How to define an ideal customer profile for B2B

Define a B2B ideal customer profile from evidence, not ambition. Pull 10 to 20 accounts you won that closed quickly, stayed and grew, and a similar number you lost or churned. Find the traits the winners share and the losers lack, write three to five must-have criteria and two or three disqualifiers, add one "why now" trigger, then turn it into filters and a fit score. Check it against real win rates every quarter and change it when the numbers say so.
This guide gives the steps, the criteria that belong in an ICP, a scoring method, a worked example, what to do with few customers, and the mistakes that make an ICP useless. It draws on what Apollo, Unify and other outbound vendors publish, and adds the parts they skip, such as disqualifiers and a fit score. We make Nagent; how NORA, our outbound research lead, uses a profile is in its own section.
What is an ideal customer profile in B2B?
An ICP describes the kind of company that gets the most value from what you sell and gives you the most value back: it buys quickly, pays well, stays and expands, and does not consume a disproportionate share of support. It is about the account, not the person.
Three terms people mix up:
ICP: the company. Industry, size, region, tools, situation.
Buyer persona: the people inside it. Role, seniority, goals, objections. Apollo draws the same line: company traits for targeting, personas for messaging (Apollo).
Total addressable market: every company that could buy. Your ICP is the slice of it you go after first.
What data do you need before you start?
Closed-won deals from the last 12 to 24 months, with deal size and sales-cycle length.
Retention and expansion for those customers: who renewed, who grew, who churned and why.
Closed-lost and churned accounts with the stated reason.
Support load per customer, if you track it.
Ten minutes with sales, success and product each. They know which customers are easy and which are painful.
If your CRM is thin, start with a spreadsheet of your 20 most important accounts and fill in the columns by hand.
How do you define an ICP, step by step?
Pick your best customers. Unify suggests 10 to 20 closed-won accounts weighted toward those that renewed and expanded, and the same number of closed-lost or churned ones (Unify).
Describe each one on the same fields: industry, employee band, revenue or funding stage, region, key tools, how they bought, why they bought, deal size, time to close, retention.
Find the pattern. Which values appear in most winners and few losers? Those are your criteria.
Keep it short. Three to five firmographic criteria and one or two technographic ones is Unify's guidance; more than that and every account fails some test.
Write the disqualifiers. Two or three traits from your lost and churned list that rule an account out regardless of fit, such as a competing tool on a long contract or a compliance need you cannot meet.
Add a trigger. The event that makes an account worth contacting now: a funding round, a new leader in the buying role, a hiring spree, a tool change.
Write it as a paragraph a new hire could repeat, then as filters a tool can run.
Score and test. Turn it into a fit score, run it against your current customers and your lost deals, and see whether it separates them.
What criteria belong in a B2B ICP?
Firmographic: industry or vertical, employee band, revenue band or funding stage, headquarters region, business model.
Technographic: the CRM, data warehouse or competing tool they run, and what you integrate with.
Situational: growth stage, team structure, a recent change that creates the need.
Operational: how they buy, who signs, how many approvers, typical procurement steps.
Value: expected deal size, likely expansion, support cost.
Disqualifiers: permanent reasons not to sell, kept separate from temporary exclusions such as existing customers or people contacted last month.
Apollo's ICP template covers industry, size, revenue, tech stack, pain points, buyers and triggers. It does not cover disqualifiers or weighting, which are the two parts that stop a list from filling with near misses.
How do you turn an ICP into a fit score?
A simple, transparent method:
Give each criterion a weight that adds up to 100. Put the most weight on the traits that separated winners from losers most clearly.
Score each account by adding the weights of the criteria it meets. Partial credit is fine for ranges, such as an adjacent employee band.
Apply disqualifiers last. Any disqualifier sets the score to zero.
Set tiers. For example: 70 and above is tier A, worth research and personal outreach; 40 to 69 is tier B, worth lighter-touch outreach; below 40 is not contacted. These cut-offs are an example; set yours from your own data.
Back-test. Score last year's won and lost accounts. If won accounts do not score clearly higher, the weights are wrong.
What does a finished ICP look like?
An invented example for a company selling sales-enablement training:
Paragraph: B2B software companies with 200 to 2,000 staff in India and Southeast Asia, selling to other businesses, with a sales team of at least 20 and a head of sales enablement or revenue operations in post.
Criteria and weights: B2B software (30), 200 to 2,000 staff (25), India or Southeast Asia (20), sales team of 20 or more (15), uses a major CRM (10).
Disqualifiers: fewer than 10 sellers; already on a multi-year contract with a direct competitor; requires on-site delivery outside the region.
Trigger: hiring for sales enablement in the last 30 days, or a new head of sales in the last 90.
Persona: head of sales enablement, with the head of sales as economic buyer.
What if you have few or no customers?
Under 20 customers: Unify suggests building a lookalike model from your two or three strongest accounts.
No customers yet: write the ICP as a hypothesis from the problem you solve and who feels it most, then test it with outbound. Track which segments reply and which book meetings, and rewrite after 200 to 300 contacts per segment. That sample size is our suggestion, not a benchmark.
One big logo: do not build an ICP from a single account. One customer is an anecdote.
How do you know your ICP is working, and when should you change it?
Compare ICP-fit accounts with the rest on reply rate, meeting rate, win rate, deal size and retention. Unify's stop rules are a useful start: re-mine the ICP if it has not been revisited in two or more quarters, and audit the disqualifiers immediately if ICP-fit win rate falls below the non-ICP baseline.
Change it when:
Fit accounts do not convert better than non-fit ones.
Sales keeps winning deals outside it.
Your product or pricing changes who it is for.
What mistakes make an ICP useless?
Too broad. "B2B companies with 50 to 5,000 employees" describes a market, not a customer.
Aspirational. Built around logos you want rather than customers you keep.
No disqualifiers, so the list fills with accounts that look right and never buy.
Never tested against won and lost deals.
Living in a slide instead of in the filters your tools run.
Where Nagent fits
We make Nagent. "Nagent is multiplayer AI for growth: people and AI agents in one shared context, delivering the number." NORA is the AI Outbound Research Lead on SERA's Sales team, and she works from the profile you write.
Write it in your words. On the Who you sell to screen you describe your customer in a few sentences. Work out the targeting then reads that description, up to 20 crawled pages of your site and your active products, and returns two groups: what you wrote, and what Nagent concluded, labelled so you can check it.
Scoring you can tune. Segments, keywords, weights and size bands live on the scoring screen, and enabled segments appear in NORA as suggested audiences (Beta).
Fit before contact. NORA scores each company 0 to 100, with industry, size and geography weighing most, and reveals contacts only at 70 or more.
Learning from outcomes. Insights breaks results down by industry, persona, signal and score band, and once enough leads have closed a run tuner proposes changes to the weights.
What it does not do: prospect discovery does not run for consumer audiences, and NORA does not decide your ICP for you; a person writes and corrects it. Work is metered in credits; see pricing.
Frequently Asked Questions
What is the difference between an ICP and a buyer persona?
An ICP describes the company that fits; a persona describes the people inside it who buy. You need both: the ICP picks accounts, personas shape the message.
How many ICPs should a B2B company have?
Usually one primary ICP, sometimes two if you sell distinct products to distinct markets. More than that spreads the team thin.
How often should you update your ICP?
Review it every quarter against win rates and retention, and rewrite it when fit accounts stop converting better than the rest.
How do I use AI to find accounts that match my ICP?
Give the tool your written criteria and disqualifiers, let it score accounts and research the top tier, and review its reasons. See find ICP-fit accounts and research them with AI.
See NORA score accounts against your profile: book a demo.