How to Turn Customer Evidence Into a Narrow ICP
Founders with a first batch of interviews, pilots or early deals should turn that evidence into a narrow ICP by listing every account against what it actually did, finding the traits shared by the accounts that committed and missing from those that did not, and rewriting each trait as a signal someone else could check from outside. The main constraint is that the ICP can only be as narrow as the committed accounts support.
The result this process should produce
This guide is for founders before or just after launch who already hold customer evidence, such as interview notes, pilots, paid trials and lost deals, but whose customer profile is still too broad to build an account list from. If you are still deciding between two or three candidate segments, start with how to choose your first ideal customer profile, which compares segments side by side. This guide starts after that choice, with the records you already have, and turns them into a definition precise enough to act on.
The process should produce three things. The first is a one-sentence ICP statement in which every term can be checked: industry, territory, the trigger event, the buyer's job title, the budget that pays and at least one exclusion, plus a size band when the evidence supports one. The second is a signal sheet that says where each term is checked, whether in a public record, on the account's website, in a job posting or on the first call. The third is a set of acceptance tests, including an account list built from the signals by someone who has not read the interview notes.
Narrow has a practical meaning here: the statement leaves out some accounts you have already talked to. If every account in your notes fits the profile, the evidence has not narrowed anything yet. The stage-by-stage go-to-market guide recommends narrowing first and widening with evidence. This is the step where the narrowing happens.
Line up the evidence in an ICP evidence matrix
Start with one row per account you have real evidence about, not one row per conversation. Include the accounts that declined, went silent or stopped a pilot. They provide the contrast that makes narrowing possible. Leave out accounts you have only researched. The first column records the strongest step each account completed, from paying, through running a pilot with its own data or introducing the person who approves purchases, down to an interview only or a decline. Sort the rows by that column, with the strongest commitment at the top.
| Column | What to record | Where it comes from |
|---|---|---|
| Strongest completed step | Paid, pilot with own data, approver introduced, interview only, declined or silent | Dated notes and the sales record |
| Firmographics | Industry, employee band, location, ownership and how the account sells | Website, public records and the first call |
| Trigger | The event in the months before the account engaged, and whether it left a public trace | The buyer's account, checked against a public source |
| Buyer roles | Who felt the problem, who compared options and who approved the purchase or pilot | The approval path of the actual purchase or pilot |
| Budget source | What the current workaround costs in money or paid time, and whose budget covers it | The workaround, seen or described |
| Delivery gaps | What the account needed that the product does not do | A written gap list confirmed with the buyer |
Tag every cell as said, seen, done or assumed, with the same tags as the customer interview notes template so that a reported event is not mistaken for an observed one. Then draw a line under the accounts that committed, meaning those that paid, ran a pilot with their own data or introduced the approver. The working question is not what the committed accounts have in common. Many traits are shared by every account you met, simply because you chose whom to contact. The question is what the committed accounts share that the accounts below the line lack. A trait present on both sides of the line does not narrow anything.
Prepare an observable buying trigger
For each committed account, write down the event that came before it engaged, in the buyer's own words where possible. Then look for a trace of that event the account left in public: a job posting, a permit, an announcement of new equipment or a new location, a leadership hire, a regulation taking effect or a funding announcement. Daniel's go-to-market mentoring page describes selecting accounts on signals such as hiring, funding, regulatory changes, leadership moves or visible evidence of the manual process being replaced.
Turn the event into a signal with three parts: what happened, where it can be seen and how recent it has to be. A trigger that appears only above the line is a narrowing term. A trigger that buyers described but that leaves no public trace still belongs in the profile, as a qualifier checked on the first call rather than a filter for building the list. Mark which is which, because a list built on filters that nobody can check from outside turns back into guesswork.
Work through the specific buyer role
Read the buyer columns of the committed rows only. For each one, write the job title of the person who brought you in and the title of the person who approved the purchase or pilot, in the form the account itself uses on its website or in job postings, so that list building can match them. If the approver title is the same across the committed accounts, it goes into the statement. If it differs, check whether the rows split into two groups with different approval paths. Two groups usually means two segments, and the one with more completed steps comes first. If you cannot name the approver for most committed accounts, the narrowing stops here until the next conversations fill that column.
Resolve budget and delivery fit
Budget and delivery work mostly as exclusion rules. Compare the budget source above and below the line. If the committed accounts pay for a workaround in money or paid staff time and the others tolerate the problem for free, write that down as a term: accounts without a paid workaround are out. Do the same with delivery gaps. If one group of accounts needed integrations or custom work that the committed accounts did not, exclude that group explicitly, even if it is large.
Industry and size terms need to be stated precisely enough to count. The Census Bureau describes the North American Industry Classification System (NAICS) as the standard used by federal statistical agencies in classifying business establishments, organized from two-digit sectors down to six-digit national industries. Use the most specific level that still contains every committed account. The Census Bureau's County Business Patterns program presents data on the number of establishments by industry detail and by employment size of the establishment, at levels from states and counties down to ZIP codes. A count for your industry code and size band in your territory shows whether the narrowed profile still leaves enough accounts for the next round of outreach. It is a count, not a list of names, and it says nothing about which establishments are in market.
A hypothetical worked example from brewery evidence
The following is a hypothetical example with invented details. It illustrates the method and does not describe a real company, customer or measured result.
A founder has built a tablet app that replaces the paper fermentation logs kept at independent craft breweries and turns the readings into weekly production reports. Over two months the founder has gathered evidence from twelve breweries. At one of them, the head brewer opens a sample valve on a stainless fermentation tank, fills a tall glass cylinder, floats a hydrometer in it and reads the number aloud while the founder checks it against the paper log clipped to the tank. The matrix, grouped by strongest completed step, looks like this:
| Strongest step | Firmographics | Trigger | Buyer roles | Budget source | Delivery gaps |
|---|---|---|---|---|---|
| Paid: two breweries | Independent; one sells kegs to bars beyond its taproom, one sells only in its taproom | Both added fermentation tanks in the previous six months, announced on their websites (Seen) | Head brewer chose the app; the owner approved (Said) | Head brewer copies the paper logs into a spreadsheet each week in paid hours (Said; spreadsheet seen at one) | None |
| Pilot with own data: three breweries | Independent; two sell beyond the taproom | Two added tanks; one posted a job for a second-shift cellar worker (Seen) | Head brewer chose; the owner approved at two, an operations manager at one (Said) | Weekly spreadsheet copying (Said) | One asked for a link to a lab instrument |
| Interview only: four breweries | Independent; three sell beyond the taproom | No expansion in the past year (Seen: no announcements or postings) | Owner is also the head brewer at three (Said) | Paper logs only, no copying (Seen at two) | None mentioned |
| Declined or silent: three breweries | Two owned by a larger brewing group; one independent taproom | One group brewery was adding tanks (Seen) | Software chosen at group head office (Said) | The group already pays for a production system (Said) | Needs integration with the group system |
Selling beyond the taproom appears on both sides of the line, so it does not narrow anything. Recently added capacity appears in all five committed rows and in none of the independent rows below the line. The only expanding brewery below the line is group-owned and buys through head office, so group ownership becomes an exclusion. A paid workaround, the weekly spreadsheet copying, appears above the line and not below it. The owner approved at four of the five committed breweries, with the head brewer choosing the product, so the statement names both roles and the first call checks who signs.
The narrow ICP statement reads: independent craft breweries within a day's drive, not owned by a larger brewing group, that added fermentation tanks or a production shift in the previous six months, where the head brewer keeps the production logs by hand, copies them into a spreadsheet and chooses the tool, and the owner or an operations manager signs. The founder then writes the signal sheet:
| Term | Signal | Where it is checked | Use |
|---|---|---|---|
| Independent ownership | No parent group named on the about page or in ownership news | Website and local news | List filter |
| Added capacity | An announcement of new tanks, or a posting for a cellar or second-shift production role, within six months | Website, social posts and job boards | List filter |
| Territory | Address within a day's drive of the founder | Website | List filter, checked against an establishment count for the industry code |
| Paid workaround | Head brewer copies handwritten logs into a spreadsheet | First call | Qualifier |
| Approval path | Head brewer chooses; the owner or an operations manager signs | First call | Qualifier |
A part-time assistant who has not read the interview notes builds a list of twenty breweries from the three filters. The founder writes down the assumption each term depends on: that expansion, not size, is what makes the paper logs painful; that whoever signs does so without a formal procurement step; and that the spreadsheet copying is common among expanding breweries rather than a habit of the five committed head brewers.
Acceptance criteria and the next action
The statement is ready to use when it passes these tests:
- Fits the committed accounts. Every account that paid, ran a pilot with its own data or introduced the approver matches every term.
- Excludes most of the rest. Most accounts below the line fall outside the statement. If they still fit, the profile is no narrower than the list you started with.
- Every term is checkable. Each term is either a filter someone can check from outside the account or a qualifier answered on the first call.
- Someone else can build the list. A person who has not read the notes produces an account list from the filters alone, and an establishment count for the industry and size terms suggests there are enough accounts in reach for the next round.
- Each assumption has a test. Every assumption behind a term is written down with an observation that would prove it wrong.
The next action is outreach to the blind list, together with a small comparison group of accounts that miss one filter, recording outcomes in the same matrix with the same tags. If accounts that fit the statement reach a pilot or payment more often than the comparison group, keep the profile and increase volume. If they do not, change the term the evidence contradicts rather than adding new ones. Rewrite the statement when the matrix changes, not when a large prospect outside it appears.
Common mistakes and limits
- Narrowing on traits every account shares. Traits common to the whole matrix describe how you chose contacts, not why accounts committed.
- Using a private trigger as a list filter. An event only the buyer can see belongs on the first call, not in list building.
- Treating one paying account as a pattern. A single strong account is a lead to investigate, not a profile.
- Mixing up the user and the approver. The person who likes the product and the person who signs often hold different titles.
- Widening the statement because the count looks small. A small but reachable segment teaches faster than a large one that nobody can find.
The method has limits. With a handful of committed accounts, any pattern is a hypothesis for the next round, not a finding. It suits products sold to identifiable organizations with an approval path; self-serve and consumer products rely more on usage and purchase data. Public counts confirm that accounts exist, not that they will buy. The matrix, signal sheet and acceptance tests are an editorial framework, not a measured result.
Related guides and next step
- Comparing candidate segments before this step: How to Choose Your First Ideal Customer Profile
- Recording the evidence this process uses: Customer Interview Notes Template for Founders
- Placing the ICP in a wider plan: How to Build a Startup Go-to-Market Strategy by Stage
Daniel's go-to-market mentoring page describes defining a narrow ICP, buying trigger and decision-maker as the first step of his process, and says the weekly read refines the ICP, the positioning, the sales collateral and the roadmap. That account is self-reported on his site.
Bring your evidence matrix and signal sheet to a free 30-minute strategy session to review your ideal customer profile with your actual evidence and constraints.
Frequently asked questions
How do you turn customer evidence into a narrow ICP?
List every account you have evidence about in one matrix, sorted by the strongest step each completed, from paying down to declining. Find the traits that the committing accounts share and the others lack, then rewrite each one as a signal that can be checked from outside the account or on a first call. Keep only the terms the committed accounts support.
Which customer records belong in an ICP evidence matrix?
One row per account with real contact: paying customers, pilots, accounts that introduced an approver, interview-only accounts and accounts that declined or went silent. Record firmographics, the event before they engaged, the approval path, the budget behind the current workaround and product gaps, each tagged as said, seen, done or assumed.
What are common mistakes when narrowing an ICP from customer evidence?
Narrowing on traits every account you met shares, using a trigger only the buyer can see as a list filter, treating one paying account as a pattern, mixing up the user's title with the approver's, and widening the statement again because the account count looks small.
How do you know a narrow ICP is ready to use?
When every committed account fits it, most accounts that did not commit fall outside it, each term can be checked from outside or on a first call, and someone who has not read your notes can build a usable account list from the filters alone.