What a discount actually does to an indie game on Steam

Discounting into the summer sale is widely held to be either the best or the worst thing a small game can do. Compared across games the evidence looks decisive, and it points the unhappy way: in June and July a discounted month runs at 1.04 while a quiet one runs at 1.20, so discounting appears to do actual harm. Outside those months the order is the ordinary one, 1.00 against 0.82.

That comparison is wrong, and it is wrong in a way worth naming, because it is the same mistake most discount advice rests on. The games that discount in summer are not the same games that discount in November. Comparing each game against itself, across 2,678 that did both, the difference disappears: the median gap is 0.00 and 50 percent do worse in summer, which is a coin flip.

When is not the question. What and how deep are.

Measured over 350,907 game months from 5,618 games, a month with no discount runs at 0.87 of its own recent baseline. That is not a failure, it is the ordinary decay of a catalogue title.

A discount averaging 37.7 percent off brings that to 0.95, and one averaging 74.7 percent off brings it to 1.00. The honest way to read those numbers is that a discount does not launch a game upward. It arrests the fall.

The effect is not confined to a lucky few. Comparing each game against itself, 83 percent of 3,181 games do better in their own discount months, at 1.00 against 0.84.

Following 479 discounts that arrived after a quiet run-up, and scoring the whole window against the baseline from before the discount, the discount month lands at 1.33 and the month after at 1.46. The second number being the larger of the two is a property of the measure rather than of the sale: a review is written some time after the purchase that produced it.

The question that matters more is what happens next, and it cannot be answered without something to compare against, because a three month window drifts down whether or not anything happened in it. Across 144,730 windows of the same shape with no discount at all, the three months run 0.94, 0.93 and 0.92.

Two months out the discounted game sits at 0.94, against 0.92 for the game that did nothing. It returns to the ordinary level instead of dropping below it. There is no payback period, and whatever a discount brings in, it does not come out of next quarter.

Cutting deeper does lift more. How much more depends on how strictly the question is asked, and rather than pick the definition that flatters the result, all three are reported here.

Asked strictly, on games with a clean run-up, the gain from going deep is 0.33 across 80 games. Relaxing the run-up condition and widening the bands brings the sample to 2,128 games and the gain to 0.12. The direction holds in all three variants and the magnitude falls by roughly two thirds as the sample grows, which is the ordinary fate of an effect measured on few cases.

A study that shows how its number moves when the filters are relaxed is harder to argue with than one that reports the cell that suited it. The conclusion that follows does not depend on which variant is chosen, and that is the reason to show them.

Going from an average of 37.7 percent off to an average of 74.7 percent off multiplies copies by 1.05 and multiplies the price of each copy by 0.41. Revenue therefore lands at 0.43 of where the shallower cut left it.

To come out even, the deeper cut would have to move 146.2 percent more copies. Nothing measured here comes close, and the margin is wide enough that it survives being wrong: even if the lift were several times what any of the three variants found, the deeper cut would still take home less.

This is the finding a developer can act on tomorrow. The deepest discount is a tool for reach, for wishlist conversion and for a spot on a sale page. It is not a tool for revenue.

Almost every deep discount in this pool is a repeat: 95 percent of them follow another within 1 to 3 months, and those land at 1.00, level with the months around them. Wait 4 to 6 months and the same depth of cut lands at 1.26. Wait 7 to 11 months and it lands at 1.44, after which the gradient flattens.

Part of that gradient is scarcity and part of it is the measure: a game that discounts constantly is being scored against its own discounted months, so a baseline that is already high leaves less room to rise. The two cannot be separated here, which is a reason to read the size of the gradient with suspicion and its direction without.

The odd result sits at the other end. The first discount of a game's life lands at 1.00, no better than a repeat, on 1,192 cases. A first discount tends to arrive while the game still has momentum of its own, and momentum is exactly what the baseline is measuring.