AI startup Quilty faces scrutiny after script prediction tool misfires on box office hits
The platform, which combines multiple large language models to score unproduced scripts, has drawn criticism for incorrectly predicting the flop Christy would outperform the blockbuster Sinners.
AI startup Quilty, founded by producers Simon Horsman and Daniel Wood, has launched a service designed to score unproduced film scripts for commercial viability and narrative quality. The platform aggregates various large language models, including Gemini, DeepSeek, Claude, and ChatGPT, to generate reports on budget, story beats, and character analysis. Despite claims that the tool can predict box office success and identify cultural resonance, early testing revealed significant inaccuracies, such as predicting the flop Christy would outperform the blockbuster Sinners. Critics describe the technology as a jumbled mix of pre-existing systems lacking genuine analytical ability, while founders maintain it is designed to assist human decision-making rather than replace it.
Quilty’s pricing model costs $50 per individual analysis, with discounted rates available for multiple analyses. The founders claim the tool can accurately determine how a project “addresses the cultural moment,” citing Revenge of the Nerds as an example of a film that would receive a lower score due to modern sensitivities regarding its depiction of sexual assault. Horsman and Wood attribute the incorrect prediction for Christy versus Sinners to Sydney Sweeney’s star power and the lower production costs of biographical dramas compared to fantasy features.
The company uses VADER (Valence Aware Dictionary and sEntiment Reasoner) as part of its “sentiment engine” to measure text positivity versus negativity. This open-source software helps assess the degree to which text comes across as positive or negative, serving as a proxy for audience reaction in the absence of genuine human intuition. The founders argue this approach, combined with the aggregation of different models, provides a comprehensive view of a script’s potential.
Quilty does not train its own models but relies on context prompting to reduce hallucinations, allowing for easy integration of new models like the upcoming “Claude Mythos.” This modular approach allows the company to swap in superior systems as they become available, with co-founder Daniel Wood noting that the strategy ensures the software improves as underlying technology advances. The platform uses Gemini for structure and patterns, DeepSeek for financial modelling, and a combination of Claude and ChatGPT for narrative and character analysis.
While the founders position Quilty as a tool to democratise access to development insights, the technology has yet to prove its analytical depth. The company admits it cannot foresee unforeseen factors, such as an actor’s fall from grace or viral internet phenomena, which can significantly impact a film’s performance. Nevertheless, Horsman and Wood maintain that the tool is intended to provide as much information as possible to help writers, producers, and financiers make informed greenlight decisions.
