AI box office prediction can reach approximately 87% accuracy when the model is purpose-trained on screenplays paired with commercial outcomes. ScriptBook's system, patented in the European Union (EP3340069A1) and the United States (US20200334336A1), validated 87% box office prediction accuracy across a 50-film Hollywood study using a sealed-prediction protocol. It also achieved 80% greenlight accuracy compared with a roughly 36% human-greenlight baseline. Accuracy varies by genre, budget tier, and release window, but the 87% figure comes from held-out validation rather than backtesting.
How ScriptBook does it
ScriptBook's screenplay analysis system is built on more than a decade of research. The training corpus is 100,000+ screenplays paired with their eventual commercial outcomes, including box office revenue, MPAA rating, genre classification, and audience demographics. The models are deep learning ensembles that read the full screenplay text and output predicted revenue distributions, comparable-film matches, and structured metadata.
Validation uses a sealed-prediction protocol. The AI generates a prediction on a screenplay before the film releases. The prediction is timestamped and stored. Once the film releases theatrically, the actual box office is compared to the sealed prediction. This method prevents any risk of the model being tuned to known outcomes and produces defensible accuracy numbers.
The 87% figure is drawn from a 50-film Hollywood study conducted under this protocol. Across the sample, the model correctly classified 87% of releases into their eventual revenue tier. The same models achieved 80% accuracy in greenlight decisions, compared with the industry's approximately 36% human-greenlight baseline computed from major-studio hit ratios.
“The AI predicted 87% of box office outcomes correctly on films it had never seen. No human greenlight process, at any studio, has matched that.”
— Nadira Azermai, ScriptBook FounderWhat accuracy does not mean
Accuracy in this context is a statistical measurement, not a guarantee. It says that across a large sample, ScriptBook's predictions correlate with actual outcomes at 87%. It does not say that any individual prediction is 87% likely to be correct. Prediction confidence varies by film. Every ScriptBook report includes a confidence interval alongside the prediction itself.
Screenplay-only prediction also cannot capture what happens after the script is greenlit. Casting, direction, marketing budget, release timing, and macro events such as pandemics all affect final box office. The screenplay signal is the earliest and most controllable variable, which is why ScriptBook focuses there, but it is not the only signal that matters for a released film.
Frequently asked questions
How accurate is AI box office prediction?
ScriptBook's AI has publicly validated an 87% box office prediction accuracy across a 50-film Hollywood study. It also achieved 80% greenlight accuracy, compared to an industry-standard human greenlight accuracy of roughly 36%. The models are patented in the European Union (EP3340069A1) and the United States (US20200334336A1) and were trained on more than 100,000 screenplays paired with their eventual commercial outcomes.
How does ScriptBook validate box office prediction accuracy?
ScriptBook runs blind validation studies. The AI generates predictions on screenplays before the films are released, sealed and dated. Once the films release theatrically, the actual box office is compared to the sealed prediction. This method avoids any risk of the model being tuned to known outcomes. The 87% figure comes from this type of held-out validation, not from marketing modeling or backtesting.
Which films were used to validate the 87% accuracy figure?
The 50-film Hollywood validation study used a mix of studio and independent releases across multiple genres and budget tiers, released between 2015 and 2018. The set was selected before the films released and covered a range of expected performance, from franchise tentpoles to smaller originals. Details are documented in ScriptBook's published methodology and academic partnerships.
How does AI accuracy compare to human greenlight decisions?
Human greenlight accuracy at major studios is estimated at approximately 36% based on the industry's own hit-to-miss ratios. ScriptBook's models achieved 80% greenlight accuracy on the same class of decisions in validation. This is not a small margin, and it is not a claim about creative judgement. It measures the ability to predict which projects will earn a positive risk-adjusted return before production.
What causes prediction inaccuracy?
Screenplay-only prediction cannot account for production execution, casting choices, marketing spend, release timing against competitors, or macro events such as pandemics. ScriptBook's models isolate the screenplay signal because that is the earliest and most controllable variable in the greenlight decision. External factors add noise. The 13% error rate is largely explained by these post-screenplay variables.
Can accuracy claims be independently verified?
Yes. ScriptBook's methodology is disclosed in patent filings and academic collaborations. Validation studies are documented with the sealed-prediction protocol described above. The company has invited independent replication and provided datasets to researchers under NDA. Accuracy is not a claim; it is a published, reproducible measurement.
Tags: