Executive Summary

We review Steam sale data for self-published games released between May 2023-May 2024 to determine the difficulty of maintaining a stable income by releasing games on the platform. We find a typical power law distribution in which 10% of games “succeed” and 90% of games “fail” based on our income velocity metric. While this leaves Steam publishing to be a rather hazardous and risky occupation, we also find no reason to believe this number is more or less damning than average entrepreneurial success rates across all industries.

Introduction

Media and entertainment is consistently talked to be the hardest entrepreneurial pathway to reach income velocity. The reasoning for this tends to be that media, unlike traditional forms of business, is tied heavily into taste and trends — two forms of buyer influence that tends to be incredibly fickle.

Video games, however, have been historically seen to be a rather strange midpoint between the “media and entertainment” industry and the “tech B2C” industry. Many have found significant fortune, or at least stable income, in the gaming world as opposed to film, music, or writing industries. These individuals tend to be self-published, forgoing publisher deals in favor of simpler crowdfunding alternatives or even bootstrapped ventures.

One of the most popular platforms to self-publish on is Steam, the most popular platform across Windows, Mac, and Linux. We decided to use Steam as the centerpiece for our data in this study.

Methods

We started with raw Steam Release data from SteamSpy from May 2023 to May 2024, considering a full year’s worth of releases. From this, we obtained around 14,185 games. Doing a simple independence check — that is, making sure that games were self-published by verifying that the developer name was the name as the publisher. This took down the total list to 10,408 games. It is worth pointing out this number is still the majority of releases for Steam for the year, which lines up with self-published books making up the majority of Amazon Kindle, or self-published records making the majority on Spotify.

From there, removing games that were free or did not have price data, and removing other miscellaneous data errors (for example, licensed games that had a leg up in revenue, or AAA games where the publisher and developer were the same) lead us to a final number of 10,179 games.

To determine the sustainability of income, we used a cutoff of $50,000 in revenue. With the average yearly cost of living in the United States being $30,000, this simply adds another $20,000 in padding. It is also roughly the average salary across the United States.

Revenue numbers were taken by multiplying the base price by the number of copies sold. Since SteamSpy uses ranges for copies, we used the midpoint for each range (for example, 50,000 .. 100,000 copies would be simply 75,000 in the data). The one exception for this would be the lowest range, 0 .. 20,000, which we opted for the more conservative assumption of 0, knowing that previous data on game sales had shown that the distribution follows a power law where the majority of games make zero revenue.

Results

[Image omitted]

The results followed the well-known power law distribution of media releases. Of the 10,000+ games surveyed, two games — Palworld and Last Epoch — made up almost 60% of total game sales. 9,196 of these games — or roughly 90% — made either no money or not enough money to sustain a single individual for a full year (that is, sub $50,000).

One thing worth mentioning is, despite media entrepreneurship being considered as more “difficult” than other more standard forms of business entrepreneurship, the data here does not suggest any difference between the two. Both follow stable power laws in which only a few market entrants make revenue, however of those who do pass an revenue threshold, almost every single one makes a sustainable income (only a 43 games in our data reached an uncomfortable midpoint of making more than 20,000 sales but also not taking in enough revenue to sustain a year’s worth of costs).

We also find that the median price for successful games tends to be $7.99, while the median price for unsuccessful games tends to be $4.99. This seems to follow the widely held belief that giving a game a “premium” look and refusing to play in price wars tends to lead to better sales. This logic also exists in other industries such as web apps or traditional businesses.

Considerations

Steam provides no public information on sale data, meaning that SteamSpy — and all other sources of data of this kind — are usually estimated and not directly sourced. This comes with several potential confounders for our results.

The first is that the revenue data does not consider discounting prices. Games on Steam — particularly self-published games — are often on sale for amounts going as far as 75%-90%, but our data only uses the snapshot of the original price point. So if a game sales at retail for 9.99 but received the majority of its sales while it was 2.99, it may be falsely assumed to pass our revenue threshold.

This is also assuming that each entrant is only releasing one game. Often indie creators release many smaller games across a year in order to diversify their portfolio. Because of this, they might contribute 3-4 games to the “failure” group but 1 game to the “success” group, and therefore being an overall success. This could be tested in a future study by grouping the data by developer name.

Lastly, Steam is only one platform, and self-publishers often diversify themselves across many platforms, such as other PC platforms like Itch.io and Epic Games Store or consoles such as the Nintendo Switch or Playstation 5. Theoretically the sales across all these systems could take a developer into the “success” range, but it is unlikely because 1) all other platforms have been shown to follow the same power laws, and 2) other platforms often have more strict, not less strict, measures of reaching their platform — meaning that less total developers are likely to host their games there compared to Steam.

Further Research

Since this project was limited to a week, there were a few key pieces of research we were interested in but unable to complete by the deadline. Our main interest was in bringing in Steam ratings data to the SteamSpy dataset. There is a rather popular hypothesis that games that score better user ratings on average rake in more revenue. Unfortunately, SteamSpy does not include Steam review data, and so this process would require us to import data from a different database, and cross reference the results. It is something that is possible, but we were unable to accomplish in the given timeframe.

Another question is whether or not repeat market entrants have a higher success rate. In Considerations we hinted at the idea that some developers release multiple games in a year, and power law statistics seem to show a user only needs one “hit” to reach the revenue threshold. Therefore, it could be hypothesized that those who release more games a year have a slight advantage over those who only release one or two. As mentioned, this could be tested by grouping the data by developer and redoing success rates based on the total sum sales of the developer.