Your sales team asks for companies with 50–500 employees in the DACH region. A week later, the list contains manufacturers, distributors, holding companies and agencies. Every row satisfies the filters. Half the team still says the accounts are wrong.
The missing piece is a researchable ideal customer profile: a description of the operating situation in which your offer makes sense, translated into rules another person can check. This guide gives you a working template, three examples and a short exercise for validating the rules before buying contact data.
What you will leave with: one account definition, explicit exclusions, a small field specification and a decision about which companies deserve further research. The templates and numerical examples below are our proposed working methods, not results from a customer study.
What is an ICP, and how is it different from a buyer persona?
An ideal customer profile, or ICP, defines the type of organization that can benefit from your offer and that you can serve commercially. A buyer persona describes the person involved in evaluating, using or buying it. Salesforce's ICP overview makes the same company-versus-person distinction.
For a maintenance software company, the ICP might be an industrial business operating several production sites with its own maintenance team. The persona might be the maintenance manager. Finding that job title at a distributor does not make the distributor an ICP match.
Keep three questions separate:
Size, industry and location narrow the search. They rarely explain the buying situation on their own. Timing and people research should come after the account definition rather than quietly changing it.
The copyable B2B ICP template
Complete this brief for one offer and one segment. Replace vague terms with a fact that can be observed, or explicitly label them as assumptions.
“Location” deserves a sentence. A German headquarters, a German subsidiary and a German customer base are different criteria. So are group-wide headcount and headcount at one operating site. If either distinction changes whether you can deliver the offer, record it in the brief.
The most useful field is often what stays unknown. Budget, internal dissatisfaction and willingness to switch usually cannot be established from a company website. Keeping them unknown prevents a plausible story from turning into a supposed fact.
Start with customer economics, not the largest available database
Before choosing filters, review a small set of your own recent accounts: successful customers, expensive implementations, lost opportunities and companies you declined. Use records you are permitted to analyze. Look for differences in the work required to achieve a useful outcome.
Ask what made an account economical to serve. Was deployment straightforward because it used one standard system? Did the customer have an internal owner? Did a multi-site operating model create enough value to justify the price? Did procurement or implementation effort overwhelm the contract value?
Write a causal hypothesis rather than a correlation dressed as a rule. “Our three largest contracts came from Bavaria” is an observation. “Multi-site manufacturers need consistent maintenance records” is a hypothesis you can investigate across regions. Keep both, but do not confuse them.
If you have no customers yet, start with the narrowest problem you can explain and test. Mark the profile as provisional. A well-written template cannot substitute for learning from real buying conversations.
Example 1: maintenance software for manufacturers
Imagine a company selling software for scheduling preventive maintenance. This is a fictional example; the requirements below are illustrative.
Weak profile: manufacturing companies in Germany with 100–1,000 employees.
Researchable profile: businesses operating their own industrial production in Germany, with multiple production lines or sites and an identifiable maintenance function. Exclude pure distributors, engineering consultancies and holding companies without operating production.
A maintenance vacancy can establish that the function exists. It cannot establish that the current maintenance system is failing. A plant expansion can justify investigating timing. It cannot establish an approved software budget.
Borderline account: a group website lists eight plants, but the candidate row is a sales subsidiary. Do not transfer the group's production characteristics to the subsidiary automatically. Resolve which legal or operating entity the sales team would actually approach.
Example 2: recruiting services for regional care providers
A recruiting agency wants to help care providers hire qualified staff in a particular region. “Healthcare companies” is too broad: it can include software vendors, wholesalers and hospitals outside the agency's delivery area.
A stronger profile names operating care facilities in specified service regions and the roles the agency can actually recruit for. Public vacancies can support the existence of a hiring need. A directory entry establishes neither current vacancies nor an urgent staffing problem.
The unit of account matters here. A care group may centralize recruiting, while individual facilities publish their own jobs. Keep the facility and parent organization related but distinct. Decide whether qualification happens at facility level and contact research at group level.
Exclude roles and regions you cannot serve even when the organization looks attractive. A profile that ignores delivery constraints produces conversations the team cannot convert into a successful service.
Example 3: integration services for B2B software companies
An integration consultancy targets software companies with a complex customer onboarding process. A simple headcount band may miss small firms with demanding enterprise implementations and include large self-service products with little integration work.
Useful observations could include public API documentation, enterprise integration pages, implementation vacancies and customer-facing onboarding material. These can support an integration-heavy business model. They do not prove that the internal team is overloaded.
An explicit exclusion might be companies whose product is consumer-only or whose implementation work requires a certification the consultancy lacks. A useful review question is whether integration is central to the offer or merely one optional feature.
Notice how the three examples require different evidence. Copying the same industry-and-size template into every market loses the operating condition that makes the offer relevant.
Translate the brief into research fields
A paragraph is good for alignment. A small field specification makes it usable in a spreadsheet, CRM review or research job. For every decisive field, define its meaning and allowed uncertainty.
Do not make “absent” mean “not found.” A website that does not mention a maintenance department does not prove the department does not exist. Likewise, do not use a filled checkbox to represent both a confirmed negative and an unsuccessful search.
If you automate a field, test the question against one clear match, one clear exclusion and one genuinely ambiguous company first. The company research prompt guide shows how to preserve that distinction in AI-assisted research.
Validate the definition with 30 accounts
Use a small exercise to discover ambiguous rules before scaling. Thirty accounts is a practical starting point for discussion, not a statistically representative estimate of your market.
- Choose 10 plausible matches, 10 borderline accounts and 10 likely exclusions. Include sparse websites and subsidiaries.
- Have two people apply the written rules independently. Hide the other's decision during the first pass.
- Preserve the decision, evidence and time spent for each account.
- Compare disagreements by criterion. Change unclear wording before changing search volume.
- Repeat the disputed cases using the revised definition.
In an illustrative exercise, the reviewers might agree on 22 of 30 accounts and disagree on eight. If six disagreements concern whether a sales subsidiary counts as a manufacturer, you have found a specific definition problem. Buying 5,000 more rows will not solve it.
Track agreement before discussion, unknown share and minutes per reviewed account. Do not optimize agreement by rejecting every hard case or forcing uncertainty into a yes/no answer. The point is a consistent, commercially useful decision process.
Avoid five common ICP mistakes
A profile that describes every possible customer. Start with one repeatable operating problem. You can maintain several distinct profiles later, with their own evidence and exclusions.
An attractive logo overrides an exclusion. Decide who may approve exceptions and why. Otherwise, the written profile becomes optional whenever someone recognizes a company name.
Headcount becomes false precision. Employee counts can refer to a group, an entity or an old estimate. Use a review band near a threshold when the distinction matters commercially.
A timing signal replaces fit. A funding announcement does not make an unsuitable company suitable. Apply the account definition first, then use the buying-signal checklist to investigate priority.
Contact availability determines the market. An easy-to-find email is not a reason to accept an account. Qualify the company before spending on contact discovery; the enrichment cost model explains the economics.
Use the profile in Leadbase
Leadbase's company search documentation describes structured company discovery. Use search filters for the parts of the profile they actually represent. Investigate the remaining operating conditions with focused typed enrichment, supplying the company identity and one clear question.
Inspect the returned explanation and sources. An accepted research job or a correctly typed answer does not by itself establish ICP fit. Retain the acceptance decision in your chosen working system and pass only approved accounts to people or contact research.
This article is published by Leadbase. The template is a general operating method; it does not require Leadbase, and it does not imply that every public fact can be found automatically. Explore the research catalog if translating your criteria into focused research is the bottleneck.
Questions teams ask before using the template
How many ICPs should we maintain?
Use a separate profile when the operating problem, evidence or delivery model changes materially. A different industry label alone may not justify one. Begin with the segment you can explain and serve most clearly.
Should revenue or employee count be mandatory?
Only when the threshold changes buying economics or serviceability and you can interpret the available evidence responsibly. Otherwise use it for discovery, preserve its source and avoid treating a rough estimate as a hard exclusion.
How often should the ICP change?
Review it when you learn something consequential from wins, losses, implementation or repeated research disagreements. Keep a version and effective date so old and new qualification decisions remain explainable.
What should we read next?
If the account definition is ready, continue with B2B buying signals: seven events worth investigating. If reviewers still disagree about evidence, use the AI company research prompts and audit checklist before expanding the list.
Editorial methods and product documentation checked on 17 October 2026. All company scenarios and sample counts in this guide are illustrative.
