ICP Scoring: How to Build a 0–100 Ideal Customer Profile Rubric

What an ideal customer profile actually is

An ideal customer profile is a written definition of the company you should be selling to. Not the company you can sell to — the one that produces your best outcomes on the metrics you actually care about: win rate, deal size, time to value, retention, expansion, and cost to serve.

The words matter here because three terms get used interchangeably and shouldn't be. An ICP describes an organisation. A buyer persona describes a role inside that organisation. A buying committee is the specific group of people at one account who will collectively approve or kill the deal. You need all three, in that order, and each one requires a different kind of data.

“Ideal client profile” is the same thing

Services businesses — agencies, consultancies, law firms, recruiters — usually say ideal client profile rather than ideal customer profile. It is the same document with the same structure; only the noun changes to match how the industry talks about the people it bills. Everything in this guide applies unchanged, and Example B further down is written for exactly that model.

The B2C equivalent is a different animal. A consumer profile is built on demographics, household composition, purchase frequency and channel behaviour rather than on firmographics and buying committees, and the reachability problem barely exists because you usually hold the contact details already. If you are selling to consumers, most of the mechanics here will not transfer.

That last clause is the test most ICPs fail. “Mid-market SaaS companies with a modern data stack” is not an ICP; it is a mood. “B2B software companies, 200–1,000 employees, headquartered in the UK, Germany or the Netherlands, running Salesforce, with an outbound SDR team of five or more” is an ICP, because it is a query. If you cannot turn your ICP into a filter that returns a countable list of companies, you do not yet have one.

Why the ICP is the highest-leverage document in go-to-market

Every downstream decision inherits its assumptions. Your list-building queries, your ad targeting, your lead scoring model, your routing rules, your sales-capacity plan, your pricing tiers, and your product roadmap all point back at a definition of who you serve. Get the definition wrong by 20% and every one of those systems is wrong by 20%, quietly, for as long as the definition stands.

The cost shows up as symptoms rather than a single failure: sales complaining about lead quality, marketing complaining about follow-up, a CAC payback period that keeps sliding, a churn cohort nobody can explain. Those are usually ICP problems wearing a disguise.

ICP vs buyer persona vs TAM vs segment

These four terms describe four different-sized things, nested inside each other. Confusing them is the most common reason two teams disagree about targeting while both believing they are following the ICP.

The bottom row is where go-to-market plans stop being theoretical. Gartner's research on the B2B buying journey puts the typical buying group for a complex solution at six to ten decision-makers, each arriving with independently gathered information. Forrester's 2024 State of Business Buying report puts the average higher still, around 13 stakeholders, with the large majority of purchases crossing two or more departments.

Translate that into an operational requirement and it is uncomfortable: if your ICP contains 2,000 accounts and each one needs six to ten reachable people, you are not sourcing 2,000 contacts. You are sourcing 12,000 to 20,000 — and every one of them needs a verified email or a mobile number attached, or that account is nominally “in the ICP” and practically unworkable.

The six dimensions of a B2B ICP

Most ICP frameworks stop at two dimensions, firmographic and technographic, because those are the two that data vendors have historically sold. A profile that predicts revenue needs six. The last one is the one nobody writes down.

Dimensions one and two are available from most company data providers — our guide to firmographic data covers how those fields are sourced and where they degrade. Dimensions three to five come from your own CRM and finance systems. Dimension six is the one that decides whether the other five ever leave the spreadsheet, and it is measurable: for any list of target accounts, you can count what percentage of the required roles you can actually contact today.

Where each dimension's data actually comes from

The most common reason an ICP stalls between definition and execution is that nobody checked whether the attributes were obtainable at list scale. This is what each dimension typically requires.

How to build an ICP in seven steps

This is a data exercise, not a workshop. If your team has more than roughly 20 closed-won accounts, the profile is already in your CRM — the work is extracting it, not imagining it. If you have fewer than 20, you are building a hypothesis, and you should label it as one and set a date to revisit it.

ICP scoring: building a 0–100 rubric

A binary ICP — in or out — wastes the middle of your market. A scored ICP lets you spend expensive effort on the accounts that deserve it and cheap effort on the rest, which is the entire point of segmentation.

The rubric below is an ICP builder you can use immediately: six weighted criteria totalling 100 points. Rate each criterion from 0 to 5 for a given account; the score is the weighted sum. Adjust the weights to your own economics — a product with heavy implementation should weight operational fit higher; a self-serve product should weight it near zero.

Reading the tiers

The knockout list sits outside the score

Run knockouts first, as a filter, before anything is scored. An account that fails a knockout gets no score at all — not a low one. This matters because a high-scoring account with a disqualifying attribute will otherwise keep surfacing at the top of your lists and quietly consuming your best rep's week.

• Jurisdiction: a country where you cannot lawfully process the data, deliver the service, or contact people on the channel your motion depends on
• Scale floor or ceiling: below the headcount where your pricing produces a viable deal, or above the size where your implementation model collapses
• Contractual lock-in: a multi-year commitment to a directly competing platform that has just been renewed
• Prohibited industries: sectors excluded by your own policy, your insurer or your customers' contracts
• Existing relationship conflict: a partner's account, a current customer's direct competitor where you have an exclusivity clause

Write the negative ICP too

The negative ICP — sometimes called the anti-profile — is the description of accounts you have learned not to sell to. It is not the same as the knockout list. Knockouts are structural and known in advance. The negative ICP is earned: it is assembled from accounts you did sell to and should not have.

Build it from three cohorts in your own data: accounts that churned inside the first year, accounts that consumed disproportionate support or implementation effort, and deals that closed after an unusually long cycle at an unusually deep discount. The shared attributes of those three groups are your negative ICP. Write it down in the same document, because it is the section that protects your team from a quarter-end decision that feels good in March and looks expensive in September.

Three worked ICP examples

Abstract frameworks are easy to agree with and hard to use. Here are three profiles written to the standard described above — specific enough to be run as a query. They are illustrative examples, not benchmarks: use the structure, replace the values with your own.

Notice what all three have in common: each names the committee, and each includes at least one knockout tied to whether the people can actually be reached. That is not padding. It is the difference between a list your team works and a list your team quietly abandons in week three.

The reachability layer most ICPs ignore

Here is the failure that no ICP framework warns you about, because the frameworks are written by people who define profiles and not by people who work the resulting lists.

You define the profile against attributes you can buy: industry, headcount, tech stack, funding. Those attributes describe companies, and company data is comparatively cheap, stable and complete. Then you hand the list to a team whose job is to contact people, and people data is none of those things. Job changes churn through it constantly, direct mobile numbers are the hardest field in B2B data to source and verify, and coverage is wildly uneven across seniority, country and company size.

The result is an ICP that is internally consistent and externally unworkable. The account qualifies. The committee is mapped. And for four of the seven roles you need, you have a generic info@ address and a switchboard number.

What this changes about how you write an ICP

Once reachability is a measured dimension rather than an assumption, three things follow.

Segments get graded by motion, not just by fit. A segment with strong fit and weak mobile coverage is not a bad segment — it is an email-and-marketing segment. Assigning callers to it is a capacity mistake, not a targeting one.

Coverage becomes a procurement question. If the segment you want most is the one your current provider covers worst, the answer is a second source rather than a smaller ambition. That is the argument for waterfall enrichment: query providers in sequence until one returns a verified match, rather than accepting a single vendor's blind spots as your market's boundary.

“Match rate” stops being one number. A single global match-rate figure hides exactly the variation you need to see. What matters is match rate for your ICP — by country, seniority and company size. Our guide on what match rate to expect from a contact enrichment API covers how to test that properly, and the 2026 accuracy benchmark shows how far the figures move between segments.

What changes when an agent does the scoring

Two parts of ICP work are now routinely handled by AI agents rather than by people with spreadsheets, and it is worth being precise about which two, because the difference matters.

Judgement-heavy classification is a good fit. Reading a company's website, its job postings and its recent announcements to decide whether it matches a written profile is exactly the kind of fuzzy pattern-matching that a language model does well and that a SQL filter does badly. Agents can score a list of 5,000 companies against a prose ICP definition in a way that firmographic filters cannot.

Fact retrieval is not. Asking a model to recall a specific person's email address or mobile number produces confident, plausible, wrong answers. That step has to be a tool call against a live data source with verification attached. This is what the Model Context Protocol is for — the agent handles the reasoning, the enrichment tool returns the verified record. If you are building this, see how to enrich contacts inside Claude, ChatGPT and Cursor and the MCP server comparison for B2B enrichment.

The practical shape of an agentic ICP workflow: an agent qualifies and scores the account, writes the score to the CRM, then calls an enrichment tool to resolve the committee roles into verified contacts — and flags any role it could not resolve rather than inventing one.

Seven ways an ICP breaks in production

Most ICPs are not wrong on the day they are written. They break later, in predictable ways.

1. It describes who you have, not who you want

If your first 40 customers came from the founder's network, the shared attributes of that cohort describe a network, not a market. Separate accounts by source before you derive anything from them.

2. It is written at the wrong altitude

“Enterprise companies that value data quality” cannot be queried, cannot be scored and cannot be disagreed with. If two reps could read it and build different lists, it is too vague to be operational.

3. Nobody removed anything

ICPs written by committee accumulate criteria and never lose them. Every added attribute shrinks the list geometrically. Six criteria at 60% pass rate each leaves you 4.7% of the market. Count your criteria; then justify each one against closed-won data.

4. It ignores the timing dimension entirely

Fit tells you whether an account should buy. It says nothing about whether they will buy this quarter. Without behavioural triggers, a perfect-fit ICP produces a perfectly-targeted list of people who are not in market, which reads to the sales team as a data-quality problem.

5. It never gets refreshed

Companies change headcount, get acquired, change stack and move markets. An ICP list built 18 months ago and never re-run is scoring accounts against a snapshot of a world that has moved. Re-run the filter quarterly; re-derive the weights annually.

6. It exists in a document rather than a system

If the ICP lives in a deck, it is applied inconsistently by whoever remembers it. If it lives as a scored field on the account object, it is applied identically by routing, sequencing, scoring and reporting. See CRM data enrichment for Salesforce, HubSpot and Dynamics 365 for how the fields get maintained once they exist.

7. It assumes contact data is free

This is the expensive one. Every ICP implies a contact-sourcing bill, and the structure of that bill decides how aggressive you can afford to be. If you pay per lookup or per prepaid credit, every unmatched record is money spent on nothing — and unmatched records concentrate in exactly the segments that are hardest to reach and often most valuable.

The point is not that one model is always cheaper. It is that prepaid and per-lookup structures make your hardest segments your most expensive ones — which pushes teams to quietly narrow the ICP toward whatever the incumbent vendor happens to cover well. That is a data contract shaping your go-to-market strategy, which is precisely backwards.

What the ICP means for sales, marketing and RevOps

The same document does three different jobs depending on who is holding it. Most internal arguments about the ICP are really arguments about which of these three jobs is being served.

The RevOps failure is the one that quietly wastes the most money. A scoring model that reads six fields will produce a defensible score only when those six fields are populated. If employee_count is missing on a third of your accounts, a third of your ICP scores are guesses dressed as arithmetic — and every routing decision downstream inherits that. Fill the fields before you weight them.

What the big platforms mean when they say “ICP”

Every major sales-intelligence platform now ships something called ICP tooling, and the label covers three quite different products. It is worth knowing which one you are buying.

Look-alike modelling — you upload your customer list, the platform finds statistically similar companies in its database. ZoomInfo and Apollo both do this well at scale. The output is a list, and its quality depends entirely on how representative your uploaded customers are.

Intent and signal layers — the platform tells you which accounts are researching your category. This addresses the timing dimension rather than the fit dimension, and it is a genuinely different product from an ICP filter, priced accordingly.

Contact resolution — turning the qualified account into reachable people. This is where the platforms diverge most, particularly on direct mobile coverage outside North America, and it is the layer worth testing on your own list rather than accepting from a coverage claim. If you are comparing options, our breakdowns of ZoomInfo alternatives, Cognism alternatives and Lusha alternatives compare them on that basis, and the enrichment tools roundup splits them by job rather than by brand.

Operationalising the ICP in your CRM

Turning the profile into working infrastructure takes five fields and one scheduled job.

The scheduled job re-runs the filter and re-enriches contacts on a cadence — monthly for high-velocity segments, quarterly for stable ones. Whether that runs through an enrichment API, a native CRM integration or a scheduled bulk file matters less than that it runs at all. Contact data that is never refreshed is a depreciating asset with an unusually steep curve, which is the same reason cold-call connect rates decay long before anyone changes the script.

One thing to check before you build any of it

If you operate in or contact people in the EU or UK, business emails and personal mobile numbers are personal data. Your ICP work needs a lawful basis, a source you can evidence, and a notification path — and channel permissions differ by country in ways that will reshape which segments are outbound-viable. Our guide to European email and phone enrichment providers covers the compliance mechanics by country. Sort this before your first list, not after your first complaint.

Is your ICP working? Four numbers that tell you

An ICP is a hypothesis about which accounts produce the best outcomes. Hypotheses get tested. Four measurements, run quarterly, will tell you whether yours is real or whether it is describing your existing customer base back to you.

The fourth measurement is the one teams most often misread. Consistent revenue from outside the profile does not mean the ICP should be abandoned. It usually means you have two ICPs and are managing them as one — which shows up as a scoring model that satisfies nobody and a sales team that has quietly stopped consulting it.

A copy-paste ICP template

Fill this in, put it where the team can find it, and version it with a date. If a section is empty, that is the section to work on next.

ICP: [Segment name]                      Owner: [Name]   Date: [YYYY-MM-DD]
Derived from: [n] closed-won accounts, [date range]

━━ KNOCKOUTS (binary — fail any, exclude the account) ━━━━━━━━━━━━━━━━━
  1. Region not in: [list]
  2. Headcount below: [n]
  3. Industry in prohibited list: [list]
  4. Locked into: [competitor / contract type]
  5. [Your fifth]

━━ SCORED CRITERIA (weights total 100) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Firmographic fit .......... 20   [industry, size, geography, structure]
  Commercial fit ............ 20   [expected ACV, budget owner, terms]
  Technographic fit ......... 15   [stack, integrations, switching cost]
  Timing signal ............. 15   [hiring, funding, leadership, expansion]
  Committee access .......... 15   [can we name the 4–6 roles?]
  Contact reachability ...... 15   [verified email + mobile coverage]

  Tier A 75–100 | Tier B 50–74 | Tier C 30–49 | Tier D 0–29

━━ BUYING COMMITTEE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Economic buyer ..... [titles, incl. variants]
  Champion ........... [titles]
  Technical evaluator  [titles]
  End user ........... [titles]
  Likely blocker ..... [titles]

━━ REACHABILITY (measured, not assumed) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Sample size tested ........ [n accounts]
  Verified email coverage ... [%]   by country: [breakdown]
  Verified mobile coverage .. [%]   by seniority: [breakdown]
  Workable accounts ......... [%]   (≥ [n] roles reachable)
  Channel permitted ......... [email / phone, by country]

━━ EVIDENCE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Top-quintile win rate ..... [%]  vs overall [%]
  Median sales cycle ........ [days] vs overall [days]
  Net revenue retention ..... [%]  vs overall [%]

━━ REVIEW ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Filter re-run ............. [monthly / quarterly]
  Weights re-derived ........ [annually]
  Next review date .......... [YYYY-MM-DD]

If you are starting from zero customers, fill in the knockouts and the committee first. Those two sections are the ones you can write from product knowledge alone, and together they are enough to build a first list and start measuring. For the wider process of turning that list into pipeline, see our guide to B2B lead generation and the practical mechanics in lead enrichment.

Frequently asked questions

Sources referenced: Gartner, The B2B Buying Journey (buying groups of six to ten decision-makers for complex solutions); Forrester, The State of Business Buying, 2024 (average of ~13 stakeholders per purchase; majority of decisions crossing two or more departments). The interactive models in this guide are calculators, not benchmarks — all default values are illustrative and intended to be replaced with figures measured on your own data.