What factors predict the success of a Steam game?

I've seen quite a few discussions, comments and questions on /gamedev about what determines a game's success. How much does quality matter? Is establishing market awareness before launch the only thing that matters? Does a demo help or hurt? If your game has a poor launch, how likely is it to recover? Is it possible to roughly predict the sales of a game before launch?
In preparation for my game's launch, I spent a lot of time monitoring upcoming releases trying to find the answer to these questions. I compiled a spreadsheet, noted followers, whether it was Early Access or not, and saw how many reviews it received in the first week, month and quarter.
I'm sharing this data now in the hopes that it helps other developers understand and predict their games' sales.
First some notes on the data:

• One of the important sources of data are the number Steam reviews. There is good evidence that these correlate strongly with copies sold, with frequently cited ratios of 50 sales per Steam review, but there's a wide range. It seems likely that the majority of Steam games fall between 25 and 120 sales per Steam review, but there are outliers. Also, games with a very small number of reviews are much more likely to be outliers in this respect. My own game is the only game I have hard sales numbers for. You can read my lengthy Reddit post on its release, but the relevant numbers are that it sold 1587 copies in the first week and 3580 copies in its first quarter.
• The total number of games in the sample was 115.
• I selected games semi-randomly from from both Popular Upcoming and All Upcoming. This favors the popular upcoming tab somewhat and this was deliberate: I wanted a diverse sample but also one not completely dominated with titles that sold zero copies.
• Games are ordered by their release date which range from 10/26/18 to 12/20/18.

Game	Price	Launch Discount	Week Guess	Week actual	3 Month	3 Month/week	Followers	Early Access	Demo	Review Score
Pit of Doom	9.99	0	7	27	43	1.592592593	295	Y	N	0.8
Citrouille	9.99	0.2	16	8	12	1.5	226	N	N
Corspe Party: Book	14.99	0.1	32	40	79	1.975	1015	N	N	0.95
Call of Cthulhu	44.99	0	800	875	1595	1.822857143	26600	N	N	0.74
On Space	0.99	0.4	0	0	0		4	N	N
Orphan	14.99	0	50	0	8		732	N	N
Black Bird	19.99	0	20	13	34	2.615384615	227	N	N
Gloom	6.99	0	20	8	17	2.125	159	N	N
Gilded Rails	5.99	0.35	2	3	7	2.333333333	11	N	Y
The Quiet Man	14.99	0.1	120	207	296	1.429951691	5596	N	N	0.31
KartKraft	19.99	0.1	150	90	223	2.477777778	7691	Y	N	0.84
The Other Half	7.99	0	2	3	27	9	91	N	Y	0.86
Parabolus	14.99	0.15	0	0	0		16	N	Y
Yet Another Tower Defense	1.99	0.4	20	22	38	1.727272727	396	N	N	0.65
Galaxy Squad	9.99	0.25		8	42	5.25	3741	Y	N	0.87
Swords and Soldiers 2	14.99	0.1	65	36	63	1.75	1742	N	N	0.84
SpitKiss	2.99	0	3	1	2	2	63	N	N
Holy Potatoes	14.99	0	24	11	22	2	617	N	N	0.7
Kursk	29.99	0.15	90	62	98	1.580645161	2394	N	N	0.57
SimpleRockets 2	14.99	0.15	90	142	272	1.915492958	3441	Y	N	0.85
Egress	14.99	0.15	160	44	75	1.704545455	7304	Y	N	0.67
Kynseed	9.99	0	600	128	237	1.8515625	12984	Y	N	0.86
11-11 Memories	29.99	0	30	10	69	6.9	767	N	N	0.96
Rage in Peace	12.99	0.1	15	10	42	4.2	377	N	N	0.85
One Hour One Life	19.99	0	12	153	708	4.62745098	573	N	N	0.81
Optica	9.99	0	0	2	3	1.5	18	N	N
Cybarian	5.99	0.15	8	4	18	4.5	225	N	N
Zeon 25	3.99	0.3	3	11	12	1.090909091	82	Y	N
Of Gods and Men	7.99	0.4	3	10	18	1.8	111	N	Y
Welcome to Princeland	4.99	0.1	1	15	55	3.666666667	30	N	N	0.85
Zero Caliber VR	24.99	0.1	100	169	420	2.485207101	5569	Y	N	0.73
HellSign	14.99	0	100	131	334	2.549618321	3360	Y	N	0.85
Thief Simulator	19.99	0.15	400	622	1867	3.001607717	10670	N	N	0.81
Last Stanza	7.99	0.1	8	2	4	2	228	N	Y
Evil Bank Manager	11.99	0.1		106	460	4.339622642	8147	Y	N	0.78
Oppai Puzzle	0.99	0.3		36	93	2.583333333	54	N	N	0.92
Hexen Hegemony	9.99	0.15	3	1	5	5	55	Y	N
Blokin	2.99	0	0	0	0	0	10	N	N
Light Fairytale Ep 1	9.99	0.1	80	23	54	2.347826087	4694	Y	N	0.89
The Last Sphinx	2.99	0.1	0	0	1	0	17	N	N
Glassteroids	9.99	0.2	0	0	0	0	5	Y	N
Hitman 2	59.99	0	2000	2653	3677	1.385978138	52226	N	N	0.88
Golf Peaks	4.99	0.1	1	8	25	3.125	46	N	N	1
Sipho	13.99	0	24	5	14	2.8	665	Y	N
Distraint 2	8.99	0.1	40	104	321	3.086538462	1799	N	N	0.97
Healing Harem	12.99	0.1	24	10	15	1.5	605	N	N
Spark Five	2.99	0.3	0	0	0	0	7	N	N
Bad Dream: Fever	9.99	0.2	30	78	134	1.717948718	907	N	N	0.72
Underworld Ascendant	29.99	0.15	200	216	288	1.333333333	8870	N	N	0.34
Reentry	19.99	0.15	8	24	78	3.25	202	Y	N	0.95
Zvezda	5.99	0	2	0	0	0	25	Y	Y
Space Gladiator	2.99	0	0	1	2	2	5	N	N
Bad North	14.99	0.1	500	360	739	2.052777778	15908	N	N	0.8
Sanctus Mortem	9.99	0.15	3	3	3	1	84	N	Y
The Occluder	1.99	0.2	1	1	1	1	13	N	N
Dark Fantasy: Jigsaw	2.99	0.2	1	9	36	4	32	N	N	0.91
Farming Simulator 19	34.99	0	1500	3895	5759	1.478562259	37478	N	N	0.76
Don't Forget Our Esports Dream	14.99	0.13	3	16	22	1.375	150	N	N	1
Space Toads Mayhem	3.99	0.15	1	2	3	1.5	18	N	N
Cattle Call	11.99	0.1	10	19	53	2.789473684	250	Y	N	0.71
Ralf	9.99	0.2	0	0	2	0	6	N	N
Elite Archery	0.99	0.4	0	2	3	1.5	5	Y	N
Evidence of Life	4.99	0	0	2	4	2	10	N	N
Trinity VR	4.99	0	2	8	15	1.875	61	N	N
Quiet as a Stone	9.99	0.1	1	1	4	4	42	N	N
Overdungeon	14.99	0	3	86	572	6.651162791	77	Y	N	0.91
Protocol	24.99	0.15	60	41	117	2.853658537	1764	N	N	0.68
Scraper: First Strike	29.99	0	3	3	15	5	69	N	N
Experiment Gone Rogue	16.99	0	1	1	5	5	27	Y	N
Emerald Shores	9.99	0.2	0	1	2	2	12	N	N
Age of Civilizations II	4.99	0	600	1109	2733	2.464382326	18568	N	N	0.82
Dereliction	4.99	0	0	0	0	#DIV/0!	18	N	N
Poopy Philosophy	0.99	0	0	6	10	1.666666667	6	N	N
NOCE	17.99	0.1	1	3	4	1.333333333	35	N	N
Qu-tros	2.99	0.4	0	3	7	2.333333333	4	N	N
Mosaics Galore. Challenging Journey	4.99	0.2	1	1	8	8	14	N	N
Zquirrels Jump	2.99	0.4	0	1	4	4	9	N	N
Dark Siders III	59.99	0	2400	1721	2708	1.573503777	85498	N	N	0.67
R-Type Dimensions Ex	14.99	0.2	10	48	64	1.333333333	278	N	N	0.92
Artifact	19.99	0	7000	9700	16584	1.709690722	140000	N	N	0.53
Crimson Keep	14.99	0.15	20	5	6	1.2	367	N	N
Rival Megagun	14.99	0	35	26	31	1.192307692	818	N	N
Santa's Workshop	1.99	0.1	3	1	1	1	8	N	N
Hentai Shadow	1.99	0.3		2	12	6	14	N	N
Ricky Runner	12.99	0.3	3	6	13	2.166666667	66	Y	N	0.87
Pro Fishing Simulator	39.99	0.15	24	20	19	0.95	609	N	N	0.22
Broken Reality	14.99	0.1	60	58	138	2.379310345	1313	N	Y	0.98
Rapture Rejects	19.99	0	200	82	151	1.841463415	9250	Y	N	0.64
Lost Cave	19.99	0	3	8	11	1.375	43	Y	N
Epic Battle Fantasy 5	14.99	0	300	395	896	2.26835443	4236	N	N	0.97
Ride 3	49.99	0	75	161	371	2.304347826	1951	N	N	0.74
Escape Doodland	9.99	0.2	25	16	19	1.1875	1542	N	N
Hillbilly Apocalypse	5.99	0.1	0	1	2	2	8	N	N
X4	49.99	0	1500	2638	4303	1.63115997	38152	N	N	0.7
Splotches	9.99	0.15	0	2	1	0.5	10	N	N
Above the Fold	13.99	0.15	5	2	6	3	65	Y	N
The Seven Chambers	12.99	0.3	3	0	0	#DIV/0!	55	N	N
Terminal Conflict	29.99	0	5	4	11	2.75	125	Y	N
Just Cause 4	59.99	0	2400	2083	3500	1.680268843	50000	N	N	0.34
Grapple Force Rena	14.99	0	11	12	29	2.416666667	321	N	Y
Beholder 2	14.99	0.1		479	950	1.983298539	16000	N	N	0.84
Blueprint Word	1.99	0		12	15	1.25	244	N	Y
Aeon of Sands	19.99	0.1	20	12	25	2.083333333	320	N	N
Oakwood	4.99	0.1		32	68	2.125	70	N	N	0.82
Endhall	4.99	0	4	22	42	1.909090909	79	N	N	0.84
Dr. Cares - Family Practice	12.99	0.25	6	3	8	2.666666667	39	N	N
Treasure Hunter	16.99	0.15	200	196	252	1.285714286	4835	N	N	0.6
Forex Trading	1.99	0.4	7	10	14	1.4	209	N	N
Ancient Frontier	14.99	0	24	5	16	3.2	389	N	N
Fear the Night	14.99	0.25	25	201	440	2.189054726	835	Y	N	0.65
Subterraneus	12.99	0.1	4	0	3	#DIV/0!	82	N	N
Starcom: Nexus	14.99	0.15		53	119	2.245283019	1140	Y	N	0.93
Subject 264	14.99	0.2	25	2	3	1.5	800	N	N
Gris	16.9	0	100	1484	4650	3.133423181	5779	N	N	0.96
Exiled to the Void	7.99	0.3	9	4	11	2.75	84	Y	N

Column Explanations
For the columns that are not self-explanatory:

• Launch Discount: Percent first week discount, 0.25 = 25% off
• Week Guess: This is my guess, made before the game launched as to how many Steam purchaser reviews it would have after exactly one week.
• Week Actual: The number of reviews that the game had after 1 week.
• 3 Month: The number of reviews that the game had after 3 months.
• Followers: The number of group followers the game had prior to launch. In some cases this recorded just before launch, in some cases up to a week before.
• Review score: The percent favorable score on Steam at the one month mark. Games needed a minimum of 20 reviews to be counted.

Question 1: Does Quality Predict Success?

There was a recent blog post stating that the #1 metric for indie games' success is how good it is.
Quality is obviously a subjective metric. The most obvious objective measure of quality for Steam games is their % Favorable Review score. This is the percentage of reviews by purchasers of the game that gave the game a positive rating. I excluded any game that did not have at least 20 user reviews in the first month, which limited the sample size to 56.
The (Pearson) correlation of a game's review score to its number of reviews three months after its release was -0.2. But 0.2 (plus or minus) isn't a very strong correlation at all. More importantly, Pearson correlation can be swayed if the data contains some big outliers. Looking at the actual games, we can see that the difference is an artifact of an outlier. Literally. Valve's Artifact by far had the most reviews after three months and had one of the lowest review scores (53% at the time). Removing this game from the data changed the correlation to essentially zero.
Spearman's Rho, an alternative correlation model that correlates rank position and minimizes the effect of huge outliers produced a similar result.
Conclusion: If there is correlation between a game's quality (as measured by Steam review score) and first quarter sales (as measured by total review count), it is too subtle to be detected in this data.

Question 2: Do Demos, Early Access or Launch Discounts Affect Success/Failure?

Unfortunately, there were so few games that had demos prior to release (10) that only a very strong correlation would really tell us anything. As it happens, there was no meaningful correlation one way or another.
There were more Early Access titles (28), but again the correlation was too small to be meaningful.
More than half the titles had a launch week discount and there was actually a moderate negative correlation of -0.3 between having a launch discount and first week review count. However it appears that this is primarily the result of the tendency of AAA titles (which sell the most copies) to not do launch discounts. Removing the titles that likely grossed over a $1 million in the first week reduced the correlation to basically zero.
Conclusion: Insufficient data. No clear correlation between demos, Early Access or launch discount and review counts: if they help or hurt the effect is not consistent enough to be seen here.

Question 3: Does pre-launch awareness (i.e., Steam followers) predict success?

You can see the number of "followers" for any game on Steam by searching for its automatically-created Community Group. Prior to launch, this is a good rough indicator of market awareness.
The correlation between group followers shortly before launch and review count at 3 months was 0.89. That's a very strong positive correlation. The rank correlation was also high (0.85) suggesting that this wasn't the result of a few highly anticipated games.
Save for a single outlier (discussed later), the ratio of 3 month review counts to pre-launch followers ranged from 0 (for the handful of games that never received any reviews) to 1.8, with a median value of 0.1. If you have 1000 followers just prior to launch, then at the end of the first quarter you should expect "about" 100 reviews.
One thing I noticed was that there were a few games that had follower counts that seemed too high compared to secondary indicators of market awareness, such as discussion forum threads and Twitter engagement. After some investigation I came to the conclusion that pre-launch key activations are treated as followers by Steam. If a game gave away a lot of Steam keys before launch (say as Kickstarter rewards or part of beta testing) this would cause the game to appear to have more followers than it had gained "organically."
Conclusion: Organic followers prior to launch are a strong predictor of a game's eventual success.

Question 4: What about price?

The correlation between price and review count at 3 month is 0.36, which is moderate correlation. I'm not sure how useful that data point is: it is somewhat obvious that higher budget games have larger marketing budgets.
There is a correlation between price and review score of -0.41. It seems likely that players do factor price into their reviews and a game priced at $60 has a higher bar to clear to earn a thumbs up review than a game priced at $10.

Question 5: Do first week sales predict first quarter results?

The correlation between number of reviews after 1 week and number of reviews after 3 months was 0.99. The Spearman correlation was 0.97. This is the highest correlation I found in the data.
Excluding games that sold very few copies (fewer than 5 reviews after the first week), most games had around twice as many reviews after 3 months as they did after 1 week. This suggests that games sell about as many copies in their first week as they do in the next 12 weeks combined. The vast majority of games had a tail ratio (ratio of reviews at 3 months to 1 week) of between 1.3 to 3.2.
I have seen a number of questions from developers whose game had a poor launch on Steam and wanted to know what they can do to improve sales. While I'm certain post-launch marketing can have an effect on continuing sales, your first week does seem to set hard bounds on your results.
Conclusion: ALL SIGNS POINT TO YES

Question 6: Does Quality Help with a Game's "Tail"?

As discussed in the last question while first week sales are very strongly correlated with first quarter, there's still quite a wide range of ratios. Defining a game's Tail Ratio as the ratio of reviews after 3 months to after 1 week, the lowest value was 0.95 for "Pro Fishing Simulator" which actually managed to lose 1 review. The highest ratio was 6.9, an extreme outlier that I'll talk about later. It is perhaps not a coincidence that the worst tail had a Steam score of 22% and the best tail had a Steam score of 96%.
The overall correlation between the Tail Ratio and Steam score was 0.42.
Conclusion: Even though there is no clear correlation between quality and overall review count/sales, there is a moderate correlation between a game's review score and its tail. This suggests that "good games" do better in the long run than "bad games," but the effect is small compared to the more important factor of pre-launch awareness.

Question 7: Is it possible to predict a game's success before launch without knowing its wishlists?

While I was compiling the data for each game, sometime prior to its scheduled launch date, I would make a prediction of how many reviews I thought it would receive in its first week and add that prediction to the spreadsheet.
The #1 factor I used in making my prediction was group follower count. In some cases I would adjust my prediction if I thought that value was off, using secondary sources such as Steam forum activity and Twitter engagement.
The correlation between my guess and the actual value was 0.96, which is a very strong correlation. As you can see in the data, the predictions are, for the most part, in the right ballpack with a few cases where I was way off.
Based on my experience, multiplying the group follower count by 0.1 will, in most cases, give you a ballpark sense of the first week quarter review count. If a game doesn't have at least one question in the discussion forum for every 100 followers, that may indicate that there are large number of "inorganic" followers and you may need to adjust your estimate.
Conclusion: Yes, with a few exceptions, using follower data and other indicators you can predict first week results approximately. Given the strong correlation between first week and quarter sales, it should also be possible to have a ballpark idea of first quarter results before launch.

Final Question: What about the outliers you mentioned?

There were a few games in the data that stood out significantly in one way or another.
Outlier #1: Overdungeon. This game had 77 group followers shortly before launch, a fairly small number and based solely on that number I would have expected fewer than a dozen reviews in the first week. It ended up with 86. Not only that, it had a strong tail and finished its first quarter with 572 reviews. This was by a wide margin the highest review count to follower ratio in the sample.
Based on the reviews, it appears to basically be Slay the Spire, but huge in Asia. 90% of the reviews seem to be in Japanese or Chinese. If anyone has some insight to this game's unusual apparent success, I'm very curious.
This seems to be the only clear example in the data of a game with minimal following prior to launch going on to having a solid first quarter.
Outlier #2: 11-11 Memories Retold. This game had 767 group followers shortly before launch, ten times as many as Overdungeon. That's still not a large number for even a small indie title. It had a fair amount going for it, though: it was directed by Yoan Fanise, who co-directed the critally acclaimed Valiant Hearts, a game with a similar theme. It was animated by Aardman Studios of "Wallace and Gromit" fame. Its publisher was Bandai Namco Europe, a not inexperienced publisher. The voice acting was by Sebastian Koch and Elijah Wood. It has dozens of good reviews in both gaming and traditional press. It currently has a 95% positive review rating on Steam.
Despite all that, nobody bought it. 24 hours after it came out it had literally zero reviews on Steam. One week after it came out it had just 10. Three months later it had demonstrated the largest tail in the data, but even then it had only climbed to 69 reviews. Now it's at about 100, an incredible tail ratio, but almost certainly a commercial failure.
This is a solid example that good game + good production values does necessarily equal good sales.

Final notes:
The big take-aways from this analysis are:

• The success of a game on Steam very strongly depends on its first week performance
• A game's first week performance is strongly correlated with its pre-launch market awareness
• Quality does not seem to strongly impact first week performance, but may have some positive effect on a game's "tail"
• All inferences regarding sales are dependent on the relationship between review counts and sales

Thanks for reading!