups directly. “The theatrical release did not come close to recouping” (You, Me & Her). “Was not profitable. P&A is truly costly” is how Lisa Hurwitz put it in her account of self-distributing the documentary The Automat (2022) into 100+ cinemas, where her cut of the $255,000 box office “nearly covered” the $100,000 P&A. The 306 Hollywood team knew from the outset it was “nearly impossible to make a lot of money from our theatrical”. “Expect to lose money” (Bomb It). What theatrical bought, in many cases, was legitimacy that led to later deals and revenue.

3. The biggest line item is not in the budget. The Beavers team valued their unpaid social labour at $150,000–$250,000. The UNION team “primarily worked for free throughout 2024”. Hammond counted 18 months as “a part-time marketer and DVD shipper” against his Sriracha profits. Warshawski toured Big Sonia for two years and reported the damage to her health. When you read about a film doing miraculously well on a low- or no-budget, look for the unpaid labour that made it happen.

4. The numbers are shaky, hidden, and rarely complete. Very few filmmakers are kind enough to share their costs, and there are no effective tools to make a reasonable guess from the outside. Even when there are publicly agreed numbers, they often don’t tell the full story. For example, the trades said Eno grossed $300k, but its actual US revenue was $730k when the filmmaker-led one-off events are included. Burn grossed more than $1 million with no distributor at all, almost none of it through channels the trades track. The documentary Age of Champions (2011) built more than $1.5 million from screening kits, speaking fees and sponsorships, almost entirely outside every reporting system in the industry.

Let’s start with a few excellent full case studies and after that go into the individual line items.

“Hundreds of Beavers”

The most recent and most complete case study is Hundreds of Beavers (2024). The film cost $150,000 to make, but when producer Kurt Ravenwood was offered an MG (minimum guarantee) of just $30k, he declined and opted for self-distribution.

The total release cost was “approximately $135,000”, meaning they spent almost as much releasing the film as making it. Kurt opened his books to Jon Reiss over a 90-minute video I highly suggest watching.

In the end, the movie grossed just over $1.5m in cinemas worldwide, plus roughly $500,000 on digital platforms and $250,000 in merchandise sales (according to Wikipedia).

This is also a good example of the value of unpaid labour. The team ran their own social media for two years at no cash cost.

Large scale social media engagement was done by producers and SRH at no cost - this would have normally cost between $150K-$250K if they were paid for it.

The unpaid labour was worth more than the entire cash release budget.

“You, Me & Her”

You, Me & Her (2025) is the same shape at a higher price point, as seen in Distribution Advocates’ case study. Selina Ringel and Dan Levy Dagerman made their rom-com for $500,000, and then raised an additional $350,000 for theatrical marketing and distribution.

That paid for a national release across more than 250 cinemas, a $100,000 marketing co-op with exhibitors, 250 influencer boxes, and digital billboards in Times Square and on Sunset Boulevard (a bargain at $3,000, “affordable due to timing and inventory”).

The theatrical release did not come close to recouping the cost to market and distribute the film. But the visibility we generated would ultimately help to secure a STARZ SVOD deal and in-flight licensing agreements with American Airlines, Qantas, and Jetstar through Encore Inflight.

Theatrical has traditionally been seen as a loss leader for many movies. The splashy, big release increases awareness and cements the movie's status as a ‘proper movie’ in the public eye. The filmmakers can then (hopefully) reap the rewards of their all-important “third window” (such as VHS, DVD, or TV), where costs were much lower.

“Sriracha”

The most complete micro-budget P&L belongs to Griffin Hammond’s Sriracha (2013), a documentary made for $12,728 that then spent $27,807 on marketing and distribution, more than twice its production cost.

Hammond published everything at 18 months and again at 36 months, revealing $136,813 of revenue, $78,186 of profit, 42 festival submissions at an average of $42 each, and the sobering discovery that ad-supported YouTube paid him $341 for 127,485 views.

The pre-pandemic Sundance case studies

The Sundance Creative Distribution case studies remain the most granular release breakdowns, and although they describe a 2017–18 world, their category splits are still the best benchmark we have at scale.

Columbus (2017) turned down a $150,000 all-rights offer against a $700,000 budget and self-released with a final P&A budget of $189,032. PR took $51,328 (close to 30% of the P&A budget), theatrical costs $50,673, and social media ads $24,000, split $8,000 theatrical, $12,000 iTunes, $4,000 boosting posts. The film grossed $1,014,533 in domestic cinemas, and per the case study:

More than twice as much money came back to the filmmakers in the self-release model as compared with a hypothetical traditional distribution arrangement.

Thunder Road (2018) did it smaller, with a distribution and marketing budget “just under $70,000” on a $200,000 film, after rejecting a $125,000 minimum guarantee. By January 2019 it had netted about $510,000 across all platforms. The single biggest revenue line was, unusually, airlines, at $166,396, at over ten times the theatrical take ($12,345).

And at the micro end, distributor Matt Grady of Factory 25 published a complete ledger for Christmas, Again (2014) on Sundance’s blog: PR $10,000, marketing $2,500, digital deliverables $3,400, an LA four-wall $5,000, and so on. $23,550 costs against combined digital and theatrical income of $20,004. They noted that they expect a Netflix deal later, see it as “the tipping point financially”.

Assorted other case studies

A few others worth a read:

• SADIE (2018): $650,000 budget; declined a 25%-of-budget MG; set aside $50,000 for marketing plus a $100,000 non-recoupable Amazon Festival Stars bonus; did not recoup. Producer Lacey Leavitt’s honest write-up is at Dear Producer.

• First Girl I Loved (2016): P&A “approached six figures”, including roughly $35,000 of social ads; approached three times its best Sundance offer. Producer Ross Putman wrote it up for Sundance.

• The Discoverers (2014): raised $100,000 on Kickstarter purely for distribution; split 31% exhibition, 37% publicity/marketing/creative, 32% advertising. The filmmakers’ account is at IndieWire.

• Red State (2011): Kevin Smith toured his $4m thriller as 15 one-night events with “less than $500 in paid advertising”, grossing $851,832 on tour (IndieWire, IGN). His stated alternative: “minimum $10 million on P&A” for a traditional release.

• Marcus Markou’s Papadopoulos & Sons (2012). Markou self-financed the £825,222 film and self-released it in the UK on a P&A budget of £35,525, itemised down to £1,000 for the BBFC certificate and £8,000 of Facebook ads targeted at people with ‘Greek interests’ living near Greek churches. It took the second-highest screen average of its opening weekend. The full costs and income are in my original article from 2015.

• Christmas, Again’s small but successful holiday release costing $24,000.

Publicists and PR

If there is one cost every self-releasing filmmaker eventually confronts, it is the cost of a publicist. The disclosed fees run from $2,000 to $54,000, and the verdicts run from “well worth it” to a lawsuit.

Part of the reason I saw such a wide range of costs across the examples is that “Publicity” isn't a single thing. It’s a collection of the full cost of activities, ranging in scope, scale, and complexity.

In the chart below, the Columbus and Unrest figures are total PR spend across the entire campaign (consultants, print ads, travel, press screenings), whereas the 1946 figure is a publicity firm’s fee within a $75,000 theatrical release. The small ones are single engagements.

Read more

One finding was that women were less likely than men to make a second film on the same scale. A few of you got in touch to ask for more details on this, so I went back to the numbers.

I studied the career paths of 1,765 directors, all of whom have directed a top-grossing movie since 1980. I looked at the professional work they had done since directing a movie that appeared on the ‘top 100 US-grossing’ chart that year. More detailed information about my methodology is at the end of the article.

Among directors who reached the top 100 and had plenty of time to go again, 53% of men directed a second top-grossing film, but only 40% of women did.

So I wanted to know why. And the answer turned out to depend almost entirely on one thing: how well the first film did.

Does the size of the first film matter?

I sorted every director by the size of their first top-grossing film, using its US box office rank that year, and then checked how many went on to a second.

When the first film was a genuine smash (i.e. one of the 33 biggest of its year) the gap disappears. In fact, women actually do a touch better, getting a second film 76% of the time, compared to 72% for men.

This means that if you’re seen as a successful director, then your gender does not seem to matter much. Money talks!

(This is the same logic as the famous orchestra experiments, in which putting a screen between the musician and the panel sharply increased the number of women hired. A hard, objective signal leaves less room for bias.)

But if you fall short of expectations, a gender divide comes into play. Among directors whose first film landed outside the very top tier, 46% of men went on to have a second top-grossing film, but just 29% of women did.

You might reasonably wonder if that is about gender at all. Perhaps the female directors simply made smaller or less commercial first films, and are being judged on the film rather than on being women. I tested this with a model that holds the first film's size and the era it came out in steady.

Even when you line up a man and a woman whose first films performed identically in the same years, the woman’s odds of a second top-grossing film are about 40% lower.

A recent study of French cinema found the same asymmetry from the other direction, with box-office success raising a woman's chance of directing again far more than a man's.

What happens after a flop?

The sharpest version of this is what happens when a first film loses money.

When a director’s debut lost money at the US box office, the men still returned to the top 100 at 35%, versus just 25% for the women. When the first film made money, the men returned at 59%, the women at 44%.

A man is forgiven for a first film that underperforms. A woman, by and large, is not.

This is not unique to film. Economists have long found the same forgiveness gap in other high-pressure jobs. Female surgeons lose far more referrals than male surgeons after a patient dies, female financial advisers are more likely to be sacked and less likely to be rehired after misconduct, and women whose start-up fails are markedly less likely than their male co-founders to get funded again. Psychologists have a name for the underlying habit: a man’s failure tends to be read as bad luck, a woman’s as a lack of ability.

This is not news to the people at the coalface. Actor and director Eva Longoria put it like this in 2023:

The problem is if this movie fails, people go, ‘Oh Latino stories don’t work…female directors really don’t cut it.’ We don’t get a lot of at-bats. A white male can direct a $200 million film, fail and get another one. That’s the problem. I get one at-bat, one chance, work twice as hard, twice as fast, twice as cheap.

But did the women just step back?

Another suggestion, following the previous article, was that female directors may be stepping back from directing due to family commitments.

Directing a huge film is brutal on your private life, no matter your age, gender or support network, with 14+ hour days for the best part of two years. If anything, the economics would predict that women would bear the brunt of this. Claudia Goldin won a Nobel prize for showing that jobs which reward long, inflexible, always-on hours push out whoever carries the caregiving load, usually mothers.

Directing a tentpole is about as greedy a job as it gets, which makes what the data show all the more surprising. So do these demands lead to more women choosing to step back than men?

The data says no.

If women were leaving in their late thirties to raise children, you would see a bump in the ages at which they stop directing. As the chart below clearly shows, there is no baby bump.

The two curves sit almost on top of each other, and both peak around 60. Women who eventually stop directing make their last film at a median age of 61, the men at 60.

Only about one in ten of either gender makes their last film between 35 and 45. If anything, the women keep going a little longer, with directing careers that run about three years more than the men’s after their first big film.

The second convincing piece of evidence debunking the ‘women chose to leave’ myth is that the women who don’t get that second top-grossing film have not, for the most part, left directing altogether.

Among directors who never returned to the top 100, the women carried on directing other things, smaller films, television, shorts, at almost exactly the same rate as the men (86% against 88%). They actually directed slightly more of them.

USC Annenberg reached the same conclusion when they asked the people who actually hire directors. In their interviews, work and family balance ranked well below gendered financing and male-dominated networks as the reasons women were missing.

The female directors did not stop directing, but the industry stopped offering them big jobs.

This fits something I found a decade ago in a study I conducted for Directors UK, that women were less likely to be handed a second, third or fourth film even after directing their first.

Epilogue

This is all rather depressing, but in the pursuit of finding an encouraging angle to end on, I will note that this is a narrow problem, and a fixable one.

The entrance is already roughly fair, and the evidence shows that women lack neither talent (their films score no worse with critics), ambition (in one study, 44% of female directors said they wanted to make big-budget films, even as most doubted they would get the chance), drive, nor staying power.

The thing that would move the numbers most is unglamorous and entirely within the industry's gift. Let women fail like men, and hire them back anyway.

Notes

For today’s research, I defined a “top-grossing film” as one of the 100 highest-grossing releases at the US box office in its initial year. A “top-grossing director” is anyone who directed a film that grossed over $100 million, and a “first-time” top-grossing director is one who reaches that group for the first time. Co-directors are each counted. My study covers first appearances from 1980 to 2025, 1,765 directors in all. This top-100 framing follows the USC Annenberg Inclusion Initiative’s annual analysis of the director’s chair, which I rebuilt to track each director’s career.

Box office data came from The Numbers, Box Office Mojo, and Wikipedia. Directors, prior roles, release years and dates of birth are from Wikidata, OMDb, IMDb, The Numbers, Opus and other data sources.

The headline second-film figures (40% of women and 53% of men) and all the splits below use directors who reached the top 100 by 2018, so everyone has had at least seven years to make another film. That is 1,399 directors.

A director is considered to have stopped directing if their last credited directing work, of any kind, was in 2021 or earlier. For the age analysis, I used directors who arrived by 2015 and have a recorded birth year. The “kept directing” figure counts any directing credit for a film, short, or television programme after the first top-grossing film, but not any additional top-100 films.

I measured the size of a first film by its US box office rank that year, splitting into the top third (ranks 1 to 33), the middle third (34 to 66), and the lower third (67 to 100). The “below the top tier” figure combines the middle and lower thirds. The female groups in these bands are small (25, 36 and 43 directors), so the individual bars are best read together rather than one at a time. The pattern runs the same way across all of them.

To check that the gap was not just women making smaller first films, I ran a logistic regression of whether a director made a second top-grossing film, against gender, the inflation-adjusted domestic box office of their first film (logged), and the year they arrived. Holding the first film’s size and the era constant, women had about 40% lower odds of a second film (odds ratio 0.60; 95% confidence interval 0.39 to 0.91; statistically significant).

The money-lost-or-made comparison uses the 936 first films for which I had a production budget, comparing the budget to the domestic box office. The group of women whose first film lost money is small (8 directors), so treat that particular pair as indicative rather than precise. The direction matches everything else.

A director’s gender was applied across their whole career, using their present-day identity. This is a very minor point from a purely data perspective, which is why it is in the Notes rather than flagged in the main article. However, I know it will animate some people on either end of a topic where feelings run high.

It’s relevant here as both Wachowskis identify today as transgender women, using the pronouns she/her, but were seen as men when they co-directed their first top 100 film The Matrix in 1999.

I cannot profess to have any special insight into the best way to tackle this. It’s a deep topic that many people feel aggrieved by. I would note that the comments section of this article is hardly the right forum to debate something as big as ‘What is a woman?’ No one is going to solve anything, nor change anyone’s mind.

Whichever perspective you take, this decision accounts for just 1.9% of the directing credits for the year 1999 and 0.27% of all directing credits I studied.

I genuinely want to hear from everyone in the comments. Feel free to share your thoughts, ask me anything, challenge me, etc. Just be nice to each other. I won’t delete comments I disagree with, but I will remove nasty ones.

If you would like a slightly more… ahem… sensible reading of the data, you may want to check out my article from a few weeks ago, titled "Are film directors getting older?"

I’ll be uploading new video essays every couple of weeks throughout 2026, so make sure you’re subscribed on YouTube to get notified when another drops (also, subscribing will really help me out with the Algorithm Gods, so it would mean a lot to me - thank you!).

https://www.youtube.com/@TheFilmDataScientist

And to help explain the channel, I also created a short channel trailer:

Today, I’m going to go through the market forces which have led to the current state of mainstream cinema and show you the data behind it.

The core mystery

Let’s start with the seemingly incongruous situation.

Hollywood as an industry has been honing its craft for well over a century. Given the power, prestige, and money on offer, it attracts some of the brightest minds of each generation. It’s also an extremely visible sector, where both hits and misses are clear to everyone, meaning there is no shortage of discussion and analysis.

Cinemas around the world have taken in close to $0.9 trillion over the past two decades, in today's money, and that is theatrical tickets alone, before a penny from streaming, discs, or television. So there are both the resources and the motivation to make better products.

Finally, good movies do exist. Humans have figured out how to make feature films which make us laugh, cry, fear, feel and get excited. Making a good movie is not easy, but it’s far from an unknowable mysterious process like dark matter.

So you would have thought the most motivated, smartest, and best-resourced folks in the business (i.e., Hollywood Execs) would have all they need to make pretty much every movie they make extremely good.

And yet a frequent refrain from audiences is that many modern movies are bad.

• NRG polling for TheWrap found that the share of moviegoers who want more original films in cinemas rose from around 60% in 2023 to 72% in 2025.

• And this isn’t a new phenomenon. In 1993 half of Americans thought films were getting worse.

• People will even tell you, flatly, that better films would bring them back. When Gallup asked in 2005, about 70% said higher-quality films would make them more likely to go to the cinema.

What gives?

Why isn’t Hollywood making better movies?

The first thing to note is that Hollywood is not trying to make good movies. The execs are selected, trained and rewarded to act more like corporate CEOs than artists or filmmakers.

As a result, they are trying to make money, not art.

Furthermore, they don’t even want to make movies. The way to sustainably make money in an art-based business is not in creating the works but in selling them.

• Making movies is expensive, risky, and you have to deal with unpredictable creatives. Every successful film is a new prototype that is hard to replicate and scale.

• Selling movies is scalable, repeatable, and you’re dealing with business-y folk. After all, cinemas and streamers are not in the art business either. The former rents seats and sells sugar, whereas the latter sells subscription-based access to bottomless content buckets.

At its core, Hollywood is a logistics and marketing business. The only reasons the studios make movies are because:

1. Supply. No one else can create the type of movie they want to sell.

2. Ancillary income. They want to be able to make money from the other rights (t-shirts, theme park rides, etc)

3. Power. They want control over what the product is and what they can do with it.

The most reliable income a studio receives from its involvement in a movie will come from charging distribution fees and exploiting the IP elsewhere, rather than taking a share of any profits.

So, if reliable income comes from selling a stream of movies, then the incentives shift from ‘Try to make a profitable movie’ to ‘Try not to lose money with any one movie’.

This creates a risk mitigation mindset. It doesn’t make sense to swing for the fences and try to make brilliant, original movies. Instead, all the incentives align to encourage executives to pursue risk reduction.

This has created a spiral. The “safest” films are the biggest ones (movies over $200 million, flop the least often, only 31% of the time), because that budget is only approved for something with a guaranteed audience.

But when one does go down, the loss can be huge (median loss is $183 million). This makes the failures more dramatic and painful, thereby fueling further fear and risk-mitigation efforts.

As movies have got bigger and bigger, so too has the pressure on Hollywood execs to have an ‘anti-flop mindset’.

A steady stream of mediocre films is good business. A series of ambitious attempts at great art will inevitably lead to some very damaging losses, which hurt pride and wallets.

So let’s look at the data in order to build a simple ‘Anti-Flop Manifesto’.

I crunched the data on 2,271 feature films, released since 1980, each with a budget over $50 million in today's money. I’m going to refer to anything that didn’t recoup twice its budget at the box office as a ‘flop’. The reality is more complicated than that (see Notes at the end of the article), but for now, just know that these are the underperforming titles most likely to result in a net loss for the studio.

The Hollywood ‘Anti-Flop Manifesto’

Here are the five things the evidence says you should do if you want to have the lowest chance of losing money on a movie.

Rule 1 - Never make anything new

The surest way to lower your risk is to sell people something they have already agreed to like.

This is the single biggest point in the Anti-Flop Manifesto. The general public will often ascribe this approach to laziness, and while there are many things you can accuse Hollywood execs of being, lazy is not really one of them.

An original is more than twice as likely to lose money - 23% of big-budget sequels flop, against 57% for an original screenplay.

Every original screenplay you green-light instead of a sequel roughly doubles your chance of a loss. So while it’s true that more flops in recent years have been sequels of one form or another, that’s not because they are riskier, but because Hollywood is making way more sequels.

By the way, this isn’t a technical point about lineage. It’s about how it feels to the audience. The recent flop Supergirl, for all its comic-book heritage, was closer to a standalone bet than a true sequel. A fresh lead, a character most people could not place, and no run of hit films behind her all make for a riskier bet.

Rule 2 - Stick to genres that people show up for

But I can see why Supergirl felt like a relatively safe movie to greenlight. In the past few decades, superhero films, the genre journalists keep pronouncing dead, are the safest kind of big film to make, with only 28% of them flopping.

Read more

There are really two claims tangled up in that reputation.

1. The first is that the list finds talent, i.e. it is very good at picking out the scripts and writers who matter.

2. The second is that it makes careers, as in, simply being on the Black List actively lifts a writer who would otherwise have stalled.

Those two things are not the same, so let’s test each in turn.

First, just how well do these scripts do?

Start with the scripts themselves. Of the 1,901 screenplays listed between 2005 and 2025, 576 have been made into films, and together they have grossed over $31bn worldwide at the box office.

That includes 103 films that grossed more than $100m each. Four Black Listed scripts have won Best Picture (Slumdog Millionaire, The King’s Speech, Argo and Spotlight), and a total of 74 have received Oscar nominations.

But raw totals can flatter. The fairer test is how these films compare to everything else released in the same era. So I plotted every Black List film against the wider field of films from 2005 to 2025, on two axes that matter: what audiences thought of them, and what they made.

The Black Listed projects are represented by the orange dots, and they are not randomly scattered. I have added dotted lines at the median film, meaning that 25% of any random group of films will fall in the top-right quadrant of films that are both well-liked and commercially successful.

For Black List films, 55% land there, more than double that base rate.

Three-quarters out-gross the average film and nearly two-thirds out-rate it. The typical Black List film earns around $23m worldwide and scores 6.5 on IMDb, against 6.3 for the field. These are, on the whole, good films that make money. As a piece of talent-spotting, that is a very strong hit rate.

I am not the first to find this. In 2019, researchers at Harvard Business School ran their own numbers on the annual Black List and reached strikingly similar conclusions. They found that listed scripts were about twice as likely to be produced, and the resulting films earned roughly 90% more at the box office than comparable titles made on the same budget.

Does the Black List spot talent?

Let’s move from the produced movies to the writers behind them. If the list really has an eye for identifying talent, then the writers it picks should be on the rise.

To check, I used my own measure of fame (based on a mix of public sources) to track our writers’ visibility around the year they first appeared on the list.

Before the list, these writers are in a much lower position, and that position is static. Once they get listed, the line climbs and keeps climbing, roughly doubling within five years.

So the answer to the first question is clear. The Black List spots talent extremely well. Its scripts succeed, and its writers go on to careers that visibly take off.

Does the Black List make careers?

The climbing visibility line we’ve just seen certainly does look like a career launch.

But we need to consider causation here. The reason the writers have ended up on the Black List is that they’ve written something that has got some serious Hollywood folks talking about it. They just wrote something brilliant, meaning that the listing lands, almost by definition, at a moment when their career was already heating up.

How much of the rise is the list, and how much would have happened anyway?

One way to answer that is to compare these writers to others who were in exactly the same position but did not make the list.

So I built that comparison. For each Black List writer, I found a matched control: a writer with a released film in the same year and a similar number of prior credits who had never appeared on the list. Then I ran the identical before-and-after analysis on both groups.

Black List writers roughly doubled their output in the five years after listing, a factor of 2.15. And the comparable writers who never made the list managed a factor of 1.96 over the same stretch.

It’s fair to say that some of the “Black List bump” is what happens to any writer you catch at a hot moment, but not all of it. Black Listed writers sit consistently above their matched twins, outproducing them by roughly 18%.

So being on The Black List provides a modest bump to an already on-the-up career.

It is not just writers it launches

Before we finish, here are two interesting tidbits I found along the way. While the Black List bills itself purely as a writers’ showcase, it is also providing fuel to directors’ careers.

More than a quarter of Black List writers have directed a feature film, and on a like-for-like measure they do so at over twice the rate of writers in general (35% vs 15%).

A full quarter of the Black List films that got made were directed by one of the script’s own writers, a group that includes Paul Thomas Anderson, Damien Chazelle, Emerald Fennell and Taika Waititi.

How quickly does it launch people?

The second data point I found concerns the delay between the script’s appearance on the Black List and its reaching cinema screens.

For the scripts that get made, the median time from listing to release is three years, and more than three-quarters arrive within five. The chances of a sale peak in years two and three and then fade away, so a script’s best window is early.

(That's looking at the share of films that are eventually made. Only about one script in three has ever been made, and most of those arrived within five years of listing).

So the launch, where it happens, is reasonably quick. It just does not happen for most scripts.

The Black List is a superb spotter of talent, a modest maker of careers, and, for the majority of the screenplays it celebrates, a beautiful credential that never becomes a film.

Reaction from Franklin Leonard

Before publishing a study, I like to reach out to one or two people to see if I’ve missed anything obvious and to provide context. For today’s article, I chatted with the man behind the Black List himself, Franklin Leonard.

He’s a math whizz, so rather than being affronted that I turned his life's work into a few charts, he really engaged with the findings. He was also careful not to overclaim on what The Black List is and is not:

I try to be very careful about what we claim, because the truth is already strong enough, as your analysis shows. I’ve never believed that the Black List makes careers. A writer’s career takes off because they wrote something great, something good enough that the right people couldn’t ignore it.

What the Black List does is put a very bright spotlight on that script so that the people who should read it actually do. And once they’ve read it, they get involved with that script or come back to the writer for the next thing, and the next thing, and so on and so forth.

Ultimately, we’re a catalyst, and a catalyst isn’t a small thing, especially in a system so broken that an almost laughably rudimentary data study can consistently beat the market.

And it’s still an incomplete solution to the deeper discovery problem. The annual list can only do what it does for scripts already in circulation, the ones that are already in the right rooms.

That’s actually why we built the website: To be the same catalyst for writers who might not be in the room yet at all. But the principle’s the same either way: If the work is good enough, it should have opportunities commensurate with its merits, regardless of who wrote it, where they live, or who they happen to know.

I’m very grateful to Franklin for his good nature and time.

Notes

I looked at every annual Black List from 2005 to 2025. That’s 1,901 scripts, 2,222 writing credits, 1,682 writers matched to a verified IMDb profile.

On the scatter plot, each point is a commercially tracked film released from 2005 to 2025 with a worldwide box office figure, an IMDb rating, and more than 1,000 votes (to exclude obscure titles). Box office is on a log scale. Black List films are those whose listed script reached the screen.

When creating the career control groups, each Black List writer was matched to one non-Black List writer with a released feature in the same year and a similar number of prior credits, drawn from the full IMDb writer population. Both groups are measured using the same before-and-after method, which ensures a fair comparison.

This research shows correlation, not proof of cause. The matched comparison strongly suggests that most of the career rise is due to selection and momentum rather than the list itself, with a smaller genuine lift on top. My measure of fame measures attention, not quality, and reacts to events such as releases, awards, deaths, and scandals.

The Harvard Business School study referenced above is by Hong Luo and colleagues. The accessible write-up is at HBS Working Knowledge, with the academic record here.

1. “We’re moments away from uploading a script and making movies!” (This is usually followed by the person trying to get me to look at a demo of their “ground-breaking” vibe-coded AI-to-storyboard start-up).

2. “It’s all hype and will burst any second, returning moviemaking to how it’s always been” (These folk also complain that GenZ don’t understand customer service and the generation below that can’t read).

Between these two outlier options sit the rest of us - curious, confused, and tired of having to work out what’s bullshit and what’s best practice.

I was recently asked by a highly experienced Hollywood VFX veteran what the lay of the land actually is today, so I thought I would dive into the facts and find out how things stand as of July 2026.

I sought out every report of a named film or show where AI genuinely did some of the work. I also looked at what the myriad of “AI-powered” tools, apps, and services are promising (taking special attention to which had real-world case studies and which did not).

My main takeaway is that the AI solutions that are working in professional settings today are mostly the boring kind. They are not the ones which make for the most exciting/scary headlines, but the ones which save time and money.

The tools shouting loudest are often the ones doing the least, whereas some of the most useful ones don’t (or maybe don’t need to) do much public-facing marketing.

Below is an illustrative chart of how I see things:

Let’s dive into the details.

Where AI actually works in visual effects

Let’s start with what some might call the thin edge of the wedge. This is where AI is fully integrated and accepted by all. VFX artists use AI today to have computers perform the most tedious manual tasks, and to do so so much faster than humans can do alone.

A good example is on Dune: Part Two, where visual effects supervisor Paul Lambert needed to turn the Fremen’s eyes spice-blue across around a thousand shots. Instead of painting each one by hand, the team trained Foundry’s Nuke tool CopyCat (a small neural network you teach on a few frames) to do it. Foundry reported that 40% of those eye shots needed no manual touch-up at all.

Other examples include:

• For Furiosa: A Mad Max Saga, the Australian house Rising Sun Pictures used its REVIZE machine-learning system to blend Anya Taylor-Joy with a younger performer across about 150 shots.

• On Kingdom of the Planet of the Apes, Wētā FX ran a facial deep-learning system across more than 1,500 shots.

• The Toronto company MARZ sells a beauty-fix tool called Vanity AI and says it has run on more than 27 productions.

AI has already come for people’s jobs, but it’s starting with the jobs no one wanted in the first place. The artist remains in charge, directing the robots toward a single, repetitive task.

This is clearly the safest ground for VFX companies to talk about AI use without causing a backlash. That’s because:

1. It is replacing work no one wants to do.

2. It does so without using anyone else’s art. The companies train on their own images and haven’t built a tool others can use to create new art.

3. We can all picture what’s happening, on a conceptual level - it’s not that scary.

Not all AI is AI, actually

Next up, we have “AI de-ageing”. We’ve been hearing about this for years, and the truth is, a lot of it wasn’t really AI. Well, not the Generative AI that we think of today, where you type in a prompt and get out a fully generated image.

In recent years, we’ve seen a de-aged Robert De Niro in The Irishman and a young Will Smith in Gemini Man. In the case of the former, ILM used a markerless capture rig and extensive traditional computer graphics, and for the latter, Wētā built a fully digital human. These are CGs made by hundreds of people, and where machine learning was used, it was more of a reference aid.

The visual effects supervisor on Gemini Man, Guy Williams, described it like this:

We didn’t even think of it truly as de-ageing; we thought of it as a complete creature replacement

A film that came closer to “AI de-ageing” was Robert Zemeckis’s 2024 film Here, which used a neural network rather than CG to de-age Tom Hanks and Robin Wright live on set. The company behind it, Metaphysic, ran the face replacement as the cameras rolled (albeit that’s not to say that what we saw in the final film was the real-time output).

So when you read that a film “used AI to de-age” a star, the question worth asking is whether they mean “Here,” or the other 90%, which is that skilled humans do the work.

Has generative AI reached professional screens yet?

But that’s not to say that Generative AI isn’t being used. An early example of such a project is El Eternauta, an Argentine Netflix series that used generative AI for a building-collapse shot. Netflix co-chief executive Ted Sarandos told investors a year ago that it was about ten times faster than the traditional method and that it was:

The very first GenAI final footage to appear on screen in a Netflix original.

So it was a year ago that just one AI-generated sequence making it to air earned a press release and headlines. Things are not as far along as some will have you believe.

Read more

The article critiques the work of the USC Annenberg Inclusion Initiative, which studies the gender of top-grossing directors, the festival circuit, animation, visual effects, and more. Hollywood Gadfly raised several objections to their work and methodology, including whether it is fair to treat the gap between the share of women directing at Sundance and the share directing top-grossing films as evidence of a broken pipeline.

The chart below is one such example of how Annenberg makes the case, highlighting the Sundance figure at 63.6% (in 2026) and the top-grossing directors at just 6.6% (2007-25).

Leaving aside that these figures cover very different periods, this raises the question:

Is there a pipeline from being selected in the US Dramatic Competition at Sundance to top flight directing?

I studied the career paths of 1,765 directors, all of whom have directed a top-grossing movie since 1980. I looked at the professional work they had done before they directed a movie that appeared on the ‘top 100 US-grossing’ chart that year. More detailed information about my methodology is at the end of the article.

Where do directors come from?

I narrowed in on each director’s first appearance in the annual top 100 grossing films at the US box office.

I looked at their past work, tagging it with several types of activity, which also means that one person could have multiple tags, since a few people just stick to one thing.

Three-quarters of them had worked in television prior to making their first top-grossing movie. Almost the same number had been credited as writers on a TV show or movie, and just over half had directed a feature film that was not one of the 100 highest-grossing movies in the year it came out.

Given how big writing is, I thought I’d split that out. Writing movie screenplays was the biggest type of writing gig that top-grossing directors did before reaching the top 100.

So does the Sundance discovery pipeline hold up?

The core idea is that people who go on to become top directors are discovered on the festival circuit. Specifically, being in the dramatic competition at Sundance is a meaningful springboard to a studio career.

I tracked eighteen of the discovery markers - the competition strands at Sundance, Cannes, Berlin, Venice, Toronto, SXSW and Tribeca, the first-feature and new-director awards, and the celebrated short-film prizes, and I followed every director who hit each one forward to see how many later directed a top-grossing film.

16% of directors nominated at the Spirit Best First Feature award went on to top-grossing status. That’s slightly above Sundance (14%), an Oscar nomination for best animated short (11%), and the Discovery strand at TIFF (8%).

On the chart below, I have put most festivals in blue but highlighted the main competitions at Cannes, Berlin, and Venice in orange, as one could argue these are not really seen as ‘discovery’ competitions.

So Sundance is the second-best choice for comparison, but that’s from within a batch of pretty weak signals.

If you present a pipeline that is just “Sundance to top-grossing”, then you would be looking at a small minority of the people who actually make it, while the meaningful markers are having worked in writing and television.

What is the churn rate for top directors?

Another thing to consider when assessing the efficacy of using a pipeline model is the churn rate. i.e., how many people each year get to direct a top-grossing film for the first time.

Interestingly, this has been fairly static for three decades, but then it has started to rise in the past ten years. Until the mid 2010, around a third of directors of top-grossing films had not previously had a title in the top 100.

2020 was a freak year, thanks to Covid, in which 71% of directors in the ‘Top 100 Club’ were new (77 of 109 directors). This is affecting the five-year rolling average, but that’s not the sole cause of the recent rise. There was also a real, sustained step-up beginning around 2016, from a long-run norm near 33% to a new normal around 40% to 45%

The feeders themselves have shifted, too. Writing and television have always led, and both have grown more common among recent arrivals, with editing and acting rising alongside them.

Some of this is down to careers becoming more multi-stranded, and it’s possible that some of it reflects the fact that recent careers are simply better documented than those that began in the 1980s. Either way, the ordering is steady throughout, with writing and television on top the whole time.

Earlier this month, I wrote about how the average age of directors has risen a lot in recent years, so you may want to check out that article for context, too.

Do the routes treat women and men the same?

Now that we have a more detailed and nuanced picture of how people become top-grossing directors, we can split the cohorts by gender and assess the extent to which gender is a factor.

Let’s start with how many women reach the top 100 at all. The share is a long, uneven story rather than a steady climb. Women were close to one in ten of new top-grossing directors in the early 1990s, slipped back through the 2000s, and only in the 2020s reached around 18%.

I pooled all the early-career routes, festival competitions, first-feature awards, and short-film prizes and followed only those who had attained them by 2012, so everyone has had more than a decade to convert.

One in ten male directors from that pool went on to be a top-grossing director, compared with just 4% of female directors.

On the main feeders, women and men look broadly alike. Both come overwhelmingly from writing and television. If anything, women lean into those two even more than men (85% of women had worked in television first, against 75% of men).

The biggest skew towards men was among cinematographers stepping up to directing, which was a small route in and picking from a field that has historically been massively male.

The festival discovery route is where we see a big difference in outcomes. Men out-convert women at the Sundance dramatic competition, at Cannes, at Venice, at Toronto's discovery strand, and at the Oscar short categories.

The direction is the same almost everywhere, and it holds across four decades, so it is not a matter of women having arrived too recently to have converted yet.

The women who do come through these routes are exactly the names you would expect - the directors who went on to make some of the most admired films of recent years. So the route does work for some women. It simply works for a far smaller share of them than of the men who start in the same place.

The other biggest difference between the genders is getting a second crack at the big director’s chair. Reaching the top 100 once is hard for everybody, and most directors never do it twice. But the rate at which they get a second turn differs.

Among directors who arrived by 2018, with plenty of time to make another film, 53% of the men went on to a second top-grossing film. For women, the figure is 40%.

Over a whole career, men in this group average about 2.8 top-grossing films each, women just 1.9.

What does all this tell us about directing careers?

We’ve gone through a lot in this article, so I thought I’d put a handy list of things we’ve learned along the way:

• Getting the job as a director on a movie that ends up in the top 100 highest-grossing movies of its year most commonly comes to men in their early forties, after years of other work.

• Women and men enter the top 100 through the same main doors, those of writing and television.

• A third of the seats turn over each year, while a stable group of returning names holds the rest.

• The festival and award routes contribute a real but minor share, and none of them comes close to being the pipeline.

• The gap between genders emerges at the discovery routes, which carry men through at more than twice the rate, and after the first film, where men are markedly more likely to get a second.

• Some directors do go from being in the official selection at Sundance to becoming a top-level director, but not many. It’s one of the biggest routes within the festival discovery framing, but a poor one overall.

Notes

I’m very grateful to both the pseudonymous Hollywood Gadfly and to Dr Katherine Pieper, Program Director of the Annenberg Inclusion Initiative, both of whom generously talked this through with me over Zoom.

For today’s research, I counted a “top-grossing film” as one of the 100 highest-grossing releases at the US box office in its initial year of release. A “top-grossing director” is anyone who directed such a film, and a “first-time” top-grossing director is one reaching that group for the first time. Co-directors are each counted. My study covers first appearances from 1980 to 2025, 1,765 directors in all. This top-100 framing follows the USC Annenberg Inclusion Initiative’s annual analysis of the director’s chair, which I rebuilt in order to follow each director’s career forwards.

Box office data came from The Numbers, Box Office Mojo, and Wikipedia. Directors, prior roles, release years and dates of birth are from Wikidata, OMDb, IMDb, The Numbers, Opus and other data sources. The festival and award records, including the dramatic competitions and first-feature and short-film prizes, are from my awards database covering 1978 to 2026; for each route I took the directors of every film in it, not only the prize winners.

When I mention film in competition at Sundance in today’s piece, I am specifically referring to films in the Sundance US Dramatic Competition. Sundance has a number of selections beyond this.

A minor footnote on the Sundance figures is whether the festival film itself reaches the top 100, or if the director gets there later with a subsequent film. It is almost always the latter, with only about one in seven cracking the top 100 with the film they premiered at Sundance. That share is inflated thanks to the strange pandemic years, when the box-office bar fell low enough for small Sundance titles like Minari and Zola to slip in.

For each route, a director “converts” if they directed a top-grossing film after the route year, so the route plausibly preceded rather than followed their establishment. Measuring it at any time gives higher figures for the main competitions, because established directors keep playing them, which is why the directional measure is used and those routes are flagged. Eighteen routes were tested in all; the gender comparison for them is shown for people who reached the route by 2012, with thirteen or more years to convert, and the gap holds in that long-runway pool (women 4%, men 10%). Per-route female counts are small for some routes, so the pooled figure is the reliable one.

Prior roles come from credited work, so uncredited jobs, foreign-language work and very early-career roles can be under-recorded, and this is more true the further back you go. Recent arrivals show more prior credits of every kind than those from the 1980s, which is partly a real change and partly better record-keeping, so the prior-role shares are best read as a floor, especially for older directors, while the relative ordering, with writing and television on top, is more reliable than any single figure.

A director’s gender was applied across their whole career, using their present-day identity. This is a very minor point from a purely data perspective, which is why it is in the Notes rather than flagged in the main article. However, I know it will animate some people on either end of a topic where feelings run high.

It’s relevant here as both Wachowskis identify today as transgender women, using the pronouns she/her, but were seen as men when they co-directed their first top 100 film The Matrix in 1999.

I cannot profess to have any special insight into the best way to tackle this. It’s a deep topic that many people feel aggrieved by. I would note that the comments section of this article is hardly the right forum to debate something as big as ‘What is a woman?’ No one is going to solve anything, nor change anyone’s mind.

Whichever perspective you take, this decision accounts for just 1.9% of the directing credits for the year 1999 and 0.27% of all directing credits I studied.

I genuinely want to hear from everyone in the comments. Feel free to share your thoughts, ask me anything, challenge me, etc. Just be nice to each other. I won’t delete comments I disagree with, but I will remove nasty ones.

Further reading

If you would like to read more about the statistics behind gender in film, then you may enjoy the following: (ok, bad choice of words - you may wish to read the following):

• A major new study into gender inequality in the UK film industry (2016)

• UK films with public funding hire more women (2016?)

• Major new study into gender inequality among UK film and TV writers (2018)

• Are movies dominated by a small number of old male composers? (2018)

• Are women less likely to direct a second movie than men? (2019)

• What percentage of film producers are women? (2019?)

• How often are women hired in key film departments? (2019)

• How well are women represented among short filmmakers? (2020)

• Do male and female critics judge films differently? (2021)

• What percentage of movies are written by women over 40? (2024)

• Is Jason Blum right that there is a shortage of female horror directors? (2025)

• How do women fare in gender-neutral Oscar categories? (2025)

In the past couple of months, two films made silly amounts of money. Obsession was made for about $750,000 and has already grossed $297 million worldwide, and Backrooms earned $257 million worldwide on a $10 million budget.

This is the dream every film investor has when funding a movie - a relatively small bet pays off hugely.

Despite seeming similar on the surface, these two movies took a very different approach to financial risk. Obsession was sold to Focus Features for around $15 million at last year’s Toronto Film Festival, eight months before release. Conversely, A24 financed Backrooms itself (with Chernin Entertainment) and kept both the risk and the upside.

Taken together, these films reveal a lot about how the film industry treats risk, money and approaches to business.

So I thought it was a good time to tackle one of the most common questions I get asked:

How do you make money investing in movies?

There are many answers to this, so today I’m going to focus on how people make a sustainable business investing in movies, given that:

1. The movie industry has massively asymmetrical returns, with ‘the hits’ providing huge returns, while ‘the rest’ delivering big losses with almost no post-release assets to recoup from. (If you buy a house and house prices tank, you still have a house. If you invest in a film and it doesn’t hit, then all you have is your ‘Associate Producer’ credit and maybe a crew jacket).

2. It is very hard to predict whether any single film will be a hit. This is even harder than the industry likes to admit, and provably so, especially given that the key things which can be useful mild indicators of success are rarely known at the financing stage.

3. This means you need a business model that doesn’t rely on luck. You need to find a model that goes beyond picking winners and pays off over time. (Either this or be honest and just enjoy the gambling).

So let’s dig into these things and learn how some people, professionally, generate returns over time.

Only a small number of movies matter

Possibly the single most important idea to grasp about the film business is power laws. This is where a small number of examples take home most of the rewards.

Most things in life cluster around an average. A good example is human height. Almost everyone is near the middle, nobody is fifty feet tall, and the average tells you what to expect from the next person.

Film money does the opposite. A few films tower over the thousands beneath them, and the gap is enormous.

Below are the worldwide box office receipts for 49,834 films (converted to 2026 dollars). As you can see, there is no “typical” film.

Another way to look at the same data is to line up every film from the lowest-earning to the highest-earning and ask what share of the total each slice represents.

The top 1% of films take 31% of the total box office. The top 5% take just over 70% of it.

And this effect has actually been growing in recent decades.

So a tiny number of films carry the whole industry. Which means most films must lose money, and they do. Big time. Statistically, if someone asks you to invest in a movie and you are interested in making money, your answer every time should be a resounding “Hell no!”.

They may be able to show you averages from the past of their type of movie, but in reality, the “average film” is a fiction. There is the tail, and there is everyone else.

Soooo… how do we work out which movies will sit in that vaunted ‘tail’ position so we can back them?

People have tried very hard to predict it

For forty years, economists, marketing professors and computer scientists have tried to crack film prediction, using budgets, stars, scripts, trailers, tweets and, lately, AI.

I have read that whole body of literature and written a companion guide you can download, study-by-study, with what each one found and whether it holds up.

Subscribers can download a free copy of my “Predicting Movie Hits" report at the end of this article, summing up the academic literature on the topic.

Here’s the overview of the picture from academics:

The numbers in brackets are the year of publication. This matters a lot as the film economics have changed dramatically over the past decade, so many of these studies would not be replicable with modern movies.

• The biggest useful single factor in estimating a film’s eventual revenue is the budget. This can be used to explain 70% of revenue, but, sadly, only 22% of profit, because bigger budgets have higher bars to clear. (1999)

• A screenplay carries enough signal to tilt a slate of bets. (2007)

• Other meaningful factors include genre, an Oscar and good reviews. (1983)

• Activity on a film’s Wikipedia page a month before release accounts for roughly 77% of the opening weekend. (2013)

• A film’s stars do not reliably move the return once everything else is accounted for, a result that keeps reappearing across the decades. (1993)

• Web search data and social media buzz add almost nothing once you know the budget and how wide a film opens. Most “signals” are just another way of measuring what the studio has already spent. (both 2010)

• Film critics predict success, but do not cause it. They spot the quality the audience would have found anyway. (1997)

• Many of the most eye-catching results lean on the opening weekend to “predict” the total, which jumps the explainable share from 45% to 97%. To my mind, that is reading a result, not forecasting one. (2000)

• Underneath it all, the maths itself (1999) is brutal. Box office is so heavy-tailed that forecasts have, in the authors’ words, “zero precision”, and even today’s best AI (2025) is at its worst on exactly the smaller films an investor most needs help with.

Outside of academia, there is a cottage industry of services within the film industry, happy to sell you tools as you try to illuminate the unknown. These fall into five broad groups:

• Theatrical forecasting firms predict box office for films already on the release calendar. These include Gower Street Analytics, Boxoffice Pro, The Numbers, Numero, and fellow Substacker Shawn Robbins at Box Office Theory.

• Tracking and testing firms survey audiences before release to gauge awareness and intent, which becomes the trade’s tracking numbers, such as National Research Group, Screen Engine/ASI, MarketCast, The Quorum, Comscore.

• Signal trackers, such as prediction markets (Hollywood Stock Exchange, Kalshi and Polymarket) and firms inferring meaning from crowds, search and social activity (Parrot Analytics, Spectura, Emberos, RecoBee).

• Greenlight predictors have been a holy grail for decades, estimating revenue from just a script or cast package. Sounds straightforward, but many fall prey to the “average film” fallacy we saw above. It’s been a turbulent corner of the sector, with many falling by the wayside (Epagogix and ScriptBook) and some still active (Cinelytic, Largo.ai, StoryFit, Vault AI, CineCockpit).

• Comparable-title report builders provide real (or calculated) revenues of past films. These are an everyday tool of indie financing, despite producers' rampant cherry-picking when using them for ROI calculations. Companies in this space include Nash Information Services, FilmProfit, FilmProposals, Cinelaunch, Slated, Callaia.

These types of services are good at providing aggregate market-level estimates, detailing what’s happened before, and in predicting performance in the weeks leading up to release.

However, single-film pre-production revenue prediction is more of a wild west, with bold, uncredible claims lacking independent validation.

Some prediction companies have meaningful offerings, such as nifty tools, relevant advice, and a large number of data points, but many are also in the business of using marketing spin to sell false certainty to unwitting new entrants.

Regardless of the scale of snake oil in this sector, it’s certainly true to say that no one has yet been able to devise a reliable way to spot enough hits, early enough in the lifecycle of a movie, to be able to build a credible business on.

(Side note: If any company in this space wants me to independently verify their claims, then please do get in touch. Seed and Spark did this a decade ago, and I was happy to validate their claims).

How to succeed in business without really predicting

If “We only invest in hits” is not a viable business strategy in film, what is?

Here’s the key idea that allows people to overcome the hurdles we’ve seen so far:

Unpredictable is not the same as random

There is an oft-repeated phrase in the industry from screenwriter William Goldman:

No one knows anything.

The thing is, people have edited the original quote to change its intended meaning. They use it to mean “Everything is unknowable”, whereas the full quote is closer to “Nothing is certain”. Here’s the full line from Adventures in the Screen Trade:

Nobody knows anything...... Not one person in the entire motion picture field knows for a certainty what’s going to work. Every time out it’s a guess and, if you’re lucky, an educated one.

So it’s not true that no one knows anything. In fact, many people know a lot. Hollywood has run the global film industry for over a century with surprisingly little change, given how unpredictable any one movie can be.

This is because while betting on any one particular movie is a risky proposition, there are numerous things people can reliably and successfully do to build a profitable business in the film industry.

It’s about shifting the core question from “How do I pick the winner” to “How do I change my relationship to the risk”.

Below are 18 approaches professionals use to sustainably make money, none of which require predicting a film’s appeal or outcome. Think of these more as aspects of risk management than as a playbook you can just follow by the numbers.

1. Make many films, not one

An interesting quirk of the asymmetrical nature of film returns is that volume converts an unpredictable bet into a dependable distribution. In short, if funding a movie is basically gambling, making a whole bunch of them is basically owning a casino.

The chart below illustrates what happens as you stop betting on one film and start betting on many. The individual films are as unknowable as ever, but suddenly it looks a little more like an actual business.

So if you are fortunate to have access to vast amounts of capital (and nothing less risky or with better upside available to you), here are some things you may wish to consider:

• Build a slate so winners cross-subsidise losers, and you stop caring which is which. This is a core part of the philosophy of Hollywood studios.

• Co-finance slices of many slates, taking 10% of fifty films rather than all of five. A good example was Legendary’s long partnership with Warner Bros.

• Run a low-budget volume factory where one breakout pays for years of misses. Paranormal Activity, made for about $15,000, took $194m, kicking off Blumhouse’s stellar horror-based business model.

• Turn one hit into a franchise, a fresh roll with dice that already landed once. Predicting the release date of a sequel or next instalment of a much-loved franchise is a safer bet than any random film (albeit I don’t think the backers of Joker 2 would have much time for this suggestion!).

Read more

Four geologists at the University of Gothenburg spent more than a decade keeping a list of every film they could find that featured a geologist. Across the 141 movies they tracked, the on-screen geologist died 34% of the time. Usually quite early, and often moments after explaining that the volcano is about to go off.

Erwin asked me:

How does the fate of geologists in movies compare with that of other jobs?

This is exactly my kind of question as it’s core to the work I do creating new movie records for Guinness World Records.

So I turned to the data to see just how slated geologists are. Whether their luck had eroded, their prospects had hit rock bottom, and their odds of survival were, frankly, being taken for granite.

I studied 43,983 movies to track the fate of 328,538 characters who shared 10,112 different job titles and experienced a combined 69,832 on-screen deaths. I can only study the data available in screenplays, film synopses, and plot descriptions, so there’s a certain amount of bias. I have included more about my methodology and data at the end of the article in the ‘Notes’ section, which might be better called ‘Pedant’s Corner’ (or should that be Pedants’ Corner? I mean, it’s not really a corner….)

What are the most dangerous jobs in movies?

Across all the 328,538 characters for whom I could determine both a job title and their fate, 21.3% died. Bummer for them, but great for us, since we now have a baseline. Being a named character in a movie has a 1-in-5 death rate.

It’s three times more dangerous to be a Commando in a movie than an average character. They came out as the ‘Most killed’, with almost two-thirds of them not making it to the final credits. Then, a terrorist, a gunman, a revolutionary and, naturally, a sorcerer.

Half of all movie assassins die, which, for a job that is entirely about killing other people, is a remarkable number of own goals.

Jobs that seem the safest include wedding planner, matchmaker, illustrator, manicurist or sportswriter.

What even is a job?

Some of the entries on that previous chart are not jobs in the traditional sense. For example, one could see serial killing more as a calling than as gainful employment. Other roles, which may not seem like real jobs at first glance, do pass muster, such as those of an informant or a thug.

Also, if there are any young people using today’s data to inform their future career choices (as I imagine many are), then they may feel underserved by discovering that jobs such as commando, terrorist, gunman, and revolutionary have high death rates. They are people whose entire job is violence, and of course, they die, because a film about armed men is usually a film about armed men shooting each other.

So let’s remove the jobs that are centred on violence and those that people do for the love of it. Oh, and wizards and such - I removed jobs that kids today would be foolish to aspire to.

This reveals that the deadliest ordinary on-screen job, once the soldiers and the sorcerers have gone home, is that of a diver, with a 52% mortality rate.

Just in case any of you were thinking of a long career in crime, the data suggests you should aim for middle management. The pay and benefits beat frontline work, but the death rate is lower than that of the criminal C-suite.

Below are the death rates for jobs with organised crime (I left out the disorganised crime roles, as that veers into hobby territory).

It’s less about your job than the world in which you do it

So far, I have been focusing solely on the characters' jobs, whereas we get an even stronger set of survival cues from the movie in which those characters operate.

Unsurprisingly, horror films have the highest death rate, with almost half of all characters (for whom I could determine both their employment and their fate) dying.

It’s not about bad jobs, but bad people

Unlike in real life, villains die way more frequently than good people (38% vs 11%).

On screen, being the bad guy is very nearly a death sentence, which mirrors what the Gothenburg geologists found: 77% of those they called “evil geologists” died. (I am slightly suspicious of just how much harm an evil geologist could do, given the timeframes they are used to working with, but I’m not going to mansplain evil geology to leading geologists).

The traits of a deadly job

It’s at this point that I have to be honest with you - I have very little experience with proper jobs. I have what people kindly call a portfolio career, which mostly means I do several odd things at once. Over the years, my professional work has included:

• Film data analyst

• Running a story agency

• Inventing new world records

• Writing puns for headlines of the horoscopes in The Sun newspaper

• Giving Kanye West tax advice

I like to think of myself as “business feral” - i.e. hard working but barely employable.

So, when it comes to deciding what is and is not a ‘proper job’, I have very little to draw from. Case in point: see below for the job traits I thought we were worth investigating death rates for.

Other things I tested included:

• Money does not save you. A surgeon on $240k a year dies about as often as a cleaner on $34k (correlation +0.02).

• Automation risk does not matter either (+0.08), so the robots and the screenwriters are after different people.

• And the jobs children dream of are mostly safe, the lone doomed dream being that of an astronaut.

The two biggest correlations were:

1. The first is real danger, but only one kind. Cinema has no idea which jobs shorten your life in a non-visual sense (i.e., those with poor working conditions or associated with negative health outcomes). But the jobs that can kill you in a single dramatic accident, the fisherman, the pilot, the miner, the lorry driver, die on screen too (correlation +0.5).

2. The second is gender. The more men do a job in real life, the more it dies on screen (-0.43). Soldiers, pilots and bodyguards die constantly, whereas nurses, hairdressers and secretaries mostly live. It’s such a strong link that job titles ending in ‘man’ are the deadliest suffix in cinema and jobs ending in ‘ess’ are the safest.

So what have we learned?

Nothing useful.

The fact that I’m unemployable also means I’m unfirable. Sorry :/

Notes

I analysed 43,983 feature films and extracted 328,538 named characters. For each one I recorded a stated or strongly implied job, whether they died on screen, and how. I was working from plot summaries and, for a large share of the films, full screenplays.

A character’s job only got recorded when the film bothers to mention it, so the sample leans towards jobs that matter to the plot. This measures the conventions of screen storytelling, not real life. Genre drives the raw body count more than any single job, so the figures are best read as relative standings rather than your personal odds. In my tests, I found that plot summaries missed about 15% of the deaths a full screenplay would reveal, though they get the running order of jobs right.

The geologists were the obvious test, since the Gothenburg team had already counted them by hand. My method puts the geologist death rate at 31%, against their 34%, which is close enough to trust the wider net.

Eleven jobs turned up in twenty films or more without a single death between them, which seemed too good to be true, so I had each one stress-tested for counter-examples. About half cracked (a circus impresario mauled by a bear, a studio boss with his throat cut, a ski instructor lost to an avalanche, a chambermaid strangled in the billiard room, and an ice hockey player). But five held firm, with no verified on-screen deaths at all: wedding planner, matchmaker, illustrator, manicurist, and sportswriter.

The original study by Sturkell, E., Sjöqvist, A., Björklund, L. and Johnsson, A. (2026), ‘Geologists on the silver screen, the sequel’ was published in Geology Today, 42(2), pages 82 to 86, which is open access and worth a read. It is the source of the 34% geologist death rate and the 77% ‘evil geologist’ figure, and of three lovely examples: Dr No (1962), where 007 shoots the evil geologist, Prometheus (2012), where one is killed by aliens, and Dante’s Peak (1997), which somehow finds room for seven heroic geologists at once.

‘Dangerous in real life’ comes in two flavours, and they behave very differently.

• For deaths actually on the job I used the US Bureau of Labor Statistics Census of Fatal Occupational Injuries

• For the jobs that non-visually shorten life overall I used UK Office for National Statistics occupational mortality.

• Pay is from the BLS Occupational Employment and Wage Statistics

• The share of each job done by women from the BLS Current Population Survey

• Automation risk from Frey and Osborne’s ‘The Future of Employment’ (Oxford Martin School)

• Public trust, which predicted nothing, is from the Ipsos Veracity Index.

The ‘dream job’ measure was my own best calibration to childhood aspiration rather than hard official numbers. All correlations are rank-based (Spearman), so they describe whether the order matches, not the exact sizes.

That fight, though, was almost entirely about television, not film. The most pernicious version of packaging lived in TV series, where an agency could earn as much as, or even more than, the very people who created the show, and where the agency had little reason to push for higher writer pay, because its money came from the package fee rather than the writer’s salary.

Packaging fees may have ended (at least for television writers), but today I want to look into two questions:

1. How prevalent was/is packaging among movies?

2. Did it really end?

What is packaging, and how do we track it?

Despite being carried out by a small number of well-known agencies that represent the world's most famous people in the biggest movies ever, packaging is relatively hard to study.

For today’s research, I have built a fairly good proxy, but before I reveal the numbers, I need to explain a little more about what packaging is and how we might detect it.

Formal packaging is a specific deal talent agencies make with Hollywood studios. Rather than take its usual 10% commission from each client, an agency assembles the talent and charges a fee paid by the studio. In television that settled into the “3-3-10” structure, which broke down as:

• 3% of each episode’s licence fee, paid up front as episodes are made (roughly $15k–$75k an episode).

• 3% of the licence fee, but deferred until the show hits “net profits”. Due to Hollywood accounting, this would often not be triggered.

• 10% of the back end (syndication and foreign sales) over the life of the show. This is where the real money for the agencies was, as even a small number of mega hits could pay a lot for decades to come.

Here is a helpful Hollywood Reporter article which explains more.

The whole notion of packing fees skewed the incentives for agencies, and many writers felt it meant they were working against the best interests of the very clients they were meant to represent.

In his usual measured, calm style, David Simon of The Wire wrote:

Packaging is a lie. It is theft. It is fraud. In the hands of the right U.S. Attorney, it might even be prima facie evidence of decades of racketeering.

A slightly more measured came for the estimable John August:

When your agent is also the buyer, that's the clearest conflict of interest imaginable.

In response, the agencies didn’t have much more of a defence than “We like the money packaging gives us, and we want to keep doing it”. Jeremy Zimmer, CEO of UTA, said in a letter to their writer clients:

Over the past couple of decades the complexity of the business and the opportunities that have been presented have allowed agencies to decide on a path and pursue it…. We aren’t saints. We are tough businessmen and businesswomen who happen to love artists.

Karen Stuart, executive director, Association of Talent Agents told Deadline:

This development is ironic given that the Guild itself has agreed to the legitimacy of packaging for more than 43 years.

In April 2019, the Writers Guild told its members to fire their agents in protest, and over 7,000 did. After a bit of bluster, the agencies caved and signed new franchise agreements one by one. Packaging fees on Guild-covered projects were switched off from 1 July 2022.

Searching for packaging-like behaviour in film

Packaging fees were overwhelmingly a television business, built on the long tail of syndication money that almost all films lack. The Writers Guild calculated that 87% of scripted series in 2016-17 were packaged, but there has never been an equivalent figure for film.

Film packaging also works differently. It tends to live in the independent film world rather than the big studio system, and it almost always starts with a script. A writer takes that script to their agency, which then tries to attach a director and a star, and the resulting package goes out to raise financing at markets like Cannes and the American Film Market.

And it is not only agencies that put these bundles together; independent producers and production companies assemble the same script-director-star packages too.

I don’t have inside knowledge of all the deals conducted between agents and studios, neither the official ones nor the murky world of favours and influence trading. But I can look for patterns in the data.

I built a database of every movie released between 2010 and 2025 with at least 50,000 IMDb votes. I then narrowed in on the film’s core creatives, which I have defined here as the director(s), writer(s), and the top three credited cast members.

I excluded producers, not because I don’t think they’re creative, but because (a) there are a large number of empty producing credits (see more on that here), and (b) producers are more likely to represent themselves.

Finally, I sought to determine which talent agency represented each creative 18 months before the film's release. If three or more of those key creators shared a single agency, I flagged it as a “One-agency Film”.

Sharing an agency doesn’t automatically mean the film was officially packaged. This research has a lot of fiddy details and caveats, which are in the Notes section at the end, along with my methodology.

Have one-agency films increased?

One-agency films have grown dramatically, from around 10% of mainstream films in 2010 to around a quarter in the current decade.

The single highest year is 2023, when nearly three in ten big films had three of their key creators at one agency.

A rising line on its own does not prove agency action on its own. Last year, I looked at which Hollywood agencies dominate the representation of film directors and actors and found that just over half of the top actors and a third of the top directors are with one of just four agencies.

If the big agencies simply represent more people than they used to, you would expect more of the one-agency film pattern we’ve just seen to emerge through chance alone.

Could this be a coincidence?

For each year I took the actual market share of each agency among that year’s creators, then calculated how often a random four-person team would land three or more people at the same firm purely by chance.

Put simply, if everyone picked their agent out of a hat, weighted by how big each agency really was that year, how many one-agency films would you get?

The chance line sits between 7% and 11% for the whole fifteen years. The agencies did grow, and that lifts the baseline a little, but only a little.

This leads us to conclude that almost all of the rise in the observed rate is due to deliberate actions by some people.

In the early 2010s, one-agency films had about a 1.4-times the chance level. Today, they run about 2.6 times higher.

So the concentration is not only larger, it is harder to explain away as a coincidence than it used to be.

Which agency?

Almost all of the clustering comes from three firms: CAA, WME and UTA. The rest (Gersh, APA, Verve, Paradigm) turn up on a film or two a year at most.

Two things stand out to me:

The first is CAA’s 2023 spike, at nineteen one-agency films, the highest for any firm in any year of the window. That is almost certainly the CAA-ICM merger landing: CAA absorbed ICM Partners in June 2022, and a slice of what would have been ICM clusters a year earlier became CAA clusters once the books combined. By 2024-25 CAA settles back to seven or eight a year.

The second is that WME wins more years than that single CAA spike suggests. It tops 2017 and 2024, among others. Across the full fifteen years, the two are much closer than 2023 alone implies, with CAA on 138 cluster-films and WME on 117. UTA holds a steady third, strongest in animation and family-franchise work.

All three have changed hands during this period. CAA went to François-Henri Pinault’s Artémis, and WME to Silver Lake, which rebranded it the WME Group.

The writers, not the stars

When I split the overlap rate by role (how often a director, a writer or a top-billed actor shares an agency with at least one other key creator on the same film), writers cluster most, rising from 47% in 2010 to 61% in 2025.

If you want to find a one-agency film, look at who wrote it rather than who is on the poster.

This is exactly what we would expect if feature packages really do begin with a script. If the writer is the first element in the room, they become the anchor around which an agency builds the rest of the film.

The Writers’ Guild fight shows up here too, faintly. The writer line dips and stalls around 2019-21, exactly when the action was live, and a chunk of the writing population had no agent at all. Then it climbs again, and by 2023 it is at a fresh high.

This is what happens when you target just one aspect of a problem. It’s like when you say to a kid, “Don’t let me catch you doing that!” and they reply, “Ok, I won’t let you catch me”.

In this instance, the Writers’ Guild said, “You shouldn’t be charging a fee for not working against the best interests of your clients” and the agencies replied, “Ok, we’ll stop invoicing”.

Notes

Today’s research examines 2,058 films released between 1 January 2010 and 31 December 2025 that received at least 50,000 IMDb votes. Of those, 1,989 had at least one key creator whose deal-time agent could be resolved. The films that drop out are mostly international productions whose creators have no US agency footprint, or films whose entire teams were first-timers not yet on the signing record.

The tracked ‘key creators’ were the credited director, all credited writers, and the top three billed cast.

I also dropped anyone whose recorded date of death predates the film’s release, which removes source-material credits such as Bram Stoker for Nosferatu or Mary Shelley for Frankenstein.

I use the release date minus eighteen months as the proxy for when a film was assembled. My per-person agent timeline behind it contains 47,562 records of (person, agency, date range) across 18,757 people, built from trade-press signing announcements in Variety, The Hollywood Reporter, and Deadline dating back to 2010, then overlaid with a May 2026 representation snapshot.

In today’s research, the term ‘agency’ refers only to talent agents. People often have separate representatives for commercials, voice, theatre and literary work, plus separate managers, lawyers and publicists. Counting all of those would multiply the clusters without telling us much.

The chance baseline is calculated using a binomial distribution. For each agency, take its share of all deal-time-repped creators in the cohort that year, work out the probability that a four-person team would land three or more people at that firm by independent choice, then sum across firms. It is recomputed every year, so the growth of the big agencies is already accounted for.

It’s worth repeating that sharing an agency is not automatically the same as being packaged. This measures single-agency concentration at deal-time and the odds that it is a coincidence. It does not, and cannot, confirm a packaging deal or fee on any individual film. Historical and corporate context comes from the trade-press record and the Writers Guild’s own agency agreements.

’ll be uploading new video essays every couple of weeks throughout 2026, so make sure you’re subscribed on YouTube to get notified when another drops (also, subscribing will really help me out with the Algorithm Gods, so it would mean a lot to me - thank you!).

https://www.youtube.com/@TheFilmDataScientist

And to help explain the channel, I also created a short channel trailer:

Rather than celebrate this achievement, some have decided that it’s just not possible and have built a fringe conspiracy theory that he did not actually direct the film. What a world we live in, eh?

Kane himself seems amused by the patently false rumour, and commented on it exactly how you might expect a 20-year-old to - via a pinned YouTube comment in which he said:

This is actually all true. They don’t tell you this, but 96% of all movies released in North America and Europe are actually directed by the same person. They call him ‘The Older Gentleman’. He has had all of Hollywood in a chokehold for decades now.

I looked at the directors’ ages behind 4,690 credits to see how old they were when their movie was released.

Rather than looking at all movies made, I focused on those with at least 50k votes on IMDb. This enables meaningful coverage (many early-stage directors have biographical data available) and lets us focus on professional, working directors.

For those wanting a sense of how famous a film has to be to be included, this filter just about includes the 1998 Very Bad Things starring Cameron Diaz, the 2007 horror film Teeth, and the 1995 Jackie Chan beat’em up Rumble in the Bronx.

Forties for the win

Across the dataset, the median director age at the time of release is 45. Half of all credits fall between 39 and 52. The distribution is right-skewed, which is a data nerd way of saying a few very old men are pulling up the average.

While this is interesting already, what's under the surface of these numbers is even more revealing.

Have directors always been this age?

You might have expected the headline average to have slowly drifted upwards, given that the UK and US working populations have aged over the past forty years, and many creative industries have aged with them. But that’s not what we see in the data.

The second-highest decade for directors’ ages was the 1950s, when the median age was 47. This is the tail end of the studio era, when the people running widely-seen films were largely those who had been running them since before the war. Alfred Hitchcock, William Wyler, John Ford and George Cukor were all in their 50s and 60s, and Cecil B. DeMille was directing into his 70s.

Between the 1950s and the 1980s, the median fell five years, from 47 to 42. This is the studio system’s collapse, the French New Wave’s influence on the Americans who noticed it, the rise of film school as a career route, and a New Hollywood generation who broke in early. Peter Bogdanovich, Martin Scorsese, Francis Ford Coppola, Steven Spielberg, Brian De Palma and George Lucas were all directing widely-seen features in their 20s and 30s. Directors got younger because the gates opened.

This century, we’ve seen a reversal of the trend, and by the 2020s, the median director of a widely-watched film is seven years older than their 1990s equivalent.

So it’s true to say that film directors are getting older… but only if you compare now to thirty years ago. And we’ve now edged just past the old 1950s high, the oldest the director’s chair has ever been

The changing faces of film directors

As new types of people were being given command of commercial films, we have seen the director’s age change. But while one aspect evolved (i.e. ‘hippy types’ took over from the studio old hands), another did not - they were still mostly men.

The 1990s saw an increase in female directors, and in more recent years, we’ve seen another increase in diversity (albeit far from parity). You might reasonably ask how this change affected the average age of directors.

Male directors have got much older (from a 1980s mean of 43 to 50 today), but the average age of female directors has barely moved, hovering around 44 for the past two decades

This means we need to update the finding we stated earlier in the article. Rather than concluding simply that “Film directors are getting older”, we need to say instead that “male film directors are getting older”.

Women are pulling the overall trend down, not up. Without the male credits, the 2020s mean would sit at 50 rather than 49.

Are old people hoarding all the cash?

As with wider society, it is indeed true that most of the money is given to, and held by, the most age-challenged folk.

Joking aside, this actually makes sense. Bigger budgets are generally given to people who have proved their capabilities on previous projects, so they are more likely to be older. In addition, the types of films that can raise the biggest budgets are also those that tend to pick their key creatives from established networks.

Which genres are run by younger directors?

As well as Backrooms, last month also saw the release of box-office smash horror movie Obsession, directed by 26-year-old Curry Barker. This is a slightly more common sight, as horror has the youngest average director age of all genres, at just 40 years old.

Some of this is a function of lower budgets, and it mirrors the audience demographic, in which horror has the youngest adult audience, whereas biographies and dramas have among the oldest.

Further reading

Here are other projects I’ve conducted on age demographics:

• Why are actors getting older?

• Is the over-50 action star boom peaking already?

• How old are horror movie audiences?

• Do older audiences want to watch older actors?

• How do movie genre tastes change with age?

• What films are older cinemagoers watching?

• Can the film industry halt the decline of young audiences in cinemas?

• What percentage of movies are written by women over 40?

Notes

Today’s research examined every feature film with at least 50,000 IMDb votes that was released theatrically since 1930. That gave me 4,690 director credits across 4,203 films, of which I could find reliable birthday years for 95%.

I worked out the age by subtracting the year of birth from the film’s release year. This is coarse (i.e. a director born in December releasing in January will be out by one year) but at decade-scale trends this does not matter.

Production budgets are from OpusData (The Numbers) and are nominal, not inflation-adjusted.

For the avoidance of doubt, this is directors and co-directors, not the second-unit, assistant, art, and casting directors.

For most of film history, people publicly identified with one of two genders, so I have mirrored this binary approach in reporting the numbers. The number of people this excludes is a rounding error, so not relevant to today’s topic.

Do you know the percentage of a film festival’s budget that comes from entry fees? I have heard that it was 9% - what have you heard?

Thirteen years ago, when I interviewed 523 film festival directors, the average was under 14.8% of income.

However, when the Raindance Film Festival opened its books to me three years later, its submission-fee income accounted for 35.6% of its total income.

So I didn’t have a clean figure to provide to Barton off the top of my head. I went digging around for data sources. In the US, most film festivals are non-profits, and non-profits file a public tax return, the Form 990. Buried inside it, on the Statement of Revenue, the festivals list where their money came from.

For example, in 2023, the Portland Film Festival took in $307,194 in income, 82% of which came from fees.

So, we can already see that the answer will differ widely between festivals. Let’s take a systematic approach, looking at the details of two sources of data:

1. 2013 Festival Director Interviews. The answers given by film festival directors in 2013. Each provided a single, self-reported figure for their festival.

2. 2026 Tax Records Analysis. The tax records of US-based non-profit film festivals. Most organisations lump all income into a vague bucket marked “programme revenue”, but a minority break it out into a separately labelled line item, “Submission fees” or “Film entry fees”. Out of around 2,500 candidate organisations, 57 itemised their fee income cleanly enough to use, giving 266 festival-years of data.

The short answer is “it varies enormously”

Using the new data, I found that three-quarters of film festivals earned less than 11% of their income from fees.

The smallest share from fees was just 0.2%, and we’ve already seen the largest, at 82%, for Portland.

The average percentage across these festivals and all the tax records was 10.7%. The median was 4.4%.

One thing worth noting is that the smallest festivals are the ones most likely to rely on fees, and yet they are also the least likely to have detailed tax records.

Has it changed much?

Using the tax records we can look back over the past decade. This reveals that the median festival has fluctuated between 3% and 7% over this period.

How are film festivals funding themselves?

At this point, you might be wondering how festivals stay afloat, if filmmaker fees are such a thin slice.

Festivals earn their money from three types of income:

1. Submission fees paid by filmmakers

2. Donations, grants and sponsorship

3. Commercial income, mostly ticket sales, but some do have merch.

Fees are the smallest of those three types. Below, I’ve broken it down by festival age and put both datasets side by side: the 2013 director survey and the 2026 tax records.

The oldest festivals rely on fees the least. Veterans take just 7% of their income from fees in both datasets because they’ve had the time to build donor and regular audiences that younger festivals are still chasing.

My reading of the data reveals four kinds of film festivals:

1. The donor-funded are the mainstream, getting the bulk of their income from grants and gifts and only a few percent from fees.

2. The sponsor-and-ticket festivals are the biggest, covering their costs with a mix of donations, box office and sponsorship.

3. The small-and-fee-leaning are tiny operations scraping a third from donors and leaning on fees for close to a fifth of their income.

4. The fee-funded are tiny, donor-less, and running on filmmaker money for three-quarters or more of their income.

The biggest of the “Donor-funded” cohort is Sundance. Its audited statements reveal that fees account for only around 2% of total income. Sundance processes roughly 15,000 submissions a year and earns far more from sponsors, donors, and ticket sales, so the $1 million it collects from submission fees is a rounding error on its books.

Cork International Film Festival, Ireland’s oldest, is a useful contrast as a public-arts festival rather than a US non-profit. Its published accounts have held steady between 8% and 11% over six years.

How much money are we talking about?

I have focused on the percentage of fees earned, rather than the dollar amount, for two reasons. Firstly, that was the focus of Barton’s initial question, but also because I think it’s a more meaningful number.

However, I’m sure I will be asked about the numbers at play, so here’s a quick overview.

Last year, I studied the cost of a single entry to a film festival, discovering that the average fee ranged from $30 (for an early-submitted short film) to $87 (for a late-submitted feature film).

In today’s new research on tax returns, the festival that reported the highest income from fees was Santa Barbara, at close to half a million dollars. But Santa Barbara is a multi-million-dollar operation, so that half-million is only about 4% of its income. (There will be many festivals earning more, but their tax reporting prevents us from knowing the figures).

The Dumbo Film Festival has a much lower total but is at the other end of the spectrum. They reported that in one year all of its $51,964 funding came from fees, and 72.7% was paid out to just three people.

Below are the festivals that had the largest declared income from fees.

So what of ‘scam’ festivals?

Filmmakers have been warning each other about ‘scam festivals’ for well over a decade. These are events that exist mainly to trick filmmakers into paying to submit, under the false premise that the festival is well-respected, useful, or even real.

The concern is neither new nor fringe. Back in 2019, Screen Daily reported calls for a festival regulator after a run of scam events, and noted that FilmFreeway had already delisted 42 of them. The Hollywood Reporter has documented how below-the-radar festivals prey on struggling filmmakers, including a horrifying story of a writer-producer who borrowed money to fly from New York to London for her ‘UK premiere’ and found a near-empty room. One Rhode Island showcase ended up branded the ‘Fyre Festival’ of film festivals.

It helps to picture a spectrum. At one end sit the outright frauds, which take your fee for an event that never happens. At the other side are the ‘award mills’, which are technically real but exist to manufacture laurels. They run rolling monthly deadlines, nominate almost everyone, spin up a dozen city-branded editions of the same event, and then sell you a certificate, a trophy or a ‘social media package’ once you’ve ‘won’. The laurel, not the screening, is the product.

Sadly, the tax data I’ve been using can’t really tell the difference between a ‘real festival’ and one that just farms fees. And to be fair to the small festivals at the top of my charts, a festival funded almost entirely by fees isn’t necessarily doing anything wrong. It may simply be a small event in a town with no arts money and no wealthy donors, funding itself the only way it can.

Similarly, the fact that most of the income goes to a small number of people is not automatically suspect. We have all met the person who makes a community event happen through sheer dedication and hard graft. If they secure venues, volunteers and resources for free, then paying them for their time is good value, not a red flag.

So how do you tell the two apart? The community has settled on a familiar set of warning signs: dozens of award categories, near-automatic nominations, vague or shifting venues, screenings that turn out to be ‘online only’, and a flurry of upsells the moment you’re selected.

There are good rundowns of the red flags from Noam Kroll, Submitting to Film Festivals and Show Me Shorts. The single most useful habit is to do your due diligence before you submit your film each time. Check that the venue exists, look for photos and press from past editions, and ask what will be screened if you’re accepted.

The platforms have, belatedly, started to respond. FilmFreeway now runs ID checks, demands venue proof, and holds a new festival’s first-year payouts until it can prove the event actually took place. Filmmakers have petitioned for more, and bodies such as the Universal Film Festival Organisation have launched specifically to stamp the practice out.

The voice of the festival directors

I don’t want to end on a negative note, as the vast, vast majority of film festivals are working hard and doing well.

A filmmaker who is starting to submit to festivals might be forgiven for initially decrying the amount of money it costs to submit to festivals, but once you dig into what it takes to put festivals on, the fees seem far more defensible.

Running a film festival is a lot like making a film. I.e., it makes no rational sense, so the only people who do it are those who are most passionate, most resourceful, and who really believe in the mission.

In preparing today’s article, I reached out to a number of the festivals in my dataset, especially those leaning most heavily on fees, to ask how it looks from their side of the desk.

Josh Leake, Executive Director of the Portland Film Festival, rightly pointed out that tax filings only capture cash, not the volunteer and community support that keeps a small festival running. He said:

The submission-fee revenue can appear larger on the filing than it would if one were looking at the festival’s broader support base as a whole.

Jeff Ross of SF IndieFest explained why a festival’s fee reliance can climb without anyone raising a single fee.

Before the pandemic, our revenue was mostly from ticket sales, with some grant money and submission fees making up a much smaller percentage. The percentage of revenue from submission fees has gone up, but that’s mainly because our revenue from ticket sales has declined so much. The overall event budget has declined about 50% since 2020. We don’t have sponsors, donors or a membership program. We are truly a grass roots DIY arts org.

At the other end, Leslie Raymond of the Ann Arbor Film Festival, whose fee reliance has actually fallen over the decade, said:

Submission fees are an important part of our funding mix, but our goal has always been to build a sustainable festival through a diverse range of revenue sources rather than relying too heavily on any single one.

And Kitty Boone of Aspen Film showed how one organisation can do both. Its feature festival is invitation-only, with no submission fees at all, whereas its shorts festival, Aspen Shortsfest, charges one, because:

…we receive well over 3,300 submissions, [and] it requires a lot to program a festival such as this.

The same themes came through thirteen years ago, in the anonymous answers to my original survey:

“It is a labor of love. No one is getting rich off your entry fees.”

“Generally, I think filmmakers believe that festivals have more money than they have… there usually is no ‘profit’ to share. The vast majority of film festivals barely get by.”

“There is very little funding, and small festivals like us rely on the nominal entry fees to let us run a festival.”

“We know people don’t like paying submission fees, but without them our festival (and I assume many others) would not be able to make ends meet.”

Notes

I built the festival list via ProPublica’s Nonprofit Explorer and the IRS Business Master File. That gave around 2,500 candidate organisations. I matched them to the IRS e-file index and read the underlying Form 990 XML directly. 57 festivals itemised submission or entry fees on their own line (Form 990, Part VIII, line 2), giving 266 festival-years across the 2015 to 2024 tax years. The denominator throughout is total revenue (Part VIII, line 12).

I only used a festival year where entry fees sat on their own labelled line. Where a festival bundled them into something mixed, like “Tickets and entry fees” or a generic “Festival fees”, I left it out rather than guess.

These are fees as booked (i.e. gross, not net), before FilmFreeway’s commission of around 12%, payment processing, sales tax or fee waivers. The cash retained is somewhat lower, so if anything, these figures slightly overstate reliance.

Organisations below the filing threshold submit a Form 990-EZ or a one-page postcard, neither of which itemises fees. Of 609 festivals that filed the short form, only about one in eighteen mentioned submission fees anywhere, and never as an isolable number. They are exactly the festivals most likely to live on fees, so the true long tail sits higher than this sample, not lower.

Sundance and Cork come from their own published accounts, not the IRS data. Sundance’s audited statements carry a “Submission” line and Cork’s accounts, filed as an Irish charity, carry a “Submission Fees” line.

A few readers reached out to ask whether I had taken the films’ quality into account. I had, but what stuck with me was how many people clearly assumed that a famous cast meant a better film. The cynic in me assumed the opposite!

Let’s briefly consider the two perspectives, and then see which the data reveals to be correct.

Fame = quality?

You can see why someone might think that famous actors would be in better films.

Famous actors are, on the whole, richer and far better supported than those still on the way up. They have readers, managers and agents whose job is to find good scripts. Not only can they seek them out, but they will also be constantly offered more of them because most filmmakers want/need a famous name in their film.

They also have power - Tom Cruise can pretty much star in almost any film he wants (if he is willing to be flexible on his fee, of course).

So it might be reasonable to conclude that more famous actors tend to make better films.

Fame ≠ quality?

But there is a counterargument that famous actors are not trying to make good films. They are trying to make films that are good for them. This can be a very different thing, tied to their brand and money.

When asked about his role in panned Jaws: The Revenge, Sir Michael Caine famously said:

I have never seen the film, but by all accounts, it was terrible. However, I have seen the house that it built, and it is terrific.

It’s a reasonable assumption that the more famous the cast, the bigger a film’s budget will be, but I doubt many people instinctively believe that bigger budgets automatically mean better scripts.

Then we have to consider that the biggest stars tend to work on films with the most competing forces around them, such as marketing, release-date strategy, and merchandising.

None of these forces aims to improve the film, and some directly pull against it.

Scores ≠ quality?

One thing to note upfront is that we cannot actually measure whether a film is good. We can only measure whether people liked it. Today I’m going to be using two quick proxies:

1. Metacritic score, which averages professional critics’ reviews into a number out of 100.

2. IMDb user rating, which captures the general audience’s verdict out of 10.

So in truth, we’re tracking reception rather than art, but they are the best signals we have of how a film landed. At the very least, they let us see which factors correlate and which do not.

I crunched the numbers on 2,756 feature films released over the past quarter-century to see if there is a connection between how audiences and critics rate them and the cast’s star power. The details of my methodology are in the Notes section at the end.

A famous cast is a weak sign of quality

The core finding is that fame is a pretty bad proxy for whether a film is likely to be any good.

Films led by a top-tier name score ever-so-slightly higher than those led by a non-famous actor, with both audiences and critics, but the gap is small and only with the most famous names.

Top-tier-led films average 6.65 out of ten with audiences, compared to 6.28 for ordinary-led films, and a similarly modest step with critics. Once you account for the film’s budget, genre and era, the famous cast still helps, but only barely.

It depends heavily on the genre

That faint average hides a sharp split by genre. A famous cast is a meaningful hint of quality in some kinds of film and no hint at all in others.

In adventure, a more famous cast tends to go with better-reviewed films. Everywhere else, the link is weak, and in drama, romance, and horror it is essentially zero, even tipping slightly negative among critics. These are exactly the genres that live on concept, script, and word of mouth, where a familiar face adds little and can even raise suspicions that a film is relying on its casting.

Over the entire period, the cast-quality link has remained weakly positive, neither strengthening nor collapsing, hovering between +0.08 and +0.22 across eras.

Which actors are a good sign of quality?

The link is much stronger with some actors than with others. Some actors appear, again and again, in films that audiences and critics both rate highly. Others have built long careers in films that neither does.

The scatter plot below places each of the 2,730 famous actors I looked at on a chart of average IMDb and Metascore for their past films. They fall into four camps:

• Everyone’s favourites are in the top right, including Leonardo DiCaprio, Tom Hanks, Cate Blanchett, Denzel Washington, Christian Bale, and, among newer arrivals, Florence Pugh.

• Critical darlings are in the top left, and include Greta Gerwig, Robert Pattinson and Kristen Stewart, whose choices lean towards the kind of film reviewers admire more than the multiplex crowd.

• Crowd-pleasers sit in the bottom right, containing actors whose films are appreciated by audiences but much less so with critics, such as Will Smith, Ryan Reynolds, Sandra Bullock, Liam Neeson and Gerard Butler.

• The… er… others. In the bottom corner, we have actors whose films are poorly rated by both audiences and critics, such as Adam Sandler, Nicolas Cage, Bruce Willis, Dwayne Johnson, and Jason Statham.

Adam Sandler appears completely unbothered by how his movies are reviewed, saying in 2013:

I didn’t get into movies to please the critics. I got into it to make people laugh and have fun with my friends.

Others in that final quadrant draw a clearer distinction between the film and their contribution to it. When IndieWire asked Nicolas Cage about the run of films he made to pay off a multi-million-dollar debt, he said:

Even if the movie ultimately is crummy, they know I’m not phoning it in, that I care every time.

If you do want to seek out good films based only on the famous names starring in them, below are the people with the famous names who have the highest correlation with quality:

1. Leonardo DiCaprio

2. Andy Serkis

3. Florence Pugh

4. Philip Seymour Hoffman

5. Warwick Davis

6. Carey Mulligan

7. Viggo Mortensen

8. Clint Eastwood

9. Tilda Swinton

10. Helen McCrory

11. Benedict Cumberbatch

12. Tom Hanks

13. Andrew Garfield

14. Quentin Tarantino

15. Christopher Lee

16. Domhnall Gleeson

17. Brad Pitt

18. Tom Hiddleston

19. Léa Seydoux

20. Tom Cruise

Bonus Finding

As I had the data to hand, I took a quick look at what factors do seem to matter.

The strongest single signal is the film’s runtime, albeit we’re still not in a ‘usefully predictive’ space. R-rated, adult-aimed films score higher than gentler ratings, and films where the director also wrote the script score higher, especially with critics.

Everything I looked at sits in the shaded zone, where the effect is small. I have sorted them by signal strength, meaning that the cast’s fame is among the least predictive.

The fact that longer films score higher is pretty consistent, rather than just the average being pulled up by the canons of Christopher Nolan and Martin Scorsese.

Notes

Today’s research looks at 2,756 feature films released between 2002-24 with:

• A budget on record,

• A worldwide box office figure

• A lead whose fame could be measured before release

• Both an IMDb user rating and a Metacritic score.

The actor chart uses a broader set of 2,730 prominent actors, each with at least fifteen rated theatrical films released since 2000.

Cast fame is a relative score for each film’s lead, built from publicly visible popularity signals gathered across several websites in the year before release, ranked against films of the same era.

Quality is the IMDb audience score (out of 10) and the Metacritic critics’ score (out of 100). Both are gathered after release, so they describe how a film was received rather than some inherent quality fixed in advance.

To be featured on the scatter plot, the actors had to have been among the 1,000 most famous actors at some point and to have appeared in at least 15 films released since 2000 that had both an IMDb audience score and a Metascore.

The signals chart shows a regression of each score on cast fame, budget, runtime, producer count, franchise status, adaptation, rating, whether the director also wrote it, animation, genre and era. Bars show each factor’s strength on a standardised scale, allowing them to be compared directly. With several thousand films, even small bars are statistically real rather than chance, so the shaded zone simply marks effects small enough to matter little in practice. Runtime and budget are associations, not proven causes.

The actor chart leans towards prominent, mostly English-language leads with enough films to be measured. Films that neither audiences nor critics have scored are absent, which skews the set towards titles that got a real release.

The list of the highest correlated stars at the end focuses on actors (not pure voice actors) and the more famous names.

Further reading

I’ve written on the topic of quality a few times over the years, including:

• Whether film length signals quality,

• Which actors appear most in excellent and awful films,

• Whether fewer producers make better films,

• Whether writer-directors make better films,

• Whether the timing of reviews predicts quality.

It might be the oldest argument in the film business, and the two sides go like this:

• One camp says a famous face is the closest thing to insurance a producer can buy. Audiences seem to see them as consistent brands in which every film is a prototype and genre labels are fuzzy at best.

• The other camp points out that the famous names are only there to protect the risk-averse gatekeepers within the production and distribution supply chain. Once movies hit audiences, they just want it to deliver and be entertaining.

While Obsession may have provided a powerful visual, the plural of anecdotes is not data. I went back to the data, studying the box office performance of 3,052 films this century, in relation to the fame of their lead cast (more details in the Notes section at the end of the article).

Does having a star help your film succeed?

Given that we can’t know how much an actor is paid upfront (or how much of the income they also get), we can’t talk about true value for money. For every star, there will be a price at which their involvement is a net positive to the bottom line, and after which they cost more than they help.

But we can measure the difference in performance between films with famous, semi-famous and non-famous names. For this, I am using the worldwide box-office multiplier relative to the budget. Let’s not worry too much about the exact numbers; instead, focus on the difference between the cohorts.

Films led by an actor in the top 10% of fame for their era grossed a median of 2.54 times their budget worldwide. Films led by someone in the 50th to 75th percentile, the group I am calling ‘Low Fame’, grossed 1.91 times.

So we can see that fame does have an effect.

The difference between ‘not famous’ and ‘kinda famous, I guess’ is nothing. It’s only as we start to talk about actors in the top quarter that we see an uplift, with the largest impact on the biggest name.

Given that those are also the highest-paid actors, it’s fair to assume that some of that uplift doesn’t go to the producers, studio, or investors, instead ending up in the pockets of the actors (and their agents, managers, stylists, doctors, and spiritual gurus).

Are there other ways having a famous name helps?

So many of the decisions in the film industry are made in response to perceived risk. Some can be seen as professionals making smart, informed decisions, but in a large number of cases, it’s people acting instinctively, reactions rooted in fear. So it makes sense that having a famous name can feel reassuring.

So while the previous graph SHOULD be the most important one, I suspect the one below is more interesting to film professionals. It shows that having a famous name helps reduce the risk of an embarrassing box-office failure.

A star does not guarantee a hit, but the floor under the downside is noticeably higher.

Where fame pays, and where it barely matters

But these effects are not spread evenly across genres. Obsession appeals to a younger, horror-friendly crowd.

When we split our data by genre, we find that while an A-lister is associated with higher returns across every major genre, the size of the effect varies widely.

The biggest payoff is in romance, where top-tier-led films return a multiplier of 3.82 versus 2.08 for lower-tier leads. A romantic film is largely a question of whether audiences want to spend two hours in a particular person’s company, and a recognisable, well-liked lead is close to the whole product.

Comedy, drama, thriller and adventure sit in a tighter band of roughly half a multiplier.

The genre where fame does least is horror, where the gap is only about 0.25x. This is the same pattern that Bruce Nash and I found when we looked at 877 independent films and the odds of a theatrical release a decade ago. Horror was the genre least dependent on a name and most dependent on concept.

The franchise does most of the lifting

Eagle-eyed readers may have noticed I didn’t comment on the effect of action in the section above. Action had the second largest effect of a star, after romance. However, it is worth asking which famous leads those are actually appearing in. Increasingly, the answer is a franchise.

The share of top-tier-led action films that belong to an established franchise has climbed from around 40% early in the period to about 79% in recent years.

In top-tier action, films in a franchise return a median ratio of 3.47, while those that are not return 1.99. The famous winners are nearly all franchise instalments, from Spider-Man: No Way Home and Jurassic World to The Hunger Games and Aquaman. The famous flops are the films that asked a star to carry a brand-new idea on their name alone: Jonah Hex (less than a quarter of its budget back), Argylle and Mortdecai. The famous name and the franchise usually travel together now, and when you can separate them, the franchise is doing most of the lifting.

A study from 15 years ago concluded that the revenue gap between star and non-star films was driven almost entirely by those films having bigger budgets and wider releases, not by audiences turning up for the face. It is also why some people contend that intellectual property is killing the movie star, and why studios and actors openly disagree about who is actually generating the money, the person or the logo.

Two bonus findings

Before we go, I want to share two minor findings I came across along the way that I found interesting.

The first is that the value of a famous name is stackable:

• Films with no top-quartile fame grossed a median of 2.04 times their budget.

• Add one famous lead, and that rises to 2.30.

• Put two genuinely famous names at the top of the call sheet, and it reaches 2.91.

I hear a few people in Cannes this year complained that while they already had a famous name in their proposed film, sales agents and distributors were not interested until they added a second famous name. While I appreciate how frustrating that must be for the filmmakers, the data seem to support what these sales professionals have been saying.

Secondly, it’s not just how famous the actors are in the moment; it’s also about the wider story of their fame.

Among leads who were already famous, I split them by whether their public attention was rising, steady or falling in the year before release.

Rising stars outperform. At the same level of established fame, films whose lead was climbing returned a median 2.69, against 2.28 for steady names and 2.21 for those whose moment was fading.

Catching an actor on the way up, when audiences are actively getting more interested in them, beats booking an equally famous name on the way down.

Notes

Today’s research examines 3,052 films released between 2000 and 2025, each with a recorded production budget, a positive worldwide box office, and a lead actor whose fame could be measured before release.

Budgets and box office come from The Numbers. The sample leans towards English-language theatrical releases, because those are the films whose budgets are publicly reported. There is a growing (and worrying) trend of reduced budget availability, which I’ve written about before.

In addition, today’s research tracks the worldwide theatrical box office. This is a useful signal, but it doesn’t capture all of a film’s income, and over the period I studied, the industry's economics have shifted.

That said, given that my numbers are comparative rather than trying to be comprehensive, I don’t feel that either effect is skewing the overall findings.

For each film, I built a relative fame score for its lead actor from a blend of publicly visible popularity signals gathered across several sources, averaged over the year ending one month before release, so the film’s own launch could not inflate it. Each film took the score of its most famous lead, and films were ranked against others from the same five-year window, so ‘top 10%’ means famous for its time rather than famous today. Leads were also classified by whether that attention was rising, steady or falling over the same window.

The median ratio is used throughout because the average is skewed by the occasional billion-dollar hit. Some genre and tier cells are small, in particular top-tier horror (10 films) and romance (22 films), so those specific figures are best read as indicative. Stars are also cast into better-financed films in the first place, so part of any apparent star effect is really a selection effect, which is exactly why the franchise comparison matters.