Every competitive intelligence product on the market is good at the same thing: telling you that something happened. A changelog updated. A job posting appeared. A press release went out. The category has spent a decade getting very good at detection, and detection is genuinely the easy half.
The half nobody reports on is what happened next. So we measured it.
What we measured
Across nine sectors and 83 companies, our engine scored 2,418 distinct competitor moves — ships, pricing changes, hiring shifts, releases — over a single reporting period. For each one, it then looked for public reaction and bound that reaction to the specific dated launch that earned it, rather than counting mentions loosely against a company name.
982 reactions came back coupled to a specific move, and 290 moves drew at least one. That is 12.0% of everything those companies shipped.
Put the other way round: 88.0% of competitor activity produced no measurable public response whatsoever.
Why that number should change how you read a feed
If roughly seven in eight competitor moves land to silence, then a tool that alerts you on all of them is, by construction, about 88% noise. That is not a tuning problem you can fix with better filters on the detection side, because the input signal genuinely does not contain the answer. Whether a launch mattered is not a property of the launch. It is a property of the response.
It also means the default ranking in most competitive intelligence dashboards — reverse chronological — is close to the worst available ordering. It puts this morning's cosmetic changelog entry above the Monday release that a market actually reacted to.
The rate is not the same everywhere
The aggregate hides a wide spread. AI coding tools landed 23.8% of their moves. B2B fintech landed 4.2%. That is a 5.7x difference in how often shipping produces a visible reaction, between two sectors of comparable company count in the same period.
AI infrastructure came in at 20.8%, DevOps at 17.6%, cloud security at 15.9%, design and collaboration at 14.2%, data and analytics at 11.1%, developer tools at 9.1%. The ordering is stable enough to be useful: some markets are simply loud, and some conduct their business somewhere you cannot see.
The honest caveat, which matters more than the headline
This is measured from public reaction — Hacker News, Bluesky, public repositories, news. That is a real signal, and it is the same signal for every company we score, which is what makes the comparison fair. But it is not a neutral instrument across sectors.
A developer-tools launch and a B2B fintech launch do not have equally observable audiences. Fintech buyers largely do not post about procurement decisions on Hacker News. So fintech's 4.2% should be read as 'this market discusses itself somewhere we cannot see', not as 'fintech launches don't matter'. Reception measurement is a targeting problem before it is a coverage problem, and any vendor who reports a cross-sector reception number without saying so is selling you a confound.
Within a sector, against companies whose audiences behave alike, the comparison is sound. That is the scope in which we use it, and it is the scope in which we publish it.
What we do with it
Two things. First, rank by reception rather than recency, so the two or three moves that actually earned a response sit at the top of the week and the rest can be ignored with a clear conscience.
Second — and this is the part that changes roadmaps — look specifically for the moves that scored high on significance and low on reception. A competitor shipping something genuinely substantial into total silence is the most actionable signal in the dataset. It means the need wasn't felt, or the framing missed, or the lane is still open. A detection-only feed files that same event as a competitor win.
Every figure above is reproducible from the published sector reports, each of which cites the source URL and timestamp behind every line. We would rather you check the numbers than trust them.
Revised after a coupling audit: reactions were being bound to launches without a causality check, so an old forum thread naming a company could attach to a recent unrelated release. 1,099 of 2,885 couplings were removed and the coupler now rejects them. The direction of the error was always the same — every landed rate here is lower than first published.