Almost every company has an ICP document. Almost none of them can tell you, right now, how many companies currently match it — or which ones started matching it last week.
That gap is the difference between a marketing artefact and an operating input.
The document problem
A written ICP says things like "mid-market B2B SaaS companies experiencing rapid growth with a modern data stack". Every word of that is unactionable. Mid-market is a range somebody has to pick. Rapid growth is a threshold somebody has to define. Modern data stack is a list of tools somebody has to enumerate.
So each team picks their own. Paid targets one thing, outbound another, and sales complains about lead quality in both directions. Nobody is wrong, because nothing was ever specified.
What an executable ICP looks like
The same profile, written as a query, forces every ambiguity into the open:
- Headcount between 80 and 800, from a source you have measured, not a self-reported field.
- Headcount grew more than 15% in the last two quarters.
- At least one of a named list of tools detected in the last 90 days.
- At least one open role in revenue or data functions.
- Not currently a customer, not in an open opportunity, not closed-lost within 90 days.
It is less elegant than the slide. It is also countable, testable, and identical for every channel that consumes it.
It is less elegant than the slide. It is also countable, testable, and identical for every channel.
The number you get for free
Once the ICP is a query, it returns a count. That count is one of the most useful numbers in the business and most teams have never seen it.
If it returns four hundred accounts, your entire go-to-market strategy is different from a team whose query returns forty thousand — different channels, different economics, different sales motion. Teams routinely run high-volume outbound against a market of a few thousand companies and burn it out in two quarters, because nobody counted first.
Version it like code
Because the query is executable, you can change it and observe what happens. Loosen the headcount band and watch reply rate. Add a signal and watch win rate three months later. Keep the versions; each one is a hypothesis with a result attached.
That is the real unlock. A document cannot be wrong in a way you can measure. A query can — which means it can also get better.