What is the Boxleiter method?

It is a way to estimate the copies a Steam game sold from its review count. The formula is: copies sold = number of reviews × a multiplier. The multiplier is the number of copies sold for each review. People also call it the Boxleiter number, the review multiple or the sales per review ratio.

The method works because Valve does not publish sales, but the review count of each game is public. Simon Carless of GameDiscoverCo updated the number in August 2020 and called his result the NB number, for New Boxleiter number. The two names mean the same method.

A high multiplier means few buyers wrote a review. A game with 100 reviews and 5,000 copies sold has a multiplier of 50. The multiplier is different for each game, so the honest result is a range, not one number.

Sources: GameDiscoverCo (3 August 2020); GameDiscoverCo (28 August 2023).

Who created the Boxleiter method, and when?

The idea is credited to the developer Mike Boxleiter, but we could not find his original post. The earliest public text we found is by Jake Birkett of Grey Alien Games, dated 12 March 2015. Birkett wrote that he discussed the idea with Ryan Clark: "He tells me the idea came from Mike Boxleiter."

So we cannot give a date or a first multiplier from Boxleiter himself. In 2018, Birkett wrote: "I think someone mentioned the Boxleiter number was 50x". That is a second-hand report, not a source.

In the 2015 post, Birkett took the 50 most popular Steam games of 2014 and divided owners by reviews. Paid games had an average of 106 owners per review, and 81 without two outliers. His conclusion was a guide of 30 to 100, with 70 as a middle value.

On 4 May 2018, Birkett published a second analysis with sales data from developers: more than 50 paid indie games. The average was 82 copies per review and the median was 77, with a range of 30 to 150. He also saw that newer games had a lower ratio. The 2017 launches had a median of 65, from only 8 games.

On 3 August 2020, Simon Carless published a GameDiscoverCo survey of 237 games and named the result the NB number. On 2 August 2021, Karl Kontus of Video Game Insights added a study of more than 10,000 games. Later updates came from GameDiscoverCo in November 2022 and from Gamalytic in 2023. The next section gives each number.

The sources disagree on one date. Birkett in 2018 and Video Game Insights in 2021 both call the first post an analysis "from 2014". The post itself is dated 12 March 2015. It uses games from 2014.

Sources: Jake Birkett, Grey Alien Games (12 March 2015); Jake Birkett, Game Developer (4 May 2018); GameDiscoverCo (3 August 2020); Video Game Insights, Game Developer (2 August 2021).

How has the multiplier changed by source and year?

It fell by about half. Surveys before 2019 found a median near 60 to 80 copies per review. Surveys of games released from 2020 found a median near 30 to 40. The main cause is a change in Steam: at the end of October 2019, Steam started to ask players to review games they had played.

The table gives each published number with its sample. The list below the table links each row to its source.

The sources disagree for the same release years. For games launched in 2020, GameDiscoverCo measured a median of 38, and Video Game Insights gave about 30. For games launched in 2019, GameDiscoverCo measured a median of 51. The benchmarks page of How To Market A Game lists 36 for 2019 and 31 for 2020, and names both studies as its research. The samples and the dates of measurement are different, so use the range.

We found no newer public survey of developer sales data than the GameDiscoverCo survey of November 2022. In April 2026, Jake Birkett wrote about the "Boxleiter ratio" again, but he gave no new number.

Copies sold per Steam review, by source and date. Each number is from the source in the row, read 19 November 2026.
Source and dateSampleCopies per reviewRange
Mike Boxleiter, date not foundWe could not find the original postNot found. Birkett heard "50x" second hand.Not found
Jake Birkett, 12 March 2015The 50 most popular Steam games of 2014; owners, not salesAbout 70 as a middle value30 to 100
Jake Birkett, 4 May 2018More than 50 paid indie games, sales data from developersAverage 82, median 77. 2017 launches: median 65 (8 games)30 to 150
GameDiscoverCo, 3 August 2020 (NB number), all years237 games, sales data from developersAverage 63, median 5825 to 100 for the middle 80%
GameDiscoverCo, 3 August 2020, launched before 2017Part of the 237 gamesAverage 81, median 74Not given
GameDiscoverCo, 3 August 2020, launched in 2019Part of the 237 gamesAverage 57, median 51Not given
GameDiscoverCo, 3 August 2020, launched in 2020Part of the 237 gamesAverage 41, median 3820 to 60
Video Game Insights, 2 August 2021Leaked 2018 data for 11,445 games, plus about 100 games of its ownAbout 30 for 2020 releases; more than 70 for 2014 and earlier20 to 55 from 2020; 40 to 100 before 2014
GameDiscoverCo, 14 November 2022About 200 games released from 2020 to 2022Median 29 for 2022 releasesMedian 20 under 1,000 copies; 59 at 50,000 to 99,999 copies
Gamalytic, 13 May 2023About 120 games with public sales numbersMedian about 35Not given
GameDiscoverCo with Gamalytic data, 28 August 2023Gamalytic estimates, not a developer survey36 under 100 reviews; 52.8 at 1,000 to 10,000 reviewsNot given

Sources: Jake Birkett, Grey Alien Games (12 March 2015); Jake Birkett, Game Developer (4 May 2018); GameDiscoverCo (3 August 2020); Video Game Insights, Game Developer (2 August 2021); GameDiscoverCo (14 November 2022); Gamalytic (13 May 2023); GameDiscoverCo (28 August 2023); How To Market A Game benchmarks; Jake Birkett, Patreon (22 April 2026).

Which multiplier should I use today?

Use 20 to 60 copies per review for a paid game released from 2020, with 30 as the middle. Use a higher range for an older game: Video Game Insights gave 40 to 100 for games released before 2014, and GameDiscoverCo measured a median of 74 for games launched before 2017.

Our tools use 20, 30 and 60 as the low, middle and high multipliers. The low and the high are the range that GameDiscoverCo gave for games launched in 2020: 20 to 60. The middle is the best guess of Video Game Insights for games released in 2020, about 30. The GameDiscoverCo median of 29 for 2022 releases agrees with it.

The middle is not the median of every survey. GameDiscoverCo measured a median of 38 for 2020 launches, so 30 can be low for a game with good sales. GameDiscoverCo also wrote in 2022 that a popular new game is closer to 60 over time. If the game sold well, read toward the high end. If it has few reviews, read toward the low end.

These multipliers are fixed in our code. The Steam sales estimator, the Steam game stats lookup and the Steam revenue calculator all use the same three numbers, so their results agree.

Sources: GameDiscoverCo (3 August 2020); Video Game Insights, Game Developer (2 August 2021); GameDiscoverCo (14 November 2022); Manalens tool pages.

Why does the ratio differ from game to game?

Because the share of buyers who write a review changes with the game, its price, its age and its players. In the GameDiscoverCo survey of 2020, the middle 80% of games sat between 25 and 100 copies per review. That is a spread of four times.

The table lists the known causes. Where two sources disagree, the table says so.

Steam itself changed the ratio. Steamworks documentation says that after a player has played a game for a period of time, Steam asks the player to leave a review inside the Steam client. GameDiscoverCo dates this prompt to the end of October 2019. Gamalytic wrote that it "cut the sales-review ratio in half". Valve also tells developers not to ask for reviews from inside the game.

Free weekends are a gap in the data. Steamworks says a purchase is not required to write a review: a customer can review any product that they have launched through Steam. We found no Valve statement on how reviews from a free weekend count, and no study of their effect. Be careful with a game that had many free weekends.

What moves the copies per review ratio, with the source of each finding.
FactorEffect on copies per reviewSource
Release yearLower for newer games. Median 74 before 2017, 38 for 2020, 29 for 2022.GameDiscoverCo, 2020 and 2022
Steam review promptLower after the end of October 2019. Old games also got late reviews: one game had a ratio of 67 for 2017 to 2019 and 14 for 2020.GameDiscoverCo, 2020
PriceHigher for cheap games and lower for expensive games. Free games have almost twice as many players per review.Video Game Insights, 2021; Gamalytic, July 2023
Discounts and bundlesHigher. A buyer at a deep discount writes a review less often, and the same is true for bundles.Gamalytic, July 2023; GameDiscoverCo, August 2023
GenreSources disagree. Birkett saw higher ratios for roguelikes, sims and strategy, and lower for story games. Video Game Insights saw higher for MMOs and lower for sports and racing. Gamalytic found genre less important than expected.Birkett, 2018; Video Game Insights, 2021; Gamalytic, July 2023
Game sizeMedian 20 under 1,000 copies, 36 at 1,000 to 9,999, 49 at 10,000 to 49,999, 59 at 50,000 to 99,999, and 48 above 100,000.GameDiscoverCo, 2022
Review scoreAbout 30 above 90% positive, close to 60 near 70% positive, about 30 again under 60%. Birkett saw no clear link in 2018.Gamalytic, July 2023; Birkett, 2018
PlaytimeLower for games that people play longer.Birkett, 2015; GameDiscoverCo, August 2023
Early AccessHigher. GameDiscoverCo adds 25% for a game in Early Access.GameDiscoverCo, 2022
Regional mixPlayers in Asia review a little more often, but Gamalytic says the link is not statistically significant.Gamalytic, July 2023

Sources: Jake Birkett, Game Developer (4 May 2018); GameDiscoverCo (3 August 2020); Video Game Insights, Game Developer (2 August 2021); GameDiscoverCo (14 November 2022); Gamalytic (23 July 2023); GameDiscoverCo (28 August 2023); Steamworks documentation, User Reviews.

Which reviews does the method count?

Count the reviews from people who bought the game on Steam, in all languages. That is the count the surveys used, and it is the count our tools read.

Reviews from Steam keys are not in the score. Steamworks documentation says: "Only purchases (not key activation) made by the reviewing account are included in the review score." Birkett built his 2018 ratio on Steam sales only, with the "All Reviews" count at the top of the store page, which leaves out reviews from keys.

So the method estimates copies bought on Steam. It does not count keys sold on other stores, keys in bundles, or keys given to press and backers. A game that sold many keys has more owners than the estimate shows.

Check the language too. On 19 August 2025, Game Developer reported that Steam now shows a score from reviews in the language of the user, for games with more than 2,000 public reviews and at least 200 in one language. The count at the top of a store page can then be lower than the total. Our lookup asks Steam for the total in all languages.

Sources: Steamworks documentation, User Reviews; Jake Birkett, Game Developer (4 May 2018); Game Developer (19 August 2025); Manalens Steam sales estimator.

How do I estimate copies, gross revenue and net revenue?

Multiply the reviews by 20, 30 and 60 for copies. Multiply the copies by the price for gross revenue at list price. Then take off discounts, refunds, tax and the Steam fee for net revenue. Every number below is an estimate, and the example game is not a real game.

The example: a paid game released in 2023 with 1,000 reviews and a price of $14.99. Copies: 1,000 × 20 = 20,000 (low), 1,000 × 30 = 30,000 (middle) and 1,000 × 60 = 60,000 (high).

Gross revenue at list price: 30,000 × $14.99 = $449,700 in the middle. The low estimate is $299,800 and the high estimate is $899,400. No game earns this amount, because many copies sell at a discount or at a lower regional price.

Now the adjustments for the middle estimate. An average discount of 20% leaves $359,760. Refunds of 8% take $28,781 and leave $330,979. VAT and sales tax of 10% take $33,098 and leave $297,881 of net sales. Valve keeps 30% of that, $89,364. The estimate of the developer share is $208,517.

That is about 46% of the list price gross. The same math gives $139,011 for the low estimate and $417,034 for the high estimate. GameDiscoverCo gave the same warning in 2020: the money a developer receives is "30%-50% of the 'optimistic' gross number", and less if the discounts are deep.

The 20%, 8% and 10% are the default inputs of our Steam revenue calculator. They are planning values, not measured values for any game, and you can change each one. The calculator has one discount input, so put sale discounts and lower regional prices together in it. The 30% is the Valve share until a game earns $10 million; Valve set 25% above $10 million and 20% above $50 million in 2018. A publisher share comes off after that.

  1. Read the review count in all languages, for example with the Steam game stats lookup.
  2. Multiply it by 20, 30 and 60 to get a low, middle and high estimate of copies.
  3. Multiply each estimate by the list price to get gross revenue at list price.
  4. Take off the average discount, which includes sales and regional prices.
  5. Take off refunds, then VAT and sales tax, then the Steam fee.
  6. Write down the range and label it as an estimate.
Worked example, all estimates: 1,000 reviews at $14.99, with a 20% average discount, 8% refunds, 10% tax and a 30% Steam fee. Our arithmetic.
StepLow (20x)Middle (30x)High (60x)
Estimate of copies sold20,00030,00060,000
Gross at list price ($14.99)$299,800$449,700$899,400
After a 20% average discount$239,840$359,760$719,520
After 8% refunds$220,653$330,979$661,958
After 10% VAT and sales tax (net sales)$198,588$297,881$595,763
After the 30% Steam fee (developer share)$139,011$208,517$417,034

Sources: Manalens Steam revenue calculator and Steam fee calculator (our arithmetic); GameDiscoverCo (3 August 2020); Valve (30 November 2018), text quoted by Wccftech (1 December 2018).

How wrong can the Boxleiter method be?

Very wrong for one game. In May 2023, Gamalytic tested one multiplier of about 35 against about 120 games with public sales numbers. Only 42.74% of the estimates were within 30% of the real number, and 70.94% were within 50%. So about 3 games in 10 were wrong by more than half.

A multiplier that changes with release date, price and review score did a little better: 50.43% within 30% and 80.34% within 50%. Gamalytic ran this test itself and chose the sample, and it left out games that sold fewer than 1,000 copies.

Other authors give the same size of error. Birkett wrote in 2018 that with his median of 77 "you might have overestimated or underestimated by about 2x". Video Game Insights said its ranges were right for more than 80% of games, and that its total for more than 10,000 games was 7% above the real total. The method is good for a group of games and weak for one game.

This is why our tools show three numbers and not one. The high estimate is three times the low estimate, and a real game can still be outside the range.

Accuracy of the review method in the Gamalytic test of May 2023, on about 120 games with public sales numbers.
Method in the Gamalytic testGames within 30%Games within 50%
One review multiplier (about 35)42.74%70.94%
Review multipliers by release date, price and review score50.43%80.34%

Sources: Gamalytic (13 May 2023); Jake Birkett, Game Developer (4 May 2018); Video Game Insights, Game Developer (2 August 2021).

When should I not use the Boxleiter method?

Do not use it for a game in its first week, a free game, a game from before 2014, or a game with fewer than about 50 reviews. In each case the ratio is far from the published range.

Use better signals when you have them. For your own game, use the sales report in Steamworks: it is the real number. For another game, a public statement from the developer or the publisher is better than any multiplier. Use the method only to fill the gaps, and compare several games, not one.

  • Very new games: GameDiscoverCo wrote in 2022 that reviews need about 5 to 7 days to catch up after release. One game sold 100,000 copies in its first week with 1,148 reviews: a ratio of 87. Birkett saw the opposite in 2018, an average of 38 for games released in the last 3 days. The sources disagree, so wait.
  • Free to play games: Birkett found 201 owners per review for free games in 2015, against 106 for paid games. GameDiscoverCo removed free games from its 2020 survey as outliers. Our tools give no sales estimate for a free game.
  • Very old games: Steam reviews started in November 2013, and Birkett left out the games that launched before that date. Older games also got many late reviews after the 2019 prompt, so their ratio changes with time.
  • Games with fewer than about 50 reviews: GameDiscoverCo found a median of 20 for games under 1,000 copies, and said friends of the developer write many of those reviews. At 20 copies per review, 1,000 copies is 50 reviews (our arithmetic). Gamalytic left such games out of its test because the error is too high.
  • Games sold mostly as keys or in bundles: reviews from keys are not in the score, so the estimate is too low for owners.

Sources: GameDiscoverCo (14 November 2022); Jake Birkett, Grey Alien Games (12 March 2015); Jake Birkett, Game Developer (4 May 2018); GameDiscoverCo (3 August 2020); Gamalytic (13 May 2023); Gamalytic (23 July 2023).

What other methods estimate Steam sales?

There are four other public methods. Each one is also an estimate, and a mix of methods is better than any single one.

Followers to wishlists, for a game that is not out. An unreleased game has no reviews, but its follower count is public. In June 2023, GameDiscoverCo surveyed more than 125 games and found that most had 7 to 20 wishlists per follower, with a median of 12. This gives wishlists, not sales. Our followers to wishlists estimator uses these three numbers.

Concurrent players (CCU). Steam publishes the number of players in a game. Gamalytic adds up the concurrent players for every hour and divides by an estimate of the average playtime. In its May 2023 test, this put 64.44% of games within 30% of the real number. Our guide shows how to estimate sales from CCU, and our CCU tool turns a peak into daily and monthly players.

Owner estimates from public profiles, as SteamSpy did. A tool reads a sample of public Steam profiles and counts who owns the game. GameDiscoverCo notes two limits: Steam profiles are private by default from April 2018, and owners include keys from bundles. In the Gamalytic test, this method put 77.91% of games within 30%, but Gamalytic says it does not work for games under 20,000 players.

Third-party tools. Our free tools come first in this list because this is our site: the Steam sales estimator, the Steam game stats lookup and the Steam revenue calculator. They use fixed public ratios and give a range, with no accuracy claim. Gamalytic and Video Game Insights (now part of Sensor Tower) mix several signals. For one game with many players, a mixed estimate is likely to be closer than our range: in its own test, Gamalytic put 76.92% of games within 30% with all methods together.

Sources: GameDiscoverCo (21 June 2023); Gamalytic (13 May 2023); GameDiscoverCo (14 November 2022).

How do I see my real Steam sales and not an estimate?

Manalens is the analytics and growth engine built for games. Connect a Steamworks Financial API key (the automatic Steam sync is in early access), and a daily sync records wishlist adds, deletes, purchases and gifts, and units sold and returned, with gross and net sales in USD per day and country. Use the Boxleiter method for other games, and use this data for your own game.

Sources: Manalens product documentation.

Frequently asked questions

What is the Boxleiter number?

It is the number of copies a Steam game sold for each review. Multiply the review count by it to estimate copies sold. For games released from 2020, GameDiscoverCo found a range of 20 to 60. The name credits Mike Boxleiter, but we could not find his original post or his first number.

How many sales is one Steam review worth?

About 20 to 60 copies for a paid game released from 2020, with about 30 as a middle value. Older games sit higher: Jake Birkett measured a median of 77 in 2018, and GameDiscoverCo measured a median of 74 for games launched before 2017. Each game has its own ratio.

What is the NB number?

NB means New Boxleiter. Simon Carless of GameDiscoverCo gave the name to his survey of 237 Steam games, published on 3 August 2020. It found a median of 58 copies per review for all games and 38 for games launched in 2020. It is the same method with newer data.

Why do the Manalens tools use 20, 30 and 60?

The low and high numbers are the GameDiscoverCo range for games launched in 2020: 20 to 60 copies per review. The middle number, 30, is the best guess of Video Game Insights for games released in 2020, and GameDiscoverCo measured a median of 29 for 2022 releases. Older games need higher multipliers.

Do reviews from Steam keys count?

Not in the review score. Steamworks documentation says only purchases made by the reviewing account are included in the score, not key activations. So the method estimates copies bought on Steam. A game that sold many keys on other stores or in bundles has more owners than the estimate shows.

Can I estimate revenue from the review count?

Only as a rough range. Copies × list price is too high, because of discounts, regional prices, refunds, tax and the Steam fee. In our worked example, the developer share is about 46% of the list price gross. GameDiscoverCo puts the real amount at 30% to 50% of the optimistic gross number.

Sources and implementation