The number that tells you a window is closing
Counting how many games are already in a genre tells you whether you are late. Counting how fast that number is changing tells you whether you are about to be. We learned to watch the second one, and the lesson was mildly humiliating.
Absolute volume is a lagging indicator
If a category already has hundreds of competing products, you do not need a measurement to tell you it is crowded. You can see it. By the time supply is obviously high, the window closed some time ago, and the only question left is how expensive it is to find that out.
What is genuinely useful is noticing crowding while it is still small enough to look like an opportunity. That means watching a rate rather than a level. Of everything we changed in how we read market data, this was the one that mattered most, and it cost us nothing.
The reason is structural rather than clever. Supply grows because teams decide to enter, and those decisions are made weeks or months before a product shows up in any ranking, so the count you can read today is the visible end of a pipeline that was already full, whereas the rate of change is the earliest public trace of decisions being made right now. That is the whole argument. The rest of this post is bookkeeping.
The case that convinced us
Stress-relief and “oddly satisfying” games looked like a reasonable target. Modest domestic validation, low difficulty, low localisation sensitivity. On a snapshot reading, nothing about the category screamed danger. I remember being quietly pleased with it, which in hindsight is the warning sign I should have learned to recognise.
Then we tracked it over thirty days. The count of monitored products in the category went from 1,648 to 2,468, an increase of 49.8 percent in a single month. The entire top five of the ad-spend ranking was already this genre, with one title running 35,537 creatives on its own.
Thirty-five thousand creatives. I sat and stared at that number for a while, then went back to check the source, because I assumed I had misread it. I had not.
A market that adds half again as many competitors in a month is not a market with room in it, whatever its absolute size suggests. We marked the window as closing and stopped considering the category. Not because it is crowded today. Because of the direction. Had we looked only at the level, we would have called it open, and we would have started something we later had to abandon.
Three proxies, and what each actually measures
Nobody publishes the number you actually want, which is how many people are making money in a category. So we watch three things that are public and that correlate with crowding. We are also careful about which of them is evidence of what, because it is very easy to let one quietly turn into the other.
| Observable | What it really measures | What it does not mean |
|---|---|---|
| Count of products tracked in a category | How many teams have chosen to enter | That any of them are succeeding |
| Distinct ad creatives in a period | How much money is being spent on attention | Revenue, or user numbers, or quality |
| Cumulative impression counts | How long something has been visible | Current performance; these accumulate and never fall |
The second row is the one that gets misused most often. Ad libraries report spending behaviour. Top of the ad chart means someone is spending aggressively, which is equally consistent with a category being lucrative and with a category being fought over until nobody is making money. We use it only as evidence of crowding, because that is what it directly shows. Nothing more.
Cumulative figures need the most care. A line-drawing puzzle title showing over eighty million impressions sounds overwhelming. It is not. That number has been accumulating for as long as the game has existed and will never decrease, so it says nothing about whether the category is hot now. We treat cumulative counters as historical context. Never as a current signal. Ever.
How we turned it into a rule
Rather than eyeballing trend lines, we made it mechanical. The supply score for every category is re-derived on a fixed schedule, and a move of two points or more in a week raises a warning: upward means the window is closing, downward means supply is receding and a closed window may be reopening.
Making it mechanical matters for a reason that has nothing to do with accuracy. A rule does not get to be optimistic. If the schedule says the supply score is re-derived on Monday, then on Monday it gets re-derived, whether or not I am busy with something more interesting and whether or not I would prefer the answer to come out differently.
Two points is deliberately coarse. Our scores are subjective single digits, so pretending to read a half-point move would be false precision. A threshold wide enough that we only react to changes we could actually defend has been far more useful than a sensitive one we would have to argue about every week.
The other half of the rule is its complement: when there is no new evidence for a category, we do not change its score. A number that drifts because it feels like it should have moved is worse than a stale number honestly labelled stale. I have broken this rule. The drift always went the way I already hoped it would.
Where this goes wrong
Rates are noisy when the base is small. A category going from four products to eight is a 100 percent increase and means almost nothing. The same percentage on a base of sixteen hundred is a real event. We look at absolute change alongside the percentage for exactly this reason, and we do not let a percentage on a tiny base trigger anything at all.
There is also a mixing problem that is easy to commit and hard to notice. Different sources update on different cadences. Some ranking sites publish monthly, some weekly, some only when somebody bothers to write it up. Compare a figure from one month against a figure from a different source in another month and you get a trend that is pure artifact. The artifact is worse than useless, because it arrives looking like evidence, it has a clean percentage attached to it, and it is exactly the kind of number that survives into a decision, by which point the mismatch in time windows is three layers down and nobody thinks to ask about it again. We record the time window of every piece of evidence, and we refuse to combine figures whose windows do not match.
And the honest limitation: watching supply cannot tell you about demand. It never will. A window can be wide open because nobody has thought of the thing, or because nobody wants it, and from the supply side those two situations look identical. The only defence is the one we described in the scoring post, which is to ask why the gap exists before treating it as an opening.
What we would keep
If you take one thing from this, take the second item on the list below rather than the first, because writing down what a metric does not measure is the step everybody skips, and it is the step that actually saves you: the day you catch yourself quoting an ad-spend ranking as though it were a revenue ranking is the day you begin making decisions about a market that does not exist, and you will not notice, because the number looks precisely like the number you were hoping for.
- Track the derivative, not just the level. Direction is available earlier than magnitude.
- Be explicit about what each metric measures, and do not let a spend metric become a success metric.
- Never compare figures from mismatched time windows.
- Set thresholds coarse enough that you only react to defensible changes.
- When evidence is missing, say so and leave the score alone.
None of this makes market timing reliable. It makes it slightly less of a guess. More importantly, it gives us a written reason for having decided something, which is what the window column is supposed to be, which is the only way we have found to notice later that the reason has stopped being true. The writing down is the whole trick. Memory is far too generous to our past decisions.