How to define your ICP: build it backwards from your closed deals

Updated 12 November 2026 · Written by Reyansh Juneja and Naman Dhiman, the co-founders of SOVE7GN, who run inbound, midbound and outbound for founders

Define your ideal customer profile by working backwards from the deals you already closed, not forwards from the market you wish you had: list every closed-won and closed-lost deal, write down what was true about the companies that bought fastest and stayed, and keep only the attributes you can find on a list before you contact anyone. An ICP that cannot be turned into a list of named companies is a positioning statement, not an ICP.

What is an ICP, and how is it different from a persona?

An ICP describes the company you should sell to. A persona describes the person inside it. You need both, in that order, because the company decides whether the deal is possible and the person decides whether it happens.

A third thing gets confused with both: the trigger, the event that makes a fitting company start looking this quarter. Fit tells you who to write to. The trigger tells you when. Most lists are built on fit alone, which is why they get answered slowly. The signals worth watching are in buying signals for B2B outbound.

How do I define my ICP when I only have a handful of customers?

With fewer than about 10 closed deals you do not have a pattern, you have anecdotes. That is fine. Build a hypothesis ICP from the deals you do have plus the deals you lost, and treat it as something to disprove in 60 days.

Work in this order:

1. List every deal from the last 12 months in one sheet: company, source, deal size, days from first call to signature, who you talked to, whether they are still a customer.
2. Tag each row won-and-active, won-and-churned, or lost, and add the reason in the buyer's words.
3. For the won-and-active rows only, write the attributes that were true before you contacted them. Before, not after. "They valued our support" is not an attribute. "They had no in-house data engineer" is.
4. Check each attribute against the churned and lost rows. Keep the ones that are common in the first group and rare in the others.
5. Cross out any surviving attribute you cannot find on a public source, a database filter or a signal. Those are nice to know, not usable.
6. Write the profile as a filter, not a paragraph.

Step 5 is where most ICPs die, and it is the step that matters. "Companies with messy data" is true and useless. "Freight brokerages with 50 to 400 staff that have posted a dispatch supervisor role in the last 90 days" can be turned into a list this afternoon.

A worked ICP, built from nine deals

Example only. Harborlane is a made-up B2B software company. The deals, attributes and counts below are invented to show the method, not data from our clients.

Harborlane sells reconciliation software. Nine closed deals in 12 months, five still active, two churned, and twelve losses they tracked.

Harborlane's ICP, written as a filter: US freight brokerages, 50 to 400 employees, two or more offices, no data engineer on staff, currently paying contractors or temps for reconciliation. Their persona: the VP of Operations, measured on invoice accuracy and dispatcher headcount, who has already tried fixing it with spreadsheets.

The two churned deals are the most useful rows on the sheet. Both were single-office and both bought after a funding round rather than after an operational problem. That is how "raised recently" got dropped as an attribute, even though it looked like a buying signal.

How do I test the ICP before I build a list of 2,000 companies?

Three tests, in order. Each takes less than a day.

If the filter returns tens of thousands of companies, it is a segment, not an ICP. If it returns 20, you have described your three favourite customers. Somewhere between a few hundred and a few thousand accounts is the range where a small team can actually reach everyone with a personal message, which is the point of the exercise. The volume math behind that is in how many cold emails per day.

What does an ICP change once it is written?

If the document changes nothing, it was not worth writing. A real ICP changes five things in the same week: the outbound list, what you say in the first line of a message, which inbound leads get a call versus a reply, what the founder posts about, and which feature requests you ignore.

It also changes who you turn down. Writing down the companies you are not selling to this quarter is part of the one-page GTM strategy framework, and it is the line that keeps the list honest when a quarter gets slow.

Running the work that follows, keeping the list current, writing to each person individually, answering every reply the same day, takes more hours than most founder-plus-one teams have. SOVE7GN GTM does that part for B2B startups from seed to Series C: we match every fitting company with the right person, write one personal message each, and handle the replies, while the founder approves posts and takes the calls. Book a call if you want a second read on your profile; pricing is shared on the call. The rest of the guides are in the GTM guides hub.

Questions people ask