ICP template: ten questions and a scoring sheet

An ideal customer profile (ICP) is a written description of the companies and people most likely to buy and stay. Answer ten questions about your best customers, turn the answers into a scoring sheet with points per trait, and grade every lead before you contact it. On our own LinkedIn outbound the audience changed meetings per person tenfold, with the same agents and offer.

Most ICPs are one paragraph on a slide that nobody uses. A useful one is a tool: it tells you who to write to, who to skip and which message to send. This post gives the template, the sheet and the way to test both.

Why does the ICP matter more than the message?

Because the same message sent to a different group gives a different result. On our own LinkedIn outbound, with the same agents and the same offer, people who had engaged with posts from AI SDR vendors and automation creators booked 2.6% to 3.7% meetings per person messaged. People who had engaged with enterprise sales voices booked under 0.4%. Replies stayed near 40% for both: the second group was perfectly polite and not buying.

The Green Shield Logistics study shows the same idea from another side. Named heads of quality, ESG, procurement and logistics replied at 8.2%, almost four times the 2.2% of clinic and lab inboxes, and produced 17 of the 18 positive replies. The owner of the problem answered, the generic inbox did not.

What are the ten questions?

Answer them from your best five or ten customers, not from what you wish your customers looked like.

1. Which customers are happiest? Name five. Why are they happy?
2. What do they have in common? Size, sector, geography, stage.
3. What problem do they have when they buy? In their words, not yours.
4. What triggers the purchase? A hire, a funding round, a new regulation, a failed vendor, a season.
5. Who owns the problem? Title and department. Who signs?
6. Who else must agree? Another team, a finance approver, a technical reviewer.
7. What do they use today? A competitor, a spreadsheet, nothing.
8. Where do they spend time online? LinkedIn, an industry forum, a newsletter, a group.
9. Who is not a fit, even if they would buy? Too small, too complicated, wrong country, wrong risk.
10. What would make you stop writing to this group? The number of messages or weeks without a reply that signals a wrong guess.

Question 9 is the one people skip, and it is the one that saves the most time. Exclusion rules protect the sender as well: Sellinger's rule packs and persona templates let you write "do not contact" conditions into the agent.

How does the scoring sheet work?

Turn the answers into traits and give each trait points. A lead is scored by adding the points for the traits it has.

Rules for the sheet:

• Keep it short. Seven traits is plenty. More makes the sheet hard to defend.
• Weight the trigger. The trait that says why now should carry the most points.
• Use a negative for blockers. A blocker should be able to cancel a lot of good traits.
• Set two lines. For example, 10 or more is contact now, 6 to 9 is contact later, below 6 is skip.

Then grade a sample of 100 leads by hand. If more than half are in the skip group, your list source is wrong, not your sheet. The glossary entries for ICP and lead scoring give the short definitions.

What does a filled in example look like?

This one is made up, to show the shape. It is not a customer.

A company that sells scheduling software to dental clinics answers: happiest customers are clinics with three to ten chairs, owners who also manage the front desk; the problem is missed appointments; the trigger is opening a second location; the owner signs; the clinic uses paper or a basic calendar; clinics in the same city; not a fit: hospital groups, because they need a long procurement.

Its sheet gives 3 points to dental clinics, 2 to three to ten chairs, 3 to owner titles, 4 to a second location announced, minus 5 to a hospital group. A clinic owner with a new second location scores 12. A hospital group director scores zero or less and is skipped.

How do you test the ICP?

Treat it as a hypothesis.

1. Build two lists that differ on one trait. For example, clinics with a recent second location against clinics without.
2. Send the same message to both. Keep the days and the sender comparable.
3. Count replies and meetings per person messaged. Not meetings in total, because the lists may differ in size.
4. Keep the winner and change one thing.

Our own data is why we recommend this. The audience produced a tenfold range in meetings per person, which no change of wording matched. On the other hand, a reply rate told us little: both audiences replied near 40%. Count meetings, not only replies.

If you do not know where to start, Sellinger can draft an ICP from your website address, and the post on building an ICP from a website URL explains how to read the draft against your own customers. The AI sales plan tool gives a first version without an account.

What goes wrong?

• An ICP written from hope. If none of your five happiest customers match it, rewrite it.
• A sheet nobody uses. If grading takes longer than writing the message, shorten it.
• Too many segments. Start with one. Add a second when the first works.
• No trigger. A list of people who fit but have no reason to move now is a list of polite replies.

How often should you revisit the ICP?

After every hundred meetings, or every quarter, whichever comes first. The answers to the ten questions change as you win customers you did not expect and lose the ones you did. The Sellinger case study is a good example of drift: replies held at about 40% all year, while meetings per person messaged peaked at 2.4% in November 2025 and drifted below 1% from June 2026. A stable reply rate hid a changing audience, and only the meeting count showed it.

Keep a dated copy of each version of the sheet. When results move, you can see which trait changed and which did not.

A checklist

• [ ] I answered the ten questions from my real best customers.
• [ ] The sheet has about seven traits, one of them a trigger, one a blocker.
• [ ] I set contact now, contact later and skip lines.
• [ ] I graded 100 leads by hand and read the result.
• [ ] I plan one test: two lists, one changed trait, same message.
• [ ] I will measure meetings per person messaged, not only replies.