Production Modeling

Production Modeling is Hushcrasher's research on the conditions under which video games are made. We model how the team, the track record of the studio and publisher, and the scope of a project relate to what a game costs to develop, and to whether it gets finished.

Studios, publishers and investors need development budgets to price a project, compare it with others or judge whether a game paid back. Yet budgets are rarely disclosed, and the industry relies on sources with known problems:

  • Disclosed budgets. Few games publish a budget, and those that do are not a random sample: large productions, exceptional cases and solo developers' games are overrepresented. Definitions also differ (with or without marketing, in nominal or constant currency), so published figures are not comparable as they stand.
  • Rules of thumb. Estimates such as team size × monthly cost × development time depend on inputs rarely known from outside (the team over time, the length of development, outsourcing), and assume one cost per person across countries and roles.
  • Survivorship. Only released games are observed. Cancelled projects, and the conditions that led to them, are missing from any dataset built on releases, which biases any assessment of risk.

As a result, the industry lacks reliable budget estimates.

We collect real budgets, make them comparable, and train a model that estimates the development budget of every game on Steam. This addresses the first two problems: budget levels that are disclosed more often count less in training, so they do not skew the estimates, and the model uses observable outputs (credits, install size) instead of internal production data (development time, salaries, monthly cost). Survivorship is the subject of our completion model, in development.

Salaries alone do not say how many people worked on a game, or for how long. Real budgets measure the total cost directly, so the model learns from them how a game's credits, size and other observable characteristics translate into cost. We are now developing a version of the model that uses country-specific salaries (from our database of more than 5,000 salaries from 2019 to 2026, across more than 15 countries) to account for labor costs that differ by country.

Methods

Budgets

Much of what determines the cost of a game can be seen from the outside: where the studio is based, the size and art direction of the game, how long it takes to play, and above all its credits, which describe the size, experience and composition of the team, the range of roles and the presence of key positions. With a large enough set of real budgets, the link between these characteristics and development cost can be learned. Progress depends on more real budgets, gathered with industry partners, and on better models of how production conditions translate into cost.

The estimates improve with every real budget we collect. Share yours, confidentially, and everyone's estimates get better. Share your game's budget →

Completion

Many projects stop before release, at different stages of development. Survival analysis provides a framework to model the probability that a project is abandoned, and at which stage, as a function of its production conditions. It requires data that releases cannot provide: cancelled projects, and projects still in development whose outcome is not yet known.

Applications

  • Estimating the development budget of any released game, and the budget range of a project before release.
  • Benchmarking a budget against productions with a similar team, scope and features.
  • Computing revenue-to-budget ratios for a game, a segment or the whole market.
  • Measuring what drives development cost: team size and composition, experience, publisher, scope and design choices.
  • Estimating the probability that a project never ships (in development).

Used in

Press coverage

Game World Observer, AUTOMATON, GameDiscoverCo, GameDev Reports, Multiplayer.it, MP1st, GameLook, 4Gamers, Youxi Tuoluo, App2Top, GoHa.Ru, PlayGround.ru, GameVicio, Game as a Podcast.

We license our datasets and model outputs for non-commercial academic research. Reach out →