We Predicted 9,000+ Games.
Here's How Close We Got.
Steam games get anywhere from zero reviews to 500,000+. Predicting where a game will land — from its store page alone, before launch — is one of the hardest prediction problems in gaming.
No other tool even attempts it. Here's how we do.
Why This Problem Is So Hard
To understand why our accuracy numbers are impressive, you first need to understand what we're up against.
1 to 500,000+
The Range Is Massive
Imagine guessing how many jellybeans are in a jar — except the jar could hold 1 jellybean or half a million. That's the range of review counts on Steam. Getting within 2x of the right answer across that range is genuinely difficult.
15,000+
Games Launch Every Year
Every game is different. Different genre, different art style, different audience, different price. A horror roguelike at $15 has completely different success patterns than a cozy farming sim at $25.
∞
Luck Is Real
One viral TikTok can 10x a game overnight. One big streamer can turn an unknown title into a bestseller. No model can predict lightning strikes — but it can predict everything else.
For Most Indie Games, We Nail the Ballpark 9 Out of 10 Times
In the 100–10K review range — where most indie games with real traction land — our v3.1 model's prediction falls within 5x of the actual result over 90% of the time. Within 3x, it's nearly 8 out of 10.
What this means: If we predict your game will get 2,000 reviews, there's a 61.5% chance the actual count lands between 1,000 and 4,000. An 80% chance it's between 667 and 6,000. And a 90% chance it's between 400 and 10,000.
In real money: 2,000 reviews roughly translates to 60,000–120,000 copies sold. Within 2x means we're telling you whether you're looking at $300K or $1.2M in revenue — not exact, but enough to decide whether to invest in marketing, scope your next project, or quit your day job.
Pick Your Game's Range. See How We Do.
Different game sizes are different prediction challenges. Pick the range your game is likely to fall in.
1K – 10K Reviews
3,422 games testedThe sweet spot. Games like Dome Keeper, Webfishing, and Unpacking. Most successful indie games land here.
Exact tier accuracy: 69.4%
Exact tier accuracy: 63%
Even When We Miss, We Don't Miss By Much
We classify games into review tiers: 0–10, 10–100, 100–1K, 1K–10K, and 10K+. In the 100–10K range, v3.1 predicts the exact correct tier up to 74.5% of the time. But even when it's off, it's almost always off by just one tier.
74.5%
Exact tier match
100–1K range
69.4%
Exact tier match
1K–10K range
99.1%
Within one tier
100–10K range
That 99.1% means: if v3.1 says your game is in the "1K–10K" tier, the actual answer is virtually never "50 reviews" or "100K reviews." It's either right on target, or one tier off.
Why Some Ranges Are Harder to Predict
0 – 100 reviews
One YouTuber can 5x a tiny game overnight. The difference between 10 and 80 reviews often comes down to a single lucky break. The model identifies these as low-traction — but exact counts are noisy.
100 – 10K reviews
Our strongest range. Games here succeeded based on store page quality, genre positioning, and pricing — exactly what the model measures. There's enough signal to separate a 300-review game from a 3,000-review one.
10K+ reviews
Blockbuster success depends on publisher budgets, platform deals, and viral moments. The model knows a game could break out, but whether it does is partly luck. v3.1 is notably better here because follower data gives early signal.
Good Enough to Act On
You don't need a perfect prediction. You need one good enough to make better decisions than guessing. SteamOracle tells you whether your game looks like a 500-review release or a 5,000-review release — and what changes to your store page could move the needle.
Free tier available. No account needed.
All accuracy numbers are from cross-validated predictions — every game was predicted by a model that never saw it during training. Tested on thousands of real Steam games. Methodology details available to Foresight subscribers.