Definition
An ideal customer profile describes the company situation in which an offer is most likely to create and capture value. It is not a buyer persona, a list of attractive sectors, or a lead score. A useful ICP states the problem context, the triggering event, the buying conditions, the adoption constraints, and the evidence that supports each inclusion or exclusion.
Closed-won deals are a strong starting point because they contain real buying journeys. They are not a complete sample. NIST describes survivorship bias as a risk when a dataset contains only cases that survived a process and may therefore be non-representative (NIST guidance on dataset construction). For an ICP, the practical correction is to compare wins with closed-lost, no-decision, and disqualified deals that faced a similar offer and market context.
Explore the Knowledge guides for sales for adjacent methods on pipeline review and commercial decisions.
Prerequisites
Choose a coherent analysis window. Do not mix an old offer, a recent pivot, several pricing models, and several routes to market in the same comparison. If the offer changed materially, split the deals into separate cohorts and explain the boundary.
Prepare one source record per deal. Useful fields include outcome, offer sold, initial problem, trigger, buyer roles, alternative considered, objection, proof requested, implementation constraint, reason for winning or losing, and the link to the original note. Mark missing information as unknown instead of filling it from memory.
Include all available outcomes. Salesforce reporting distinguishes closed opportunities through Closed Won and Closed Lost stage types (Salesforce guidance on opportunity stage types). Your CRM may use different labels, so map them into a small common outcome vocabulary before comparing them. Add no-decision and disqualified cases when they are tracked outside the closed stages.
Finally, assign a reviewer who did not own every deal. That person should challenge hindsight explanations, identify missing records, and prevent one memorable account from becoming the whole market definition.
Steps
-
Define the decision. Write what the ICP will control: account selection, message angle, discovery priority, or market test. One ICP should not silently answer every commercial question.
-
Normalize the deal set. Keep cases that concern the same offer and a comparable buying situation. Record why each case is included. Place incomparable deals in a separate file rather than forcing them into the pattern.
-
Build a deal evidence card. Copy only observable facts from notes and records. Separate the customer’s stated problem from the seller’s interpretation. Add the source link and label any absent field as unknown.
-
Compare outcomes by question. Ask which problem existed before contact, what changed, who carried the problem, what made action possible, what blocked adoption, and why the deal ended as it did. Look for contrasts, not merely traits shared by winners.
-
Write the ICP as evidence rules. For each proposed criterion, record supporting wins, contradicting losses, exclusions, confidence, and the next test. Use three labels: observed, plausible, and unknown. A firmographic trait remains plausible until the deal evidence explains why it mattered.
-
Publish a version, not a verdict. State the offer, scope, review owner, unresolved questions, and next review trigger. A new loss, a new segment, or a material offer change can reopen the profile without erasing the previous reasoning.
Worked example
Consider a hypothetical workflow software company. Its won deals are concentrated among service businesses, so the first draft says the target sector is the decisive criterion. The team then reviews lost and no-decision deals from the same sector. Several had the right size and buyer title but no operational change, no internal owner, and no reason to replace the current process.
The team returns to the source notes. In the wins, a process change had created visible coordination work, one manager owned the problem, and the buyer could name the consequence of waiting. In the losses, those conditions were absent or undocumented. The evidence does not prove that the sector is irrelevant. It shows that sector alone does not explain the contrast.
The revised ICP therefore keeps the sector as a search boundary, but promotes three contextual conditions: an active workflow change, a named owner, and a documented cost of delay. It excludes teams seeking a one-off service rather than an operating system. Each condition links back to deal notes, and the unresolved role of company size becomes a hypothesis for the next review.
This is illustrative. It is not a customer story and claims no measured performance result.
Common mistakes
Studying winners only. This makes every trait of a successful account look important. A trait becomes useful when it helps explain a contrast with a comparable loss or no-decision.
Counting revenue as fit. A large deal can hide custom work, founder involvement, slow adoption, or weak repeatability. Keep commercial value and repeatable fit as separate observations.
Mixing offers and periods. A deal won under a former proposition may describe a market that no longer exists for the current offer. Split the evidence before searching for patterns.
Turning correlation into a rule. Industry, headcount, and geography are easy to filter, but the available records may not explain why they mattered. Keep them as hypotheses until the buying mechanism is documented.
Writing a persona instead of an ICP. The buyer’s role matters, but the ICP first describes the company and buying situation. The buying committee can then be attached to that situation.
Freezing the profile. An ICP without an owner, version, and review trigger becomes inherited folklore. Preserve the reasoning so later evidence can refine it.
Tools
A spreadsheet is enough for the first pass. Use one row per deal and fixed columns for outcome, source record, problem, trigger, buyer roles, adoption condition, objection, proof, exclusion, and unknowns. Add a separate criterion table that lists supporting and contradicting cases. This prevents the final wording from losing its evidence trail.
Use the CRM as the record index, not as proof that the fields are complete. Link meeting notes, proposals, loss reasons, implementation notes, and customer feedback where available. If a reason was selected from a generic dropdown but the note says something different, preserve both and flag the conflict.
After the ICP is versioned, the search can begin. Ember Lead Intelligence prioritises opportunities from the available context. Lead Intelligence monitors signals about people and companies to keep context current. Lead Intelligence proposes the next action and channel that fit the lead situation. These capabilities help apply a stated profile to current opportunities. They do not validate the historical sample or decide which evidence is causal.
Want a product example in How Lead Intelligence Works for Founder Introductions??
When to use this method
Use this method when the team has real closed outcomes, can identify the offer that each deal concerned, and can retrieve enough source material to distinguish observation from memory. It is especially useful when the existing ICP is a broad firmographic list and the team cannot explain why those filters predict a buying situation.
The method also works as a lightweight review after an offer change, provided the old and new evidence remain separate. A small sample can still reveal useful questions and exclusions. Present those outputs as provisional, not representative of the whole market.
Structure a Weekly Outbound Review for Small B2B Sales Teams can then capture whether the profile continues to explain current conversations.
When not to use it
Do not use closed-deal analysis as the main method when the offer is new, the company has no comparable outcomes, or one unusual contract dominates the history. In those cases, start with a narrow problem hypothesis and customer discovery. Treat the first ICP as a research frame, not a conclusion.
Do not merge deals from materially different products, countries, channels, or sales motions merely to create a larger sample. More rows do not repair a broken comparison. Do not use the method to exclude a segment solely because no one has sold to it yet. Absence of evidence may reflect the previous targeting strategy.
Stop if source notes are too incomplete to reconstruct why deals progressed or failed. Improve the capture process first, then revisit the profile when new decisions have traceable evidence.
Action plan
Create the cohort definition and the blank evidence card before opening the CRM. Export the candidate deals, map their outcomes, and remove only cases that violate the written scope. Keep a removal log.
Have the deal owners fill factual gaps, then ask an independent reviewer to challenge the proposed patterns. Draft the ICP in a decision table with inclusion conditions, exclusion conditions, supporting records, contradictions, unknowns, and the next test. Choose one version owner.
Apply the draft profile to a bounded prospecting or discovery batch. Review whether each criterion helped explain relevance, not only whether a meeting occurred. Record contradictory evidence and revise the profile when the stated review trigger is reached.
Ember data
No proprietary performance figure is used in this article. No conversion lift, response rate, or market representativeness is claimed.
The product statements are limited to the current catalogue. Lead Intelligence prioritises opportunities from the available context. Lead Intelligence monitors signals about people and companies to keep context current. Lead Intelligence proposes the next action and channel that fit the lead situation. After the mission, Lead Intelligence shows the contacts analysed, signals detected and priority actions actually recorded by Ember.
Sources and methodology
The method combines three kinds of evidence. Salesforce documents how closed opportunity stage types are represented in reporting. NIST provides the general warning about selection and survivorship bias in datasets. The UK business evidence annex explains that segmentation should match its objective, rely on reliable and comparable data, and avoid unnecessary complexity while recognising exclusion risks (UK business evidence annex).
The source pages were checked during this repair. The operational worksheet, worked example, and review cadence are editorial methods, not results reported by those sources.
Types of sources used: official pages, institutions and named studies.
Sources
FAQ
Can a small B2B team define an ICP using only closed-won deals?
Closed-won deals are a useful starting point, but not a safe final sample. They reveal where the team captured value while hiding comparable accounts that lost, stalled, or were rejected. Draft initial patterns from the wins, then compare them with other outcomes. Keep a criterion only when the source records explain the contrast, and label every remaining idea as a hypothesis.
Which fields should a B2B ICP deal evidence card contain?
Record the outcome, offer, customer problem, triggering event, buying roles, alternative considered, objection, proof requested, adoption constraint, reason for the result, and the source note. Add an explicit unknown value when evidence is missing. The card should contain observations, not reconstructed certainty. A separate column can hold the team’s interpretation so reviewers can challenge it.
What is the difference between an ICP built from deals and B2B lead scoring?
The ICP explains the company and buying situation that the team intends to pursue. Lead scoring ranks individual records against chosen signals or rules. Build the ICP first from contrasted deal evidence, exclusions, and unknowns. A scoring model can later operationalise selected criteria, but a high score cannot repair an ICP whose assumptions were inferred only from winners.
How should a small B2B team handle a limited ICP deal sample?
Use the sample to generate questions, exclusions, and provisional criteria, not a claim about the whole market. Preserve the exact deals behind every statement, show contradictions, and make uncertainty visible. Choose the next discovery or prospecting test that could disprove the most important assumption. Review the profile when new comparable outcomes appear instead of waiting for arbitrary scale.
What role can Lead Intelligence play after the ICP review?
Ember Lead Intelligence prioritises opportunities from the available context. Lead Intelligence monitors signals about people and companies to keep context current. Lead Intelligence proposes the next action and channel that fit the lead situation. The team still defines the ICP, checks the source records, and decides which criteria are supported. Product context helps apply a decision, but it does not make a biased sample representative.
When should a B2B ICP be revised?
Set review triggers tied to evidence: a material offer change, a new segment, a cluster of contradictory losses, or repeated missing conditions in current conversations. Keep a version owner and preserve the prior reasoning. Revision does not mean rewriting the profile after every call. It means reopening the criteria when comparable evidence challenges an assumption or changes the decision context.