Most ICP exercises fail because they start with a blank page and a brainstorm. Start with your invoices instead.
1. List your last ten clients
All of them, not the flattering ones. Next to each, write the fee, how long the project ran, and whether you’d take it again. Be honest about the last column.
2. Split them into good and bad
Good means: paid properly, respected the process, you’d take another. Bad means the opposite, regardless of how well-known the logo is. Two or three in each column is enough to see a pattern.
3. Find what the good ones share
Look for boring, checkable facts, not vibes. Sector. Headcount. Whether they had an in-house team or none. Who signed off. Whether they were growing, restructuring, or entering a new market.
The pattern is almost never "they liked our work". It’s usually a structural fact: a size band, a missing internal function, or a moment of change.
4. Write the filter
Turn the pattern into something you could hand to a stranger and have them build a list. Geography, sector, size, signals, buyer role. If any line can’t be checked from outside the company, cut it.
5. Test it on twenty companies
Build the list and look at it properly. If more than a third feel obviously wrong, your filter is off, not the market. Tighten one line, usually size or sector, and run it again.
6. Expect to revise it
An ICP isn’t a one-off document. The version that works in January will be slightly wrong by June, because your capability and your market both move. Revisit it whenever a project goes badly and ask whether the ICP predicted it.
One ICP or several?
One per distinct sales motion. If you sell two different services to two different audiences, that’s two ICPs, not one blurry one. Two sharp filters will always beat one that tries to cover everything.