Build an ICP from public evidence, not a persona workshop

Persona workshops produce a slide. On it is a stock photo, a name like Marketing Mary, a bulleted list of goals and frustrations, and a quote nobody said. The slide goes in the deck and the outbound list gets built from a headcount filter anyway, because nothing on the slide can be turned into a query.

The test for an ICP is narrow: can you look at a company from outside and decide, in under a minute, whether it belongs on the list? If the answer needs a discovery call, it is not a targeting criterion. It is a qualification criterion, and it belongs later.

Start from evidence you can see

Write your criteria so that each one points at something public. A worked example, for a company selling content research to marketing teams:

• Publishes on a schedule. Check the last eight weeks of one channel. Regular cadence, or nothing since spring.
• Runs more than one brand or market. Check for a second language, a second account, a regional site.
• Owns the channel internally. Check whether posts carry a house voice or an agency template.
• Competes on message rather than on price. Check whether the copy argues a position or lists a discount.

Each of those takes a minute per company and gets checked without asking anyone. Compare with the workshop version: "values data-driven decision-making, feels overwhelmed by content volume." True, probably. Unusable.

Behaviour beats firmographics

Headcount, sector and revenue band describe the container. What a company does in public describes the pressure inside it.

Two agencies with 25 employees each can be entirely different prospects. One publishes client case studies twice a month and argues about positioning in public. The other has not posted since a rebrand two years ago. The database row is identical. The likelihood of a reply is not.

This is the same argument as reading hiring signals instead of company profiles, applied to a different public surface. A job ad shows what a company is short of. A publishing pattern shows what a company invests in. Both beat the container.

The practical version for outbound targeting: for every firmographic filter in your list build, add one behavioural filter that requires somebody to look. The list gets shorter and the reply rate stops embarrassing you.

Write the exclusions down

Most ICP documents list who to include. The money is in who to exclude, because exclusions are what stop a campaign burning three weeks on companies that were never going to buy.

Ours, written plainly:

• Companies that post twice a quarter. They have no recurring pain to solve.
• Single-market companies with one product line and no competitor set worth reading.
• Buyers who want a report rather than a workflow. They want a consultant, and we would be a bad one.
• Anyone whose first question is whether the tool writes the posts as well. Different product, different expectation, unhappy customer.

That last one hurt to write and it saves a week a month. An exclusion list only earns its place when it removes companies you would otherwise have wanted.

Test the ICP against your own funnel

If you have customers, the ICP is discoverable rather than inventable. Take the ten who renewed, stayed engaged, or referred someone. Find the public attribute they share that the ten who churned do not.

It is rarely the attribute you expected. Often it is unglamorous: the ones who stayed had somebody whose job title made them personally accountable for the channel. Nobody puts that on a persona slide because it is not aspirational.

If you have no customers, you are writing a hypothesis. Say so out loud, in the document, with the date on it. The failure mode is a first-guess ICP that hardens into doctrine because it got a nice slide and nobody revisited it after the first fifty replies came back from the wrong companies.

Keep it narrow enough to say something specific

A narrow ICP makes the copy easy. If your list is "European B2B companies, 50 to 500 staff", the only opening line available is one that could be sent to anyone, and it reads that way.

If the list is "recruitment agencies in Poland who have advertised the same blue-collar role four or more times since June", the opening line writes itself and no competitor is sending it. Ripe Leads works this way in their staffing campaigns, segmenting down to role clusters and repeated vacancies rather than "recruitment agencies in Poland".

There is a floor. Cut too far and the segment holds forty companies, which is not enough to learn from and not enough to sell to. When that happens, keep the narrow definition for the copy and widen one criterion for the list, rather than diluting all of them at once.

The document

Two pages. Inclusion criteria, each with the public source that proves it. Exclusion criteria, with the reason. The date, and the evidence the current version rests on.

Then revisit it after every fifty conversations. An ICP is a claim about the market, and the market answers.

Next: what to do when the list is right and the message is wrong.