What is an ideal customer profile (ICP) in sales and marketing?

This guide is for founders, marketing leaders and sales leaders deciding where to focus. It explains what an ICP is, how it relates to personas and market size, how to choose between segments with real numbers, and how to turn the result into accounts. When you're ready to write yours down, use the ICP template: it's the worksheet, with two completed examples. This page helps you make the decisions that go into it.

What ICP means in sales and marketing

ICP stands for ideal customer profile. It's a description of a type of company, not a person, and "ideal" means the best two-way fit: companies that get the most value from what you sell and are the most valuable to you in return. They buy at a reasonable cost, stay, and don't need rescuing every month. The biggest logo in your market isn't automatically ideal.

A useful ICP has five parts:

• Firmographics: industry, size, location and business model, the facts you can filter a list on.
• Operating conditions: what's true about how the company works when you fit, such as running several sites, meeting a regulation or having outgrown a spreadsheet.
• Triggers: events that start a buying process, such as a new location, a renewal date or a new leader.
• Buying group: the roles that own the problem and sign off.
• Disqualifiers: traits that rule a company out even when everything else matches.

Here's what one sounds like in a sentence: commercial electrical, mechanical and roofing subcontractors in Texas with 20 to 150 staff that work for general contractors who require proof of insurance, and that start shopping when they bid larger jobs. That's a commercial insurance broker's ICP, and it's the one we build in the worked example below.

ICP, buyer persona, target market and TAM

These four terms get used interchangeably, and they shouldn't be. Each answers a different question:

They nest: the TAM contains the target market, which contains the ICP, which produces the account list. Personas cut across all of them, because the same owner-operator might run a roofing firm or a restaurant.

You can size the outer layers with public data. In the US, the Census Bureau's County Business Patterns reports establishments with paid employees by industry and employment size. It counts establishments (individual locations), so a company with five sites appears five times. In the UK, the Office for National Statistics publishes counts of VAT- or PAYE-registered businesses by industry and region. Use them to check whether your ICP leaves enough companies to sustain your plan, not as a precise count.

Why an ICP matters to each team

An ICP pays off when every team uses the same definition. Here's what it decides for each:

"ICP marketing" means running marketing at the account level. In practice, that's three shifts: build target lists from ICP criteria rather than job titles alone, keep poor-fit companies out of campaigns where you can, and report the share of pipeline and wins coming from ICP accounts instead of raw lead counts. A thousand leads from companies that will never buy is a cost, not a result.

How to build an ICP from your best and worst customers

1. Define "best" with numbers before you look at names. For example: renewed at least once, profitable after service costs, closed in a normal cycle, would act as a reference. Agree the definition first so the list isn't a popularity contest.
2. List 10 to 20 best customers and 5 to 10 worst. Worst means churned, stalled, unprofitable or constantly escalating.
3. Put their traits side by side. Industry, size, location, how they operate, what triggered the purchase, who signed and who used it.
4. Find the separators. You're looking for traits common among the best and rare among the worst. Industry and size often matter less than operating conditions and triggers.
5. Write one sentence, then the detail: firmographics, conditions, triggers, buying group.
6. Add disqualifiers from the worst customers and from deals you lost for the same reason more than once.
7. Mark what's evidence and what's assumption, then test the profile against your closed deals before you build lists from it. The ICP template walks through that validation with worked arithmetic.

If you don't have many customers yet, start from the offer: the problem you solve, who feels it most sharply and what event makes it urgent. That's a set of assumptions to test with your first 20 to 30 conversations.

Most sellers don't start with a written ICP at all. Blackpearl Group, which makes Bebop, analyzed 59,881 opening queries submitted to its prospecting app between February 2025 and May 2026 for its GTM-Bench research. The largest group, 43.8%, were domain-based searches such as "find customers for xyz.com", and 17.6% were generic lead searches. Only 6.2% described an offer and asked for help working out who would buy it. The authors note the app's input format biases the sample toward the first two types, so read this as how people start on one product, not a market survey. Still, the lesson holds: a website says what you sell. The ICP is the step that says who buys it and why, and skipping it is how lists fill up with companies that look right and never buy.

Worked segmentation example: a commercial insurance broker

Illustrative example. The company and figures are fictional, to show the method.

Fairhaven Risk Partners (fairhavenrisk.example) is a commercial insurance broker in Dallas with 220 business clients. Its producers quote anything that comes in. Leadership wants to know where to focus outbound effort, so it pulls 24 months of quotes and the renewal history for each segment:

Win rate alone says restaurants are as good as construction (both 40%). Commission alone says trucking is best ($14,000). Neither tells you where an hour of prospecting is worth the most. Combine them into five-year value per quote = win rate × annual commission × five-year retention factor.

The retention factor is the expected number of years of commission in the first five. A client who renews 92% of the time is there in year one, then in year two with a probability of 0.92, year three 0.85, year four 0.78 and year five 0.72. Those add up to about 4.26 years. At 90% the factor is 4.10, at 88% it's 3.94, at 74% it's 2.99 and at 70% it's 2.77.

Then Fairhaven looks inside the winning segment. Of its 25 most valuable construction clients, 22 do commercial work for general contractors who require certificates of insurance. Of the 9 construction clients that left, 7 were residential remodelers. That split becomes an operating condition (works for GCs on commercial jobs) and a disqualifier (residential only). The final ICP is the sentence from the top of this page.

What made this work:

• Combining win rate, value and retention changed the answer. Restaurants were the easiest to win and among the least valuable to pursue.
• The segment table found the industry; looking inside the segment found the operating condition that separates good clients from poor ones.
• The decisions aren't all "yes" or "no". Restaurants still get served when they call; they just don't get outbound effort.
• Fairhaven treated trucking's numbers with caution: 9 wins is a small sample. "Exclude for now" means revisit with more data, not never.

How to keep your ICP current

An ICP drifts as your product, pricing and market change. Watch three numbers every month or quarter: the share of new pipeline from ICP accounts, win rate inside versus outside the profile, and 12-month retention inside versus outside. Then act on what they tell you:

Give the ICP an owner, a version number and a date, so people know which one they're using. A full rebuild once a year, plus a review whenever a signal above fires, is enough for most teams. For more warning signs, read why your ICP is probably outdated.

Turning the ICP into a list of accounts

1. Translate criteria into filters. Industry codes (NAICS in the US, the SIC codes companies give Companies House in the UK), headcount bands and geography go straight into a database search.
2. Separate filters from signals. Must-haves filter the list; triggers rank it; anything you can only learn by reading the company's website or job ads becomes research.
3. Count what's left. If your team can work 300 accounts in depth this quarter and the filtered list has 40, loosen a criterion. If it has 40,000, add a must-have or require a recent trigger.
4. Tier by trigger. Tier 1: fits and had a trigger in the last 90 days. Tier 2: fits, no trigger yet. Tier 3: partial fit, monitor only.
5. Check a sample before you load it. Read 50 records and count how many genuinely fit. A longer list isn't a better one: Blackpearl Group's GTM-Bench benchmark subtracts a point for every weak record, and agents that returned thousands of loosely matched rows repeatedly ended up with negative scores.
6. Research tier 1 before contacting it, using the account profiling template.

A plain-language ICP sentence is also the input Bebop works from: it finds matching accounts and contacts and builds a playbook for each. Check its first results against your disqualifiers, as you would any list.

Is your ICP doing its job?

Run these five tests. If yours fails two or more, it needs work before it drives any spending:

• Two people classify the same 20 accounts and agree on at least 18. If not, a criterion is too vague.
• It excludes something. An ICP with no disqualifiers covers every company in the industry.
• The numbers back it. Win rate and retention are clearly higher inside the profile than outside.
• The list is workable. It leaves enough accounts for your plan, and not so many that research becomes impossible.
• Sales and marketing use the same version. If marketing's lead scoring and sales' target list use different definitions, you have two ICPs and neither is working.

Next step: write yours down with the ideal customer profile template, which has the worksheet, two completed examples and the validation arithmetic.