How to Create an ICP from Your Wins and Losses, Step by Step

To create an ICP, pull your last 20 closed-won and 20 closed-lost deals, record the same fields for each, find what the wins share and the losses lack, turn that contrast into criteria and disqualifiers, then test the draft against your open pipeline. Evidence beats opinion at every step.

Step 1: List your wins and make them comparable

Export recent closed-won deals from your CRM. BusinessMCP’s ICP guide suggests the last 20. Put them in a spreadsheet with one row per deal and the same columns for every row:

Column	Why it matters
Company, domain	Lets you enrich the rest
Industry	Check, but do not over-trust
Employee count	Often a stronger signal than industry
Location	Affects support, tax and language
Tools they used before	Signals readiness
Buyer title	Tells you who to target
Why they bought	The most valuable column
Deal size, time to close	Separates easy wins from slow ones
Months retained	Reveals customers who stuck

Missing cells are normal. Fill what you can, and for firmographics consider a lookup; the BusinessMCP companies API returns industry, size and location for a domain and charges only when a result is found, per its pricing page. Spot-check a handful by hand.

Step 2: Find what the wins share

Sort and filter. You are looking for clusters, not averages.

• Do most wins fall in a narrow headcount band?
• Do they share a tool, a recent hire or a recent event?
• Is the buyer title the same each time?
• Which wins were fast and stayed longest? Weight those higher.

Write down three to five traits shared by most wins. Resist the pull to include a trait only because one famous customer has it.

Step 3: Do the same for losses

This is the step most teams skip, and it is where the profile gets sharp. Pull the last 20 closed-lost deals, record the same columns plus the stated loss reason, and find the shared traits.

Wins and losses often look alike on industry and differ on size, stack or buyer. That difference is your criterion. A trait present in both groups is not predictive and should come out of the profile.

Key takeaway: a trait only belongs in your ICP if it separates wins from losses. Traits shared by both groups are decoration.

Step 4: Write criteria and disqualifiers

Convert the contrast into two lists.

Criteria are the traits wins share, written as ranges: “20 to 200 employees”, “uses a CRM”, “first sales hire within a year”.

Disqualifiers are the traits losses share: “under 10 employees”, “no budget owner involved”, “must self-host”.

The guide cited above argues that disqualifiers are the highest-value part of the profile, because they stop spend on accounts you will not win. Our dedicated piece on writing ICP disqualifiers has examples.

Then draft the document. You can use our ICP template or answer the six questions in the free ICP generator and edit the exported markdown. Either way, replace anything generic with something from your spreadsheet.

Step 5: Pressure-test against open pipeline

Apply the draft to deals currently in progress and to a list of accounts you have never contacted.

• Do your best open deals match the profile? If not, either the deals are weaker than they look or the profile is wrong.
• Does the profile exclude accounts your team gets excited about? Discuss why.
• Can someone outside the team sort 20 companies the same way you would?

Adjust the criteria until the answers feel honest. Then take it a step further and attach scores, as described in building an ICP scoring model.

What if your data is messy or thin

Some practical fallbacks:

1. Few deals. Use all of them and mark conclusions tentative. See ICP for startups.
2. No loss reasons recorded. Call or email five lost prospects and ask. Short, honest answers are enough.
3. Missing firmographics. Enrich by domain and verify a sample.
4. Self-serve customers with no sales touch. Use signup data and usage depth as the equivalent of deal data.

Timeline and ownership

A reasonable plan for a small team:

• Day 1: export deals, build the spreadsheet.
• Day 2: enrich and clean, find shared traits.
• Day 3: draft the one-page profile, review with sales and customer success.
• Week 2: test against pipeline, adjust, publish.

Give the document one owner and a review date. The guide suggests quarterly reviews, and revising sooner if closed or churned accounts drift from the document. An ICP built once and left alone becomes a historical record rather than a tool.

Worked mini-example of the contrast step

Imagine twenty wins and twenty losses for a fictional analytics product. After the spreadsheet is built, you notice the following.

Trait	Share of wins	Share of losses	Verdict
Software industry	most	most	Not predictive, drop it
20-200 employees	most	few	Keep as a criterion
Already runs a CRM	most	about half	Keep, probably as a soft criterion
Founder is the buyer	few	most	Candidate disqualifier or separate profile

(The words “most” and “few” stand in for your own counts; always use actual numbers from your data.) The pattern the table shows is typical: the industry that feels like the core of your business turns out to be shared by winners and losers alike, while size and buyer role carry the signal. That echoes the guide’s claim that company size, stack maturity and role often predict fit better than industry.

Handling outliers

Every dataset has outliers: the huge customer who bought through a friend, the tiny one who loves you. Do not build the profile around them, but do not ignore them either. Record each as a note: “outlier, bought via partner introduction”. If outliers cluster in some way, they may be a second ICP worth testing on purpose.

Turning the draft into action

Once the document is approved, put it to work immediately. Add its fields to your CRM so that leads can be tagged. Use it to rank a list of target accounts (see building a target account list). Brief the team in fifteen minutes, show two examples of an in-profile and an out-of-profile account, and book the first quarterly review before the meeting ends.