How these numbers are calculated

Revenue and ownership figures on this site are estimates based on public data. This page explains the calculations, their assumptions and their limitations.

Revenue

Steam does not show sales. It shows review counts, and a fairly stable share of buyers leave a review. Multiplying reviews by an estimated sales-per-review factor gives an estimate of copies sold — the Boxleiter method, named after Mike Boxleiter, later refined by Simon Carless and others.

The catch is that the share is not one number. It varies with genre, price, release year and how loudly a game asks for reviews; published estimates land anywhere between 20 and 80 sales per review. A single figure would therefore be false precision, so this site shows a range:

• Low end — 20 sales per review
• Expected — 45 sales per review
• High end — 80 sales per review

Where SteamSpy also provides an owner estimate — an entirely separate method, based on sampling public profiles — the range is narrowed to where both agree. If they contradict each other, the range is widened to cover both rather than quietly picking a favourite; the game page says so when that happens.

Which price the copies are multiplied by

Not the launch price. Most copies of most Steam games sell on sale, and plenty of games settle at a permanently lower price years after release — multiplying every copy by the launch price is the single largest known bias in a Boxleiter estimate, and it always points the same way: too high.

So the copies are spread over time and priced individually. Two public series make that possible, and between them they usually reach back to a game's release:

• Valve's review histogram gives the review count month by month over a game's whole lifetime. The growth between two months is a proxy for the copies sold in that month — the same Boxleiter logic, applied per period instead of once over everything.
• IsThereAnyDeal's price log gives every price change to the day, including every sale.

Each month's copies are therefore multiplied by the price actually in effect that month, and the results are added up. The game page states the resulting average price and how far below list price it lands. Where the two series do not cover enough of a game's lifetime to weight it honestly, the estimate falls back to the launch price and says so.

What that price still gets wrong

• Reviews lag sales. People who buy in a sale write their review days or weeks later, so some sale copies get charged at the full price that followed. This is not corrected for — the correction would be a second invented number next to the sales-per-review factor, and two stacked assumptions are harder to check than one. The bias that remains points the same way as before: high rather than low.
• Within a month, copies are spread evenly. Reviews are monthly, prices are daily, so a one-week sale enters its month time-weighted. In reality sales cluster in the discount week, which means the discount effect is understated rather than overstated.
• Before the price log starts — for older games it begins later than the reviews do — those periods still use the launch price.
• No period is ever valued above the launch price. The price log also carries edition and bundle prices for the same game, and the occasional outright bad value, and a single such point would otherwise be multiplied by a whole period's worth of copies. Prices above the launch price are therefore capped at it. Where a game became more expensive, this can understate revenue. This bias acts in the opposite direction to the timing assumptions described above.

What the revenue figure does not include

• Steam's cut (30 %, less at high volume) and VAT are not deducted — the figure is gross.
• Refunds, which Steam grants generously within two hours of playtime.
• Regional pricing. Prices here are US prices. A game costing $40 in the US may cost the equivalent of $12 elsewhere.
• Keys sold off Steam, bundles, and giveaways — these produce owners and reviews with little or no revenue.
• In-game purchases and DLC, which for some games dwarf the base price entirely.

Taken together: treat the estimate as an order of magnitude, not an accounting figure. It is most reliable for paid single-purchase games with more than a few hundred reviews, and least reliable for free-to-play titles, heavily bundled games, and anything with fewer than about 50 reviews.

Owners and concurrent players

Both come from SteamSpy, which samples public Steam profiles. Owners are reported as a range (“1M – 2M”) and shown that way here — averaging the two ends would invent a precision the method does not have. Owners include free keys and gifts, so they are an upper bound on sales, not a synonym.

The player count is labelled concurrent players, not a peak. SteamSpy documents the figure as yesterday's peak, but held against the monthly rows on SteamCharts it lands on the monthly average, not on any peak: in August 2026 SteamSpy reported 314,682 for PUBG, where SteamCharts had that game averaging 327,150 players through July and peaking at 792,142. It is a typical simultaneous-player figure, and roughly half of what the game reaches at its busiest hour.

SteamSpy's playtime figures are deliberately not used: they have returned zero for every game since Valve removed playtime from public profiles in 2018.

Tag statistics

Tags come from SteamSpy's community-vote ordering, so a game's first tag is the one players most associate with it and its eighteenth is a stray vote. That order matters and is preserved.

For each tag the site reports both the total and the median. The difference is the whole point:

Action has an enormous total revenue because it contains some of the biggest games ever sold. The median game tagged Action earns a few thousand dollars. The median helps describe a typical game in the group; the total describes the group as a whole.

Success rate is the share of games in a tag that reach at least 50 reviews — roughly a thousand-plus sales. It is not a measure of quality but of visibility: how often a game in this niche gets found at all.

The niche score

For every pair of tags, the site compares demand (median reviews per game in that combination) against supply (how many games carry it). Both are converted to percentile ranks across all pairs, and the score is the difference:

niche score = demand percentile − supply percentile

A combination in the 85th percentile of demand but only the 30th of supply scores +55. Positive scores indicate relatively high median review activity compared with the number of games; negative scores indicate the reverse. This is a relative ranking, not a direct measure of unmet demand.

Percentiles rather than raw ratios, because the distribution is extremely skewed — most tag pairs have a median review count near zero, and a ratio against that baseline sends many combinations toward extreme values. Percentile ranks compare their relative positions within the dataset.

Only tags within a game's top 6 by community vote count toward a pair, and pairs with fewer than 15 games are excluded to reduce the influence of small samples.

Use the score as a starting point for research. It does not explain why relatively few games serve a niche or predict how a new game will perform.

Next Fest numbers

Valve ranks the demos of a Next Fest by how much they are played, but publishes no figure behind the ranking — and once the fest closes, the ranking is gone. The fest page therefore mixes two kinds of data, and it is worth knowing which is which.

• Chart positions are Valve's. They are read from the store's own fest browser several times a day and stored with a timestamp. That is what makes the “biggest climbers” list possible: the store only ever shows the order of right now.
• Player counts are ours. Steam reports the number of people currently in a given app, and a demo is an app — so the figure is a direct reading, not an estimate. The peak is the highest reading over the whole fest; the current value depends on the time of day you look.
• There are no wishlist numbers. Nobody publishes them. Follower growth during the fest is the closest public substitute, and it is not the same thing — you can follow a game without wishlisting it, and the other way round.

Two gaps worth naming. Only demos whose game is in this catalogue are listed, so Valve's position numbers can skip. And a demo is measured in turn, not continuously: the leading titles are read every hour, the rest of the field over the course of a day.

Where the data comes from

• Steam Web API — the list of every application ID.
• Steam Storefront API — name, price, genres, media, release date.
• Steam review API — review counts and score.
• SteamSpy — community tags, owner ranges, concurrent players.
• Steam store fest pages — which games take part in a Next Fest, and Valve's own order of them. Same endpoint the store's fest browser uses.
• Steam Community — follower counts. Valve does not publish them, but following a game means joining a hidden group, and that group's member list is public. The count shown for an unreleased game is that member count.

Prices and review counts are refreshed on a rolling schedule — heavily reviewed games daily, mid-sized ones weekly, the long tail monthly — and every change is recorded, which is where the history charts come from. Nothing here is a live figure; the game page shows when it was last checked.

Two moments get a faster schedule than their size would earn them, because both are short and neither can be measured after the fact. Anticipated games still to come are checked several times a day, so the run-up to a launch is a curve rather than a point. And for the first week after release — the week in which reviews, price and player count move faster than they ever will again — every game is checked at least three times, however small it is.

Known gaps

• Wishlist counts are not available to anyone outside Valve. No tool has them, including this one.
• Follower history starts the day a game was first measured here — past counts are published by nobody, Valve included, so no chart can reach behind that date.
• Only games are tracked — DLC, demos and soundtracks are skipped during collection.
• How far back a chart reaches depends on the series. Reviews, prices and player counts are filled in from the public sources named above and usually cover a game from its release. Every other figure starts on the day this site first recorded the game.

Found a number that looks wrong? It probably is. The method above is public precisely so that its failures are checkable.