A 90 percent discount does not mean the game is bad
The scariest number on the store page is mostly a birthday.
There is a reflex that kicks in at about 75 percent off. If it is this cheap, what is wrong with it.
I went looking for evidence of that, expecting to find it. The catalog appears to agree at first. On August 19, 2026, sorting every live sale by how deep the cut was, the average BANG Score fell as the discount grew:
| Discount | Average BANG |
|---|---|
| 1 to 24 percent off | 39.8 |
| 25 to 49 percent | 39.2 |
| 50 to 74 percent | 37.7 |
| 75 percent or more | 36.7 |
I ran it again on September 23, 2026, over the 5,728 games on sale that day, and got the same shape: 42.1, 43.3, 35.7 and 35.6. That looks like a rule. It is a trick played by who ends up in each row.
The same table, with the shovelware removed
Restrict it to games with at least a thousand reviews, meaning games a real number of people actually bought and played, and the fall goes away. On August 19, 2026 the pattern even turned around:
| Discount | Average BANG, 1000+ reviews |
|---|---|
| 1 to 24 percent off | 56.9 |
| 25 to 49 percent | 59.2 |
| 50 to 74 percent | 62.6 |
| 75 percent or more | 61.6 |
On September 23, 2026 it did not turn around: the same four bands read 54.7, 57.4, 53.9 and 52.7. What held both times is the part that matters. Every band sits well above its row in the first table, and the four sit within six points of each other. Among games with an audience, a deep sale does not mean a worse game.
Both views are true. Only the second one is about the question you are asking.
What the first table was actually measuring
Age. From the August 19, 2026 snapshot:
| Discount | Average release year | Median review count |
|---|---|---|
| 1 to 24 percent off | 2025.0 | 480 |
| 25 to 49 percent | 2024.5 | 5,212 |
| 50 to 74 percent | 2022.5 | 4,133 |
| 75 percent or more | 2020.2 | 9,441 |
A game discounted 10 percent is, on average, a 2025 release. A game discounted 80 percent is a 2020 release. Publishers do not cut a new game hard, and they stop protecting an old one. The September 23 run showed the same slope: 2024.9, 2024.1, 2021.4 and 2019.3.
The shallow band is stuffed with brand new games, median review count 480, because a game released last month has not had time to gather any. Small, unknown, recent games drag that first row's average down, and the deep-discount row gets the games that have been around long enough for ten thousand people to have an opinion. The score gap in the first table is the gap between famous and unknown, wearing a discount label.
So what does a big discount tell you
That the game is old, and that its publisher has stopped defending the launch price. Neither of those is a defect. Most of the best value on Steam is a five year old game at 80 percent off, and the mechanism above is the whole reason why.
The discount tells you almost nothing on its own. It is a percentage off a number the publisher chose, and the publisher chooses it knowing you will read the percentage.
What to look at instead
The recorded low. Across the games I track a price history for, the lowest recorded price averages 55 percent below list. If today's cut is 40 percent and the record is 75, this is not the sale you are waiting for. Every game page here prints its own recorded low, and the Value Desk ranks today's sales by the score each price lands at.
Whether it is at the floor right now. In the August 4, 2026 snapshot of every live sale, 69 percent of the discounted games were sitting at the deepest cut their history had ever recorded. When a sale lands it usually lands at the bottom, so a cut that is nowhere near the floor is unusual and worth ignoring.
The first sale takes about four months. Measured on August 19, 2026 across 34,640 games with both a release date and a discount on record, the first discount arrives an average of 125 days after release. If you are waiting on something new, that is the number.
Where this is weak
The tables use averages, and averages hide the shape of a distribution. A thousand reviews is an arbitrary line. I picked it because it is roughly where a game stops being invisible, and nothing in particular happens at that number.
The bigger hole is that this only sees games that got discounted at all. A game nobody ever put on sale has no row in the price history, so the whole analysis is conditioned on a sale having happened at least once. That biases everything here toward games with publishers who run sales.
And the release year is the Steam store's date. Ports, re-releases and games that spent three years in early access all carry a date that means something different from the one next to it.