JournalResearchNo. 02
How to Build an ICP for B2B Outbound, Even With Few Customers
Build an ICP for B2B outbound as a testable hypothesis, even with few customers. Criteria, persona, unknown data and a free ICP worksheet.
Matija Nikolić7 min read9 sections

Contents
Published by Perpetual
To build an ICP for B2B outbound, write it as a hypothesis you can test: two to four criteria that describe the companies most likely to need your offer, one of which is the anchor every company must meet, plus one buyer persona and a handful of example companies. Then check the list it produces, keep companies with missing data separate from confirmed fits, and change one thing at a time based on the replies you get.
A common approach is to analyze your best customers and assign scores. That works once you have dozens of customers. This guide is for teams with few or none yet, which is when outbound needs an ICP most. It reflects the rules we use when a Perpetual campaign is created.
What an ICP is for in outbound
An ideal customer profile describes the kind of company that gets the most value from your offer. In outbound, it does one practical job: it decides which companies enter your list. Everything later depends on that list, so an ICP that is too broad wastes research and sending capacity, and one that is too narrow leaves you with nothing to send.
Keep three ideas separate:
- ICP: the kind of company. Industry, size, business model, region, and what it does.
- Persona: the person inside it who owns the problem.
- Signal: a recent event suggesting the timing may be right, such as a relevant job posting.
The ICP says who could buy. A signal suggests who might be ready now. Our comparison of account discovery and account monitoring covers how the two work together.
Write the ICP as one testable hypothesis
A useful outbound ICP fits in a few lines:
| Part | What to write | Example (invented) |
|---|---|---|
| Anchor criterion | The one thing every company must be. | B2B software company |
| Supporting criteria | One to three more conditions. | Roughly 20 to 200 employees; sells to other businesses in the US |
| Persona | One primary title, with alternatives. | Head of Sales; alternatives: VP Sales, Founder |
| Problem and proof | What you solve and how you show it. | New sales hires take months to ramp; a sample onboarding plan |
| Example companies | Up to three real companies that fit. | Three customers or close prospects you know |
Why two to four criteria? This is the limit we enforce in Perpetual. One criterion is too broad to make a usable list. Beyond four, most teams are guessing at the extra conditions, and each guess removes companies without evidence that it should. A hypothesis that vague or that narrow cannot be tested well.
Describe what the buyer is and sells, not what you sell. A security vendor's ICP is not “cybersecurity companies”; it is the fintech or healthcare companies that need security.
How to build an ICP with few or no customers
Scoring models built from customer data need enough customers for the patterns to mean anything. With only a handful of customers, a weighted score mostly reflects chance. Use these sources instead:
- Your offer. Write down the specific problem you solve and which kind of company has it most acutely. This is your first anchor.
- Your conversations. Who replied, who bought, who said no and why. A clear no from the wrong segment is as useful as a yes.
- Your example companies. Name up to three companies that are clearly a fit. If you cannot name three, the ICP is still too abstract.
- Disqualifiers. List the kinds of companies you cannot serve well, such as regulated industries you lack proof for. Exclusions often sharpen an ICP more than inclusions do.
Treat any weights or scores you add at this stage as a starting guess. Write the date beside them and expect to change them.
Keep unknown data separate from a fit
This is a common failure. Company databases are incomplete, and headcount, headquarters country and business model are often missing or estimated. When you filter, a company with missing data can be silently included or silently dropped.
Sort each company into three groups instead of two:
- Fits: the data confirms every criterion.
- Possible: it only looks like a fit because a key field is unknown or inferred.
- Does not fit: the data shows it misses a criterion.
Start with confirmed fits. Check a sample of the possible group by hand before including it. Do not loosen a criterion to make a list bigger; change the hypothesis openly instead. Treat company sizes as approximate bands, not precise headcounts, because sources often disagree.
From ICP to a list you can actually contact
A market that fits your ICP is not the same as a list ready for outreach. Between them sit several checks, each of which removes companies:
- Fits the ICP, from confirmed data.
- Not excluded: not a current customer, open deal, competitor or recent contact.
- Has a current person matching the persona.
- Has a verified email or another usable channel.
- Has a reason to write now, if your campaign requires one.
Record how many companies pass each step. If most fall out at step three, your persona may be wrong for the company size. If most fall out at step five, the timing requirement may be too strict for the market.
Test the ICP with one segment at a time
Run one active segment first. Before adding a second, make sure it differs in persona, problem or proof, not just in a filter value, and that you have the sending capacity and time to handle its replies.
Judge the ICP by replies and qualified conversations from companies that truly fit, not by list size. Change one part at a time and note the date, so you can tell what caused a difference. Our signal-based outbound scorecard shows how to compare segments fairly.
Download the ICP hypothesis worksheet (CSV). It has rows for the anchor, supporting criteria, persona, problem, proof, example companies, exclusions and the funnel counts above, with an invented example filled in.
Common ICP mistakes
- Describing your product instead of the buyer. The ICP is about them.
- Too many criteria. Each one removes companies; make sure it earns its place.
- Treating missing data as a match. Keep possible fits apart from confirmed ones.
- Scoring without data. Weights from a handful of customers are guesses.
- Never revisiting it. An ICP is a hypothesis; replies are the test.
Where Perpetual fits
Perpetual turns this structure into a campaign. Each campaign needs two to four criteria with one anchor, a persona and at most three example companies. Its preview separates companies that fit from those that only look like a fit because a field is unknown, and criteria are never relaxed silently. It then finds companies with relevant public hiring activity, selects one contact and drafts a message from the evidence.
It does not decide your ICP for you, and the quality of the list depends on the hypothesis you give it. See the quickstart guide for how a campaign is set up.
Frequently asked questions
What is the difference between an ICP and a buyer persona?
The ICP describes the company; the persona describes the person inside it who owns the problem. Outbound needs both.
How many criteria should an ICP have?
For outbound, two to four, with one anchor that every company must meet. Fewer is too broad to act on; more usually adds guesses.
Can I build an ICP before I have customers?
Yes. Start from the problem your offer solves, the companies you would most like to win and the ones you cannot serve, then test it with a small segment.
How often should I update my ICP?
Review it when you have enough replies from one segment to see a pattern, and whenever you change your offer. Change one part at a time.













































