Software Growth

ICP (Ideal customer profile)

An ideal customer profile (ICP) describes the customers who get the most value from your product, by the situations, triggers and deal-breakers that predict who buys and stays.

An ideal customer profile (ICP) is a description of the customers who get the most value from your product: they buy quickly, pay well, stay longest and recommend you. The one-line rule: describe them by the situation and behavior that make them need you badly, and name the customers you will turn away, instead of describing them by industry, size and location alone.

An ICP tells you who to market to, who to sell to and what to build next. Without one, every channel, every lead and every feature request looks equally attractive, and a small team spreads itself thin. This guide covers how an ICP differs from a buyer persona and a target market, why situations beat demographics, how to build one from your best customers, the deal-breakers and triggers that belong in it, a worked example, and how to put it to work across marketing, sales and product.

ICP vs buyer persona vs target market

Three side-by-side cards. Target market answers who could buy: everyone with the need who would trust each other's recommendations, for example agencies and freelancers who report marketing results to clients; used for sizing the opportunity, category and channels. Ideal customer profile, highlighted, answers who we should win first: the situations and behaviors of the customers who get the most value, for example agencies that report to 10 or more clients a month, already pay for SEO data and work in English only; used for targeting, qualifying, the roadmap and saying no. Buyer persona answers who we talk to inside: the people in an ICP account and their role in the purchase, for example the account manager who uses the tool and champions it and the agency owner who pays.
Software Growth. Example from the fictional worked example below. Schematic.

These three terms get mixed up because they all describe "who we sell to". They answer different questions:

  • Target market is everyone who could buy. Geoffrey Moore's definition in Crossing the Chasm, which Jason Cohen leans on, adds a useful test: a market is a group with a common set of needs whose members reference each other when they make a buying decision. Size it with your total addressable market.
  • Ideal customer profile is the part of that market you win first and serve best. In B2B it usually describes the account, the company or team that buys. For a solo founder or consumer product, the account and the person are the same.
  • Buyer persona describes the people inside an ICP account: the user, the champion, the person who signs. RevenueGrid's summary of the difference is a good one: the ICP works at the organizational level for choosing accounts, personas guide how you engage the people within them, and you build the ICP first.

April Dunford adds a practical point about personas. When you position a B2B product, agree first on the deal champion, the person who creates the shortlist, pulls in the right colleagues and makes the recommendation to the economic buyer. If your message doesn't land with the champion, you never make the shortlist.

Why situations beat demographics

Most ICP templates start with firmographics: industry, headcount, revenue, country. Jason Cohen argues in Your target market isn't demographic (September 7, 2025) that for most companies these dimensions describe the market wrongly. They include people who will never buy and leave out people who will.

His evidence is his own company. WP Engine's early website looked like a small business selling to small businesses, yet mid-sized and large companies bought the same product. A large bank used it so branches could publish their own content; a large software company ran its developer portal on it. The common thread was never company size. It was people with WordPress sites who valued enterprise-grade speed, security and support at a price far below an enterprise CMS. Cohen's conclusion for founders who notice big logos signing up: don't take it as a sign to go upmarket. Individual teams inside big companies buy for the same reason a small business does, and that reason is your target.

So describe your ideal customer by what predicts buying and staying. Cohen lists more than two dozen dimensions to try, including:

  • Problem ownership: who owns the problem, and is that the same person who owns the solution?
  • Displacing vs first-time: are they replacing a tool or a manual process, or doing this for the first time?
  • Emergency vs deliberate: do they buy because something just broke, or after a planned evaluation?
  • Tools used daily and integrations: what they already run tells you a lot about how they work.
  • Technical acumen, risk tolerance and decision cycle: none of these track company size.
  • The job they are measured on: what number gets them promoted or fired. This is where an ICP meets jobs to be done.

Keep a demographic attribute only when the product genuinely can't be used outside it, such as back-office software for dental practices under one country's law. Then he offers a simple check for whether two kinds of customer belong in the same ICP:

Would customer [B] be swayed by a genuine, effusive testimonial given by [A]?

Jason Cohen / Your target market isn't demographic (September 7, 2025)

If a prospect reads your best customer's testimonial and thinks "that's exactly my situation", they are in. You can test it directly: see whether testimonials make prospects light up on sales calls, or whether adding them to a landing page lifts conversions.

A narrow ICP doesn't shrink your market

The usual fear is that aiming at one type of customer turns everyone else away. Cohen's 2024 essay Selling to Carol argues the opposite. Picture your perfect customer, Carol. She has every problem you solve, needs exactly the advantages you have over competitors, and doesn't care about your weaknesses. Speak only to her, and your message becomes specific enough to be noticed.

A bullseye. The center is Carol, the ICP, who has every problem you solve, values every strength and sees your weaknesses as strengths. The next ring is Diana, who shares the trade-offs: she values your strengths and doesn't mind your weaknesses, roughly ten times as many people as Carol. The outer ring is Eddie, who weighs the trade-offs: roughly a hundred times as many as Diana compare you with alternatives and about one in ten decide you are the best fit, for a total of 20 to 100 times Carol. Outside a dashed line is the anti-market, where a deal-breaker applies and they cannot buy however much they like you.
Rings and multiples from Jason Cohen, Selling to Carol (January 7, 2024), his rough estimates. Schematic.

Other people buy too. Diana values the same strengths and is indifferent to the weaknesses; Cohen estimates there are about 10 times more Dianas than Carols. Eddie weighs the trade-offs like most buyers do, and some share of Eddies pick you because your trade-offs are clear. Cohen puts the effective market at 20 to 100 times Carol. Treat those multiples as a thought model, not data.

His real-world example is Subaru. In the 1990s, with sales falling, Subaru found that lesbians were its strongest buyer segment and advertised to them directly. Cohen reports that Subaru then grew faster than any comparable car company over the following decade, and most of the new buyers were straight. They bought for the same trade-offs: affordable, safe and rugged, at the cost of sporty and sleek.

The economics favor this too. In the workshop built on his book Hidden Multipliers, Cohen notes that at WP Engine, 5% of customers produced 40% of the profit. Finding more customers like them is worth a lot of effort.

What goes in an ICP

A useful ICP has four parts. Cohen's Find Your Carol workshop builds them in order, from an honest list of your strengths and weaknesses:

  1. Keystones: the circumstances that make someone need an extreme version of one of your strengths. Cohen's example: a fast website is a strength; an online store where every tenth of a second of load time changes revenue is a keystone. His rule is strict. If you can't name a real market segment that has the keystone, it is table stakes, and you cut it.
  2. Deal-breakers: the circumstances that disqualify a customer however much they like you. Cohen's example: no SOC 2, no deal. The people they apply to are your anti-market (more below).
  3. Triggers: what Cohen calls inciting events, the moments that turn "maybe someday" into "we need this now" (more below).
  4. Only the facts that matter: industry, size, region or tech stack, if and only if they predict buying or staying in your own data.

Each line should pass the test Cohen sets in his Define Carol step: marketing can target it, sales can qualify against it, and product can build features that thrill the people it describes. Skip invented personal details. Cohen is blunt that a backstory like "Carol has two cats and loves Mahler" is irrelevant and inaccurate.

How to build an ICP from your best customers

You don't have to invent an ICP. If you have paying customers, it is already in your data. Gemma Casey-Swift and Elena Verna describe how the product marketing team at Lovable builds theirs from usage, retention and revenue data, surveys and interviews:

Now, no one has time for an ICP that doesn’t drive value.

Gemma Casey-Swift and Elena Verna / Who's got time for Product Marketing? (April 28, 2026)

A process that works for a small SaaS team:

  1. Pick your best customers. Cohen defines "best" as high retention, upgrades, top NPS scores, unprompted reviews and advocacy, and joining beta tests. In practice: customers who have stayed a year or more, pay above your median and would be upset to lose you. Twenty is plenty.
  2. Pick a comparison group. Take customers who cancelled early, within the first 90 days. A trait that is just as common among them as among your best customers predicts nothing.
  3. Ask the people who already know. Cohen suggests asking sales which phrase from a prospect means the deal will close, asking support who the best customers are, and checking retention data for customers who stay even when you think they shouldn't.
  4. Interview your best customers. Ask about their situation before they bought, what they tried first, what made them start looking and what nearly stopped them. Ask soon after purchase, while they still remember. See customer development for how to run these conversations without leading the witness.
  5. Compare and keep what separates. List candidate traits and count how often each shows up in the best group vs the early-churn group. Keep the ones with a big gap. This is how you avoid what Dunford calls segmentation based on the wishful thinking of which customers we hoped to land.
  6. Write it down and test it. Turn the survivors into one short paragraph plus a list of keystones, deal-breakers and triggers, then use it on the next month of leads and see if it predicts who converts and stays.

Dunford ties this to positioning. Once you know the value only you can deliver, she asks which account characteristics make someone care a lot about that value. Her warning about the opposite case is short:

Companies with weak positioning often have overly broad best-fit customer definitions.

April Dunford / Positioning and Pessimistic Product Thinking (March 14, 2024)

With a broad definition, every lost deal looks like proof the product is weak, when the real problem is that you were chasing accounts you were never going to win. Her positioning book makes the same point from the other side: focus on the best-fit customers who naturally love your product, because they show patterns that the whole customer base hides.

If you haven't launched yet, you can't do this from data. Dunford's advice in January 2026 is to treat pre-launch positioning as a thesis, keep it a little loose, launch, see where the market pulls you, then tighten. The same goes for an ICP: start from a hypothesis, often your beachhead market, and rebuild it from real customers once you have 20 or so.

Deal-breakers and the anti-ICP

A deal-breaker trumps every keystone. In Cohen's Deal-Breakers step, you take each weakness and ask when it stops being annoying and becomes disqualifying, then name the segments that live in those circumstances. That group is your anti-market. The response is deliberate exclusion: don't market to them, don't sell to them, and don't build the features they ask for, however loudly they ask.

Deal-breakers also sharpen the keystones. Cohen's example: "e-commerce stores that need speed" plus the deal-breaker "can't afford premium pricing" becomes "e-commerce stores processing at least $5,000/mo", which is easier to advertise to, easier to qualify against and clearer to build for.

The costly mistake is the big customer who isn't a fit. In Selling to Carol, Cohen describes the account you keep because of the revenue: it fills support with requests, gets custom features, stays unhappy and eventually leaves, having been unprofitable all along. His sign that a team has learned the lesson is turning down what would have been its largest customer because it isn't the ICP.

Triggers: why they buy now

A perfect-fit prospect can agree with everything on a sales call and still say "call back in nine months". Cohen's explanation in the Inciting Events step is that the person with buying power has one to three top priorities, and you aren't one of them yet. Something has to happen first: the site gets hacked, a traffic spike crashes the best day of the year, a consultant says slow pages are why rankings won't move. Every real trigger links to a keystone. The keystone means they could care; the trigger makes it urgent. Mark each trigger as observed (a customer told you) or hypothesized, and note how you would find prospects in that moment: what they search for, where they post, what public signal gives them away.

Rob Snyder makes the same point with his PULL framework for finding product-market fit, discussed on Lenny's Podcast (October 25, 2026): Project, Unavoidable, List of options, Limitations. In his earlier write-up, buyers pull when there is a project on their to-do list that is unavoidable right now, and the options they look into fall short. "Unavoidable right now" is the trigger. Without it, even a perfect-fit account stays in the "someday" pile.

Sales teams call the outside evidence of a trigger a signal. Kyle Poyar's 2026 State of B2B GTM report, from a September 2026 survey, found that teams mostly act on first-party signals: content engagement (52% of respondents), website visits (42%) and product usage (40%). Third-party signals such as champion job changes (28%), hiring (25%) and leadership changes (24%) are used far less.

Signals might be the buzziest buzzword in GTM right now.

Kyle Poyar / 2026 State of B2B GTM report (October 14, 2026)

Poyar notes that nobody agrees which signals matter. An ICP with named triggers answers that for your own business: the signals worth watching are the ones that reveal your triggers.

Worked example: a client-reporting tool

Fictional example. A bootstrapped tool that builds monthly marketing reports for agencies' clients has 140 paying customers and $12,000 in MRR. The founder wants an ICP to decide where to spend the next quarter.

Flow from 140 paying customers to an ICP. The best 20 customers (12 or more months, 99 dollars or more a month, 14% of customers and 30% of MRR) are compared with 30 early churners who cancelled within 90 days. Bar pairs show the share of each group with a trait: reports to 10 or more clients a month, 85% vs 20%, ratio 4.3, a keystone; already pays for an SEO data tool, 75% vs 30%, ratio 2.5, a keystone; agency has 2 to 10 staff, 60% vs 57%, ratio 1.1, no signal; based in the US, 55% vs 53%, ratio 1.0, no signal; needs reports in another language, 0% vs 23%, a deal-breaker. The resulting ICP card: agencies that report results to 10 or more clients every month and already pay for SEO data; deal-breaker, needs reports in a language other than English; trigger, a client questioned the results, from 6 of 8 interviews; left out, team size and country, which are as common in churners.
Software Growth. Fictional example, numbers invented to show the method.

Step 1: the best customers. 20 customers have paid for 12 months or more on a plan of $99 a month or above. They average $180 a month: 20 × $180 = $3,600, or 30% of MRR from 14% of customers (20 ÷ 140). The other 120 average $70: 120 × $70 = $8,400. Total: $12,000.

Step 2: the comparison group. 30 customers cancelled within 90 days over the past year.

Step 3: compare traits. Divide the share in the best group by the share in the early-churn group:

  • Sends reports to 10+ clients a month: 17 of 20 (85%) vs 6 of 30 (20%). 85 ÷ 20 = 4.3×. Keystone.
  • Already pays for an SEO data tool: 15 of 20 (75%) vs 9 of 30 (30%). 75 ÷ 30 = 2.5×. Keystone.
  • Agency of 2–10 staff: 12 of 20 (60%) vs 17 of 30 (57%). About 1.1×. No signal.
  • Based in the US: 11 of 20 (55%) vs 16 of 30 (53%). About 1.0×. No signal.
  • Needs reports in a language other than English: 0 of 20 vs 7 of 30 (23%). The tool only produces English reports, so this is a deal-breaker.

The two firmographic traits the founder would have written first, team size and country, are as common among churners as among the best customers. They stay out of the ICP.

Step 4: interviews. The founder talks to 8 of the best 20. Six started looking within weeks of a client questioning the results in a meeting; two when they hired their first account manager. The first is an observed trigger; the second is worth testing.

The ICP: agencies that report results to 10 or more clients every month and already pay for SEO data, usually right after a client has questioned the numbers. Not for agencies that need reports in other languages.

Putting it to work. The signup form gets two questions: how many clients do you report to each month, and which SEO tools do you pay for? Next month brings 200 trials, and 50 match both keystones. Say ICP trials convert at 30% and the rest at 6%: 50 × 30% = 15 and 150 × 6% = 9, for 24 new customers. A quarter of the trials (50 ÷ 200) produce 63% of the new customers (15 ÷ 24). The founder makes onboarding calls to those 50 instead of all 200, 150 fewer calls, and writes the homepage for the agency owner whose client just asked "are these results real?"

How to use your ICP

An ICP that lives in a slide deck does nothing. It should be the first input to your go-to-market strategy and show up in daily decisions. Verna and Casey-Swift's point is that naming the ICP is the easy part; the work is getting the rest of the company to act on it.

Marketing

  • Write the homepage and ads for Carol. Cohen's test: if your ad can't grab your perfect customer's attention in a few seconds, it won't grab anyone's.
  • Lead with the trigger. Cohen notes that in advertising the inciting event often works better than the keystone because it speaks to what the customer feels right now.
  • Pick channels where your best customers came from, and use their words in your copy and value proposition.

Sales

  • Turn keystones and deal-breakers into lead qualification questions and lead scoring rules. A deal-breaker should disqualify a lead outright, whatever its score.
  • For account-based marketing or outbound, build the target list from triggers, not just from firmographic filters.
  • Track win rate separately for ICP and non-ICP deals. If they look the same, your ICP isn't sharp enough.

Product

  • Tag customers as ICP or not at signup, then segment activation, trial conversion, feature usage and churn by that tag. Cohen's rule: Carol's cancellations are the signal to investigate; customers outside the bullseye are noise.
  • Prioritize requests from ICP customers. Feature requests from the anti-market go in the "no" pile, even from large accounts.
  • Check that ICP customers reach value quickly in onboarding. If even Carol struggles, the product or the ICP needs to change.

Cohen makes the payoff explicit in his January 2026 essay Creating space: giving up on winning everybody frees the time to win the customers who fit. For a small team, the hours not spent on poorly fitting prospects go to customers the product actually serves.

What changed in 2026

  • ICP reviews moved from yearly to continuous. Casey-Swift and Verna write that Lovable reviews which segments are growing, converting, expanding and churning every week, because an ICP set in January and left to calcify can't keep up. They also found something many founders will relate to: for a general-purpose product, demand showed up from solo builders to large enterprises at once, so their ICP is genuinely broad. Their answer was more data and faster reviews.
  • Your user may be an agent. Verna asked in April 2026 whether MCP belongs in your ICP: as AI agents use products on people's behalf, the "user" may never touch your interface. If agents choose or operate tools in your category, describe what they need too.
  • Positioning shifts with model releases. Dunford wrote in September 2026 about AI founders who fear their positioning will be out of date in two weeks. Her advice: re-run the steps, because new capabilities usually strengthen an existing value theme. Only genuinely new value means starting over, and with it, rechecking who cares most about that value.
  • The method became tooling. Cohen published the steps from his book Hidden Multipliers as free AI skills (updated August 2026) that walk a team from observations to strengths, keystones, deal-breakers, triggers and a final definition. Poyar's report shows teams pairing an ICP with signals and AI agents to decide who to contact and when.

Common mistakes

  • Using only demographics. "B2B companies with 50–500 employees in North America" fits thousands of companies that will never buy. Add the situation that makes someone need you.
  • Writing the ICP you wish you had. Dunford's "wishful thinking" warning applies. Start from customers who already love you, not the logos you hope to land.
  • Keeping traits that don't separate. If a trait is as common among churners as among your best customers, it doesn't belong.
  • Averaging two segments. When your keystones point to two different customers, Cohen's Define Carol step forces you to pick a primary one. A profile halfway between two segments describes nobody.
  • Letting a big anti-ICP account steer the roadmap. High revenue and constant custom requests from a poor fit usually end in churn after a lot of wasted work.
  • Confusing ICP and persona. An ICP picks accounts; personas help you talk to the people in them. Do the ICP first.
  • Tightening too early. Before launch, an ICP is a hypothesis. Hold it loosely until real customers prove or disprove it, then revise it as the product and market change.
  • Read: Selling to Carol: Why targeting an ICP brings 10x more customers than you expected, Jason Cohen (January 7, 2024). The bullseye argument and the Subaru story in full.
  • Read: Your target market isn't demographic, Jason Cohen (September 7, 2025). The WP Engine story, the testimonial test and a long table of non-demographic dimensions to try.
  • Read: Find Your Carol, Jason Cohen. A free six-step workshop, also usable as AI skills, that takes you from honest observations to keystones, deal-breakers, triggers and a written ICP.
  • Watch: An 86-minute guide to finding PMF, Rob Snyder on Lenny's Podcast (October 25, 2026). The PULL framework from 01:13, selling to a person rather than a company from 16:10, and what churn teaches you about your ideal customer from 1:09:28.
  • Book: Obviously Awesome, April Dunford (updated and expanded edition, 2026). Her positioning process, which starts from your best-fit customers and the value only you deliver.
  • Read: 2026 State of B2B GTM report, Kyle Poyar (October 14, 2026). Which buying signals teams act on, and how go-to-market motions change with deal size.

Sources

Back to the SaaS glossary