How to build an ideal customer profile: a template, a negative ICP and a 20-account test

To build an ideal customer profile, score your existing customers by what happened after they signed, write down what your best ones share, add a negative ICP from your lost and churned deals, and test the result on 20 real accounts before you scale it. The six steps below take you from customer data to a one-page ICP template.

An ideal customer profile (ICP) describes the type of company that gets the most value from what you sell and gives the most back in revenue and retention. Gartner, through its TOPO research, defines it as the firmographic, environmental and behavioral attributes of accounts expected to become your most valuable customers.

The examples use a fictional company, Acme Billing, so no real customer data appears. Third-party figures name their source, and the full list is at the end. Sources were last checked on 17 October 2026.

ICP, segment and buyer persona: what each one covers

A useful test: if your ICP cannot become a yes or no check that someone could run on an unfamiliar company in five minutes, it is a description, not a filter.

Step 1: Build your ICP from customer data, scored by what happened after they signed

A closed-won list only shows who said yes. Gartner's account-based framework says to build the ICP from internal data such as closed-won history by account type, account-level profitability and customer satisfaction, plus sales team feedback and, optionally, external data. So record outcomes, not just attributes.

For each customer, note:

• Who they are: industry, size, tools and region.
• What triggered the purchase and who signed.
• Revenue at signing and now, and how many months they have been active.

Then sort each customer into one of five groups: expanded, held, shrank, churned, or too early to tell. Customers newer than your first renewal belong in the last group. SaaS Capital notes that young SaaS companies show inflated gross retention because their customers have not yet had time to churn, and the same logic applies to your newest accounts.

Here are ten fictional Acme Billing customers:

Ranked by logo size, the logistics firm comes first, and it churned. By outcome, the three dental groups went from $900 to $2,250 in monthly revenue (250%), the three agencies from $900 to $600 (67%) and the three largest companies (A, D and H) from $2,700 to $1,500 (56%). The next question is why the dental groups grew.

Three customers per group is too few to trust. Allston Labs' June 2026 guide says you need at least five to seven closed-won customers before a pattern means anything, and 15 to 25 before secondary attributes settle. With fewer, treat the result as a hypothesis and interview your best customers.

The same guide counts customers who expanded three to five times as heavily, with no data behind the multiplier. So re-rank at 1x, 3x and 5x. If your top segment never changes, the multiplier does not matter.

Check logo size against your own outcomes

Published benchmarks do not tell you whether bigger customers stay longer in your product. They compare SaaS vendors by price point, not customers by size. ChartMogul's 2023 report (over 2,100 SaaS businesses, 2022 data) puts top-quartile net revenue retention at 65.1% for vendors averaging under $10 a month per account and 109.3% above $500. SaaS Capital's 2023 survey found median gross revenue retention of about 93% for vendors with annual contract values above $25,000 and 90% below. These are averages across vendors, not facts about your base.

Segments can matter more than size. In one TOPO example, three verticals made up 80% of a client's customers and led on close rate, contract value and churn, yet drew only 5% of its 6,000-plus monthly leads. That is one unnamed client, not a benchmark.

Step 2: Write a negative ICP from lost and churned deals

A negative ICP (or anti-ICP) lists the traits of companies you should not pursue, written from evidence. The IAB's 2019 account-based marketing playbook recommends both a closed-won and a closed-loss analysis. Negative criteria remove the look-alikes that waste time.

1. Pull two lists. Tamtam's blog suggests customers who churned in the last 18 months and your top 20 to 30 "no decision" losses of the past year (its numbers, not a standard). Late-stage losses make a better sample than unqualified leads.
2. Fill in the same attribute columns you used for customers, looking at who they were on day one.
3. Compare frequencies. Suppose 9 of Acme's 12 late-stage losses had nobody in-house who owned billing, against 1 of the 7 active customers (75% versus 14%): a disqualifier candidate. If 8 of the 12 losses and 4 of the 7 customers had 50 to 500 employees (67% versus 57%), that is noise. As Allston says, an attribute common to both groups describes the pipeline, not the outcome. With samples this small, treat a gap as a lead to confirm in interviews, not as proof.
4. Ask the buyers. Clozd, a win-loss vendor, says it compared CRM loss reasons with buyer feedback on thousands of deals and found they matched 15% of the time. It sells buyer interviews, so read that as a reason to check, not as a measure of your CRM. Allston suggests 20-minute calls with five to seven lost buyers, 60 to 90 days after the loss.

Negative-ICP checklist

Separate hard disqualifiers (they stop outreach) from soft ones (they lower priority). One you cannot check from public information belongs in your first-call questions, not the list filter. Tamtam also suggests a required disqualification-reason field in the CRM.

Step 3: Choose firmographic, technographic and situational criteria

• Fit versus timing. Gartner Digital Markets pairs fit data (who matches your ICP) with intent data (who is showing interest). Situational criteria expire, so use them to pick whom to contact first, not who belongs.
• Size by the team that uses the product. Allston's example: a 5,000-employee company with a 12-person team is a 12-person customer.
• Keep four to seven criteria per segment. Allston warns that fewer gives a generic list and more gives a small one that breaks on data errors. Mark each as must, should or nice.
• Technographics are inferred, so a company can run a tool that leaves no trace.

Step 4: Write buyer personas for each segment

For each segment, write one to three roles to write to. Gartner says personas cover daily activities, challenges, goals, role in the decision and motivations. The IAB playbook maps them to roles such as economic buyer, champion and user.

For each role, note the problem in their own words, what they tried first, the likeliest objection and where to reach them. Take the words from won-deal notes, support tickets and calls rather than inventing them.

ICP template for B2B: steps 1 to 4 on one page

Fill in one row set per segment. The example column uses the fictional Acme Billing.

Ideal customer profile examples

Two more one-line ICPs, both illustrative:

• Security training tool: 50 to 250 employees, health or finance, no security hire yet, posted a first compliance role or announced an audit in the last six months.
• Brand design studio: software companies of 20 to 100 employees, funded in the last year, website last redesigned over three years ago.

Step 5: Test the ICP on 20 real accounts before you scale

1. Build the list by running the criteria, not by picking. Take 20 accounts the ICP says fit, from a source other than your CRM. The customers you used to write the rules will always pass.
2. Have two people check independently. Each marks every criterion met, not met or can't tell, from public sources, in about five minutes per account (see how to research a prospect). Note where the answer is "can't tell" or reviewers disagree.
3. Ask of each account: would we be glad to win it? Set the bar before you look, for example 14 of 20.
4. Back-test if you have history. Score 10 retained and 10 churned or lost accounts you did not use to write the criteria. The ICP should rank the retained ones higher.
5. Fix what failed. A criterion that is often "can't tell" is no use as a filter, and one where reviewers disagree needs clearer wording.

By the Wilson method for a 95% confidence interval, 14 of 20 is 70% with a range of about 48% to 85%, so the test cannot tell a 55% ICP from an 80% one. The same limit applies to outreach: 2 replies from 20 emails is 10%, with a range of about 3% to 30%.

Use the test to catch gross errors, not to fine-tune. Then scale in stages (20, 100, then the whole list), re-checking the "glad to win" share each time. A lookalike search from your best customers is a good source for the 100.

Step 6: Review the ICP quarterly and when things change

Sources differ on cadence. A vendor ICP guide says to review it quarterly. Allston recommends re-running the analysis both quarterly and each time your customer base doubles. Gartner's account-based framework adds that target-account list accuracy declines 20% or more each year as company and contact details change. That is data decay, not ICP decay, but it is one more reason to re-check your lists.

A workable rule: a 30-minute check every quarter, and a full rebuild when your customer count doubles or you change price, product or market.

At the quarterly check:

1. Add new customers and new churn to your tables.
2. Did any churned customer pass the ICP and escape the negative ICP? Then a rule is missing.
3. Which wins fell outside the ICP, and why? Which criteria are still often "can't tell"?
4. Write down what changed, with the date.

Does a strong ICP really raise win rates? The 68% claim

Many posts say organizations with a strong ICP have 68% higher account win rates. The original is TOPO's January 2019 Account Based Benchmark Report, summarized in a press release dated 29 January 2019. TOPO surveyed more than 150 practitioners at leading account-based organizations. More than 80% of the most successful ones said they believed they had a strong ICP, against 42% of the others. A TOPO analyst later described the comparison as organizations confident in their ICP versus those that were not.

Pages that repeat the figure credit it to different sources, including SiriusDecisions, RollWorks, HubSpot and Gartner. HubSpot's list repeats it and links the number to the TOPO report without naming TOPO in the text.

Read it as a correlation from a survey that grouped organizations by confidence, not as a measured effect of building an ICP. It is a reason to write one, not a forecast of your win rate.

How FoxKnock helps with steps 3 and 4, and where it does not

FoxKnock drafts steps 3 and 4 from your website. You give it your site address (or a short description) and, optionally, domains of your best customers and of customers you lost. It reads the homepage and up to six linked pages and proposes three to five segments, ranked by a 0 to 100 segment score.

• Criteria and personas. Each segment has criteria (industries, regions, employee range, funding stages, technologies, buying signals, excluded domains) and one to three personas with titles, seniority, pains, goals and objections. All of it is editable.
• Lost customers as negative seeds. In lookalike discovery, companies that resemble a lost customer you added as a negative seed are pushed down and flagged. Results get a 0 to 100 fit score with a reason, evidence and source link per criterion, which gives you something concrete to check in the 20-account test.
• Limits. The draft comes from your website text and the domains you paste, so check it against the customer data from step 1. FoxKnock does not read retention or revenue from a billing system, so step 1 is still yours, and the segment score is a model's ranking, not a measured win rate. It does not re-weight segments from outcomes, so step 6 is manual. FoxKnock launched in October 2026, so there are no customer results to quote.

The AI features use Claude, by Anthropic, so your site's text is sent to Anthropic.

To try it, create a free FoxKnock workspace. The Free plan has 500 one-time credits (they arrive once you verify your email), needs no card and includes AI onboarding from your URL. Compare the draft segments with the ICP you built from your own customers, and see pricing for plan details.