llms.txt
Compare with other sites' llms.txt files →# Hushcrasher
> Hushcrasher is a research-driven intelligence firm designed to understand the video game market, predict its outcomes, and optimize its key decisions. We measure. We predict. We optimize.
Hushcrasher combines exclusive data, proprietary machine learning models and methods developed with academic partners to deliver insights, tools and custom research to studios, publishers and investors. Founded in France by Julie Belzanne (CEO; formerly in venture capital and fundraising advisory, then head of a video game accelerator and member of the leadership team of a 30-person studio) and Antoine Mayerowitz (CTO, PhD in economics, Paris School of Economics).
## Free tools
- [Steam Regional Pricing](https://hushcrasher.com/tools/steam-regional-pricing): regional prices for Steam from IMF purchasing power parity, US sales tax and psychological price points; exports a Steamworks CSV.
- [Game Budget Benchmark](https://hushcrasher.com/tools/game-budget-benchmark): realistic budget range (q10–q90) for a game, from a model trained on hundreds of disclosed budgets and applied to more than 100,000 Steam games.
## Research
Hushcrasher's research areas are continuous lines of research, each with a public page on what it does, its methods, its applications and where it is used.
- [Game Taxonomy](https://hushcrasher.com/research/game-taxonomy): Hushcrasher groups video games by clustering representations of all the data available on each game (free-form text, images and structured data), with unsupervised and semi-supervised machine learning, instead of sorting them into predefined categories.
- Why: the industry classifies games with store metadata (genres, tags, features), and most market benchmarks rely on it alone. Store metadata has no field for production scope, art direction, narrative theme, niche or player audience. It is chosen by developers and publishers, so it reflects marketing intent and can differ from how players see a game. As a predefined list, it misses emerging niches, sub-genres and visual styles, rests on conventional thresholds (such as where AA ends and AAA begins), breaks time series when a store adds or removes a tag, and puts each game in a single category.
- Data: free-form text (store description), images (capsule and screenshots) and structured data (tags and their rank, genres and features, credits, install size, launch price, developer and publisher track record).
- Outputs: clusters by axis (production scope as Kei, Midi, AA or AAA; genre and niche; art direction; narrative theme; player personas, in development), the closest comparable games to any title, and clusters re-estimated by release cohort to track when groups appear, grow, split or merge.
- Paper: [A Semi-Supervised Taxonomy of Game Production Scope](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5429396) (Antoine Mayerowitz and Julie Belzanne, 2025), on production scope.
- [Production Modeling](https://hushcrasher.com/research/production-modeling): Hushcrasher estimates the development budget (marketing excluded) of every game on Steam with a model trained on hundreds of real budgets, and models whether games get finished (in development).
- Why: development budgets are rarely disclosed. Disclosed budgets are not a random sample (large productions, exceptional cases and solo developers' games are overrepresented) and use different definitions. Rules of thumb (team size × monthly cost × development time) need internal data unknown from outside and assume one cost per person across countries and roles. Datasets built on released games miss cancelled projects.
- Data: real budgets from public sources (press coverage, developer post-mortems, conference talks, web archives, court filings) and from studios under NDA, converted to US dollars at the release year's exchange rate, deflated to constant dollars, and weighted by the inverse density of their value so that often-disclosed budget levels do not skew the estimates.
- Model: a stacked ensemble of gradient boosting, elastic net and kernel ridge regression, trained on the log of the budget. It uses observable data rather than internal production data: team size, roles and experience from the credits, studio and publisher track record, install size, launch price, average playtime, platforms, languages, 2D or 3D, multiplayer. Accuracy is measured on real budgets kept out of training. Where a real budget is known, it replaces the estimate.
- In development: a version of the model that uses Hushcrasher's salary database (more than 5,000 salaries from 2019 to 2026, across more than 15 countries) to account for country-specific labor costs; a completion model built on data on cancelled projects.
- Used by the Game Budget Benchmark, The State of Profitability in the Video Game Industry, the Comparable Games Market Analysis, and doctoral research at Harvard University.
- [Market Modeling](https://hushcrasher.com/research/market-modeling): Hushcrasher's research on the economics of the video game market: how players choose between games, how producers set their prices, and how the two interact, to answer business questions with simulations. In development: no results published yet, but its first building blocks already inform Hushcrasher's pricing advisory.
- Why: market decisions (price, discounts, acquisitions) rely on tools with known limits. Competitor price averages say nothing about how demand responds to price. Benchmarks and ratios such as wishlists to sales describe past outcomes and cannot simulate a decision not yet taken. Better games are often priced higher and sell more, so comparing prices and sales across games understates price sensitivity. Store tags do not measure which games actually compete.
- Data: two levels of Steam data, game-level data (monthly market shares, net prices, discounts, game characteristics) and player-level purchase histories (wishlists, games owned and when each was acquired).
- Methods: a structural model of demand, an explicit model of how players choose which game to buy and when, given current prices and the discounts they expect, estimated on store data and on the behavior of individual players; it measures price sensitivity and how far ahead players anticipate future prices, and can simulate situations never observed, such as a merger or a subscription. It is complemented by causal machine learning, to estimate the effect of a decision such as a price change directly from observed variation, and policy learning, to derive the best decision for a given game. Carried out with Anatole Cheysson, economist at ZEW, and advised by Felipe del Canto Monge, from Harvard University.
- Applications: price elasticities and substitution between games; the revenue-maximizing price at launch and over a game's life, including discounts; defining a market from substitution; merger and acquisition simulations; platform-wide decisions such as a subscription offer and its price, revenue share or discount rules, and their effect on total revenue, on how it is split between producers and the platform, and on what players buy.
- [Papers](https://hushcrasher.com/research/papers): Hushcrasher's research papers, with how to cite them.
- [Partnerships](https://hushcrasher.com/research/partnerships): Hushcrasher works with academic researchers on the economics of the video game industry.
- Harvard University: Felipe del Canto Monge, PhD candidate, uses Hushcrasher's development budget data in doctoral research on release strategies, and shares his expertise on Market Modeling.
- ZEW (Leibniz Centre for European Economic Research): Anatole Cheysson, economist, works with Hushcrasher on Market Modeling.
- Academic researchers can request access to Hushcrasher's datasets and model outputs for non-commercial research, under a data use agreement; publications cite Hushcrasher, and co-authorship can be discussed.
## Methods
- Regional pricing: a price for each country from IMF purchasing power parity exchange rates (World Economic Outlook, updated twice a year) or from a Netflix-like price index, adjusted for the average sales tax in the US and Canada and rounded to psychological price points. Countries that share one Steam price are combined with weights based on Steam traffic.
## Reports
- [All reports](https://hushcrasher.com/reports)
- [Comparable Games Market Analysis](https://hushcrasher.com/reports/comparable-games-market-analysis): the 200 released Steam games closest to a project and how they performed (revenue distribution, revenue-to-budget, country exposure). On demand, requested by email, $499 excl. VAT. Gamevestor, an investment platform for video game projects authorised by the French AMF, uses it as a standardized first read on every project that reaches its selection committee, and its investors can request it for free on selected projects; a commercial due diligence (see Advisory) goes much deeper into a specific deal.
- [The State of Profitability in the Video Game Industry](https://hushcrasher.com/reports/state-of-profitability): what share of games recover their development budget, by genre, budget and strategy. Release February 2027; $1,999 excl. VAT before release ($2,999 after).
- [Game Industry Salary Report 2026](https://hushcrasher.com/reports/game-industry-salaries): what game developers earn by role, seniority, country and gender. Free; December 2026.
- [Game Scope Classifier](https://hushcrasher.com/tools/classification): a chart and a guide to place any game in the Kei / Midi / AA / AAA production-scope taxonomy from its credits and its size on disk. Free.
## Advisory
Dedicated expertise for studios, publishers, venture capital and M&A, plus tailored research on request.
- Pricing optimization: the price that maximizes a game's lifetime revenue, from a model of what players will actually pay, using the building blocks of Market Modeling as they come out, rather than an average of competitors' prices.
- Pre-launch revenue projection: sales scenarios for a game before it ships (best case, worst case and the odds of each), going beyond a wishlist-to-sales ratio and built on the simulation methods banks use to model risk.
- Commercial due diligence: an independent read before buying, selling or funding, pressure-testing revenue projections, market dynamics, competitive pressure and budget realism.
## Articles
- [Newsletter and articles](https://newsletter.hushcrasher.com)
- [How much does it cost to make a video game?](https://newsletter.hushcrasher.com/p/how-much-does-it-cost-to-make-a-video): a model that estimates the budget of 100,000 games, and what it shows.
- [Why video games cost so much to make](https://newsletter.hushcrasher.com/p/why-video-games-cost-so-much-to-make)
- [Did games really get more costly to make?](https://newsletter.hushcrasher.com/p/did-games-really-get-more-costly)
- [By the way, what's a AA?](https://newsletter.hushcrasher.com/p/taxonomy-of-games): the Kei / Midi / AA / AAA taxonomy explained.
- [A Semi-Supervised Taxonomy of Game Production Scope](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5429396) (SSRN, 2025)
## Founders
- Julie Belzanne, co-founder and CEO: master's degree in Technology and Innovation Management (MINES ParisTech and Université Paris Dauphine).
- Antoine Mayerowitz, co-founder and CTO: PhD in economics from the Paris School of Economics, with a thesis on innovation and competition supervised by Nobel laureate Philippe Aghion.
## Contact
- [Contact](https://hushcrasher.com/contact)
## Reuse and attribution
You may quote, copy, summarize, translate and reuse the figures, charts, methods and text published by Hushcrasher, in any medium, provided you credit Hushcrasher and link to the source page on hushcrasher.com.