ICP in sales: what it means, and how to build yours from the deals you already won

What an ICP in sales actually means

ICP in sales stands for ideal customer profile: a description of the type of company that gets the most value from what you sell and is therefore the cheapest to win and the most likely to stay. It describes an organisation, not a person, and it is written in attributes you can filter on: industry, size, country, technology, growth signals. The ICP meaning that matters in a sales context is operational rather than aspirational, because its whole purpose is to be turned into a list.

The canonical definition and its variations live in our glossary entry on the ideal customer profile. This page does the other half of the job: building yours from the deals you have already won, and showing what that costs.

Why bother at all, which is the fair question a sceptical founder asks. Because every hour of selling is spent somewhere, and without a written profile it gets spent on whoever answered. Teams that write an ICP down are not smarter about their market; they are just able to say no faster, and to say it before the discovery call rather than after the third one.

Two failure modes make an ICP useless, and both are common. The first is writing it from opinion: a workshop produces a profile everyone likes and nobody can filter on. The second is writing it once: markets move, your product moves, and a profile from two years ago describes customers you no longer win. The method below fixes both by deriving the profile from data and by making it cheap enough to redo every quarter.

ICP in sales vs buyer persona vs target market

An ICP describes a company, a buyer persona describes a person inside it, and a target market describes the whole category. These three get used as synonyms in the same meeting, and the confusion is why so many profiles are unusable. They answer different questions and they are built from different material.

The practical order is target market, then ICP, then persona. The ICP tells you which doors to knock on; the persona tells you what to say when someone opens. Skipping the middle step is how teams end up with excellent messaging pointed at companies that were never going to buy.

One more distinction worth keeping straight: an ICP is not a list of your biggest customers. Biggest and best are different. A large account that took eleven months, three discounts and a custom integration is not ideal, however good the logo looks on the website.

The attributes of an ideal customer profile, and where each one comes from

A profile is only as good as the attributes you can actually check on a company you have never met. Here is the working set, with the source of each one, because an attribute with no source is a wish.

Keep the profile to five or six attributes. A profile with fifteen criteria matches nothing, which feels rigorous and produces an empty list. A profile with two matches everything, which produces a list you cannot prioritise. The test is whether the filtered list is big enough to work for a quarter and small enough to research one by one.

Note which lines are filterable before you commit to them. Firmographics, technographics and signals can be applied to a market you have not touched yet. Behavioural attributes cannot: they only exist for accounts already in your funnel, so they belong in your scoring model rather than in the profile you prospect from.

How to build your ICP in sales from won deals, in five steps

Five steps: rank your won deals on the numbers you already have, enrich the best five, write what they share as filterable criteria, generate lookalike accounts, then score new accounts instead of filtering them. The method starts from evidence, not from a workshop. It takes an afternoon the first time and about twenty minutes every quarter after that.

Step one: pick the right winners, not the biggest ones. Take your closed-won accounts from the last twelve months and rank them on four numbers you already have: annual contract value, time from first touch to signature, retention or renewal, and how much support they consumed. The accounts that score well on all four are your evidence. Five is enough to start, ten is comfortable. If a large account took twice as long and churned at renewal, it does not belong in the sample, whatever it contributed to the quarter.

Step two: enrich them and read the attributes side by side. Put the five companies in a sheet and enrich them, so that industry, headcount, country, founding year and technology sit in columns rather than in your memory. This is where the pattern shows up, and it is usually not the one the team expected. A frequent surprise: the common attribute is not the industry at all but the structure, for example companies with a sales team between five and fifteen people, whatever they sell.

Step three: write the profile in filterable language. Turn the pattern into five or six criteria a stranger could apply. Not "ambitious mid-market companies" but "software companies, 50 to 500 employees, in the United States or Western Europe, with a sales team of at least five, hiring for a revenue role". If you cannot filter on it, it is a note for the persona, not a line in the profile.

Step four: produce the lookalike list. A profile that stays in a document is a document. Turn it into accounts: take your best-fit customers as seeds and generate companies that resemble them, then filter that output on the criteria from step three. The next section shows exactly what this costs, because we ran it. The same two actions run in a spreadsheet, from an AI assistant over the Derrick MCP server, or from the REST API when the profile has to be recomputed on a schedule.

Step five: score, do not sort. Give each criterion a weight and score the list rather than filtering it down to a rigid yes or no. A company that matches four criteria out of five and is hiring two sales people deserves a call before a perfect match that shows no movement. Scoring keeps the near misses visible; filtering deletes them silently.

What step four costs: the run we did on 18 September 2026

Building the profile from data cost us 21 credits on 18 September 2026: 1 credit to enrich the seed account, then 1 credit per lookalike company returned, for 20 named accounts. Every article on this subject tells you to analyse your best customers, and none says what the analysis costs or shows the output, so here it is, measured and not estimated.

We took one seed account, a software company in San Francisco founded in 2015 with just over two thousand employees, and ran two actions on it:

• Enrich Companies on the seed: 1 credit. It returned the attributes that make up a profile, industry, headcount, country, city, founding year, and the identifier needed for the next step.
• Find Similar Companies from that seed: 1 credit per company returned. We asked for a first batch and got 20 companies back, so 20 credits, out of 100 matches found in total.

Total for the run: 21 credits for 20 named lookalike accounts, each with website, LinkedIn URL, industry and country, and a city for 19 of the 20. The batch size sets the cost, since you pay per company returned: we stopped at 20 of the 100 matches found. Scaled to the method above, five seed accounts and twenty-five lookalikes is 30 credits if you already hold the LinkedIn URLs of your customers, and 35 if you have to look those companies up first. Either way it fits inside the free plan of 100 credits a month. That figure assumes your LinkedIn account is connected through the Chrome extension: company enrichment needs that connection to run at all. Find Similar Companies does not, and costs 1 credit per company either way.

The output was coherent, not random: from a revenue intelligence seed, the list came back full of revenue and sales data companies, spread across the United States, Europe, Australia, India, Tunisia and Brazil. We ran it without the optional country filter, which is why the list spans several markets; that filter exists if you want the output pinned to one country from the start. Everything else is filtered on your own criteria afterwards, and that filtering costs nothing.

You can look at the real file rather than take our word for it. Download the lookalike accounts as a CSV, no sign-up. It contains companies and public business information only: no personal emails, no phone numbers, no named individuals, and the file says so on its first line. Two of the twenty are left out of the file because the brand no longer trades under that name, which is itself worth knowing: a lookalike list reflects the history in the data, so a human still reads it before anyone gets contacted.

Three ICP examples, written the way a sales team can use them

Abstract advice about the ideal customer profile is easy to agree with and impossible to apply. Here are three written out, in the filterable language of step three. They are illustrative rather than borrowed from a real company's internal documents.

Example one, a sales enablement tool selling to mid-market software companies. Software companies, 50 to 500 employees, United States or Western Europe, sales team of at least eight people, currently hiring at least one revenue role, already running a CRM. Disqualifiers: agencies, companies under 20 employees, anyone without a named sales leader.

Example two, a compliance product selling to regulated French mid-caps. Companies registered in France, 250 to 2,000 employees, in financial services, health or industry, with at least two sites, present in the public company registry with a filed annual account. Disqualifiers: pure holdings, subsidiaries of groups whose compliance is handled at head office.

Example three, an agency selling outbound as a service. B2B companies, 20 to 200 employees, that raised funding in the last 18 months, with fewer than three people in sales and no dedicated SDR, selling a product above a five-figure annual contract value. Disqualifiers: consumer businesses, anyone whose average deal is too small to pay for the service.

Notice what all three have in common. Every line can be checked from outside: headcount, country, industry, registry presence, hiring activity, funding. And each one ends with disqualifiers, which are the half most profiles skip and the half that saves the most time, because knowing who to drop is faster than ranking who to keep.

An ideal customer profile template you can copy

Copy this into a document and fill it in. It fits on one page on purpose: a profile that needs two pages is a strategy paper, and nobody filters a list with a strategy paper.

The two lines people leave blank are evidence and review date, and they are the two that make the difference between a profile and an opinion. A profile with no evidence cannot be argued with; a profile with no review date will still be quoted in a year, long after it stopped describing anyone you win.

ICP drift: why the profile expires, and how to re-derive it

Re-derive the profile once a quarter: it is a twenty-minute job, not a new workshop. Profiles go stale quietly. The market moves, your product gains a feature that opens a new segment, a competitor exits and their customers become winnable. None of that changes the document on the shared drive, so the team keeps prospecting a profile derived from deals closed two years ago.

Three symptoms tell you a profile has drifted, and they show up before the pipeline does. Win rates fall inside the profile while staying flat outside it. Sales people start making exceptions, and the exceptions close. The accounts your team is most excited about keep failing one criterion, always the same one.

The fix is a recurring twenty-minute job rather than a new workshop. Once a quarter, take the deals closed since the last review, run steps one to three again, and compare the attributes to the current document. If the pattern has moved, the profile moves with it; if it has not, you have spent twenty minutes buying confidence. Because the lookalike step costs a credit per account, re-deriving the list is cheap enough to do on schedule rather than in a crisis.

The related idea worth naming is the expansion profile. Sometimes the analysis turns up a second cluster that wins well but does not match the main profile at all, for example smaller companies in a different industry that buy faster. That is not drift, it is a second profile, and it deserves its own list and its own messaging rather than being averaged into the first one.

Using your ICP in sales: scoring, prioritising, and ABM

The profile shows up in three places once it is written: qualification, prioritisation and account based marketing. A written profile earns nothing until it changes what someone does on Monday morning.

Qualification. The profile becomes a fit score attached to every account, computed from the criteria and their weights. Discovery calls then start from a number rather than from a feeling, and a rep can say no to a poor fit without arguing about taste.

Prioritisation. Sort the pipeline by fit multiplied by signal. A high-fit account showing no movement goes into nurture; a medium-fit account that just hired three sales people goes to the top of the call list today. Signals decay in weeks, so this ordering has to be recomputed, not set once.

Account based marketing. The profile is what turns a market into a named account list, which is the precondition for any ABM programme. Our guide on account based marketing covers the programme itself, and LinkedIn ABM covers the channel. Once the list exists, building the prospect list is the mechanical step, and selling to decision makers covers who to reach inside each account.

One discipline holds all three together: the profile has to be attached to the account record, not stored in a slide. If the fit score is not visible in the tool where the work happens, it will be ignored within a month, however good the analysis behind it was.

Where the ideal customer profile fits with everything else

The profile is one artefact in a chain, and it is worth being clear about what sits on either side of it so that it does not get asked to do a job it cannot do.

Upstream sits your go to market choice: which market, which motion, which price point. Our guide on go to market strategy covers that layer. An ideal customer profile does not choose your market; it sharpens the one you have chosen.

Downstream sit the list, the people and the message. The profile produces the account list; finding the right people inside each account and reaching them is a separate discipline, covered in our lead generation guide and in outbound sales. A perfect profile with a generic message still fails, just for a different reason.

We ran this on our own users and wrote up what came out in the ICP we derived from our best accounts. And in the middle sits the thing this page argues for: deriving the profile from data you already own, cheaply enough to redo it every quarter. Five enriched customers and twenty-five lookalike accounts cost 30 credits, which the free plan covers. The barrier to a data-derived profile is no longer budget, it is the decision to look at the deals you already won instead of the market you wish you had.