The ICP Fit Test: Score Your Won and Lost Deals Before You Trust Your Profile

Every company has an ideal customer profile. Almost none has evidence for it. The ICP fit test fixes that in an afternoon: take your last 50 to 100 closed-won deals, take a similar pile of closed-lost ones, and check which traits show up in the deals you won but not the ones you lost. If your profile cannot tell those two piles apart, it is not a profile. It is a wish. And you do not need a data team or a single line of code to find out, because the whole test now runs in a chat window.

Why Most Ideal Customer Profiles Are Folklore

The typical profile was written in a planning meeting, built from three or four memorable customers, and never checked against anything. That is not a fringe problem: 68% of B2B companies have no clearly defined profile at all, and plenty of the ones that do are working from a document nobody has re-opened since it was drafted.

How a Profile Goes Stale Without Anyone Noticing

• It was reverse-engineered from a handful of favorite logos rather than every deal you actually closed.
• It says things like "growing tech companies in the US," which fits tens of thousands of companies you will never win.
• It mentions traits, like revenue range or the tools a company runs, that your CRM has no field for, so nobody can ever check it.
• The product and the market both moved, and the document did not.

GTM advisor Tim Hillison put it plainly on LinkedIn: everyone is convinced their profile is right, up until the data says otherwise. The fit test is how you ask the data.

The Fit Test: One Question, Two Piles of Deals

The test asks a single question: do the traits in your profile show up in the deals you won and not in the deals you lost? A trait that appears in both piles at the same rate is a description of your market. A trait that appears in 60%+ of wins and under 40% of losses is a real qualifier, the kind worth building a targeted prospect list around.

Why the Lost Pile Is Not Optional

Studying only your wins tells you who you sold to, not who you should sell to. Say 7 in 10 of your won deals are mid-size software companies. Sounds like proof, until you look at the lost pile and find 7 in 10 of those are mid-size software companies too. The lost pile is the control group that turns a story into a test:

• It exposes traits that carry no signal, the ones both piles share, so you can delete them.
• It surfaces the accounts that look ideal on paper and keep saying no, which become explicit walk-away criteria.
• It gives you the number that actually matters: the gap between the win match rate and the loss match rate.

One Caveat: Sort Your Losses First

A deal you lost on price or timing was often a genuine fit. Tag those and set them aside; keep the losses where the account simply was not right as your comparison pile. Mixing them in waters down the signal.

How Many Deals Before the Numbers Mean Anything?

Fifty won deals is the floor, 90 or more is the comfortable zone, and everything should come from the last 12 to 18 months. Under 50, one odd account can swing a percentage by double digits and send you rewriting the profile around a fluke.

• Take every win in the window. Hand-picking the memorable ones rebuilds the exact bias you are trying to remove.
• Stop at 18 months back. Older deals closed against a different product and a different price. They describe a company you no longer are.
• Keep the piles roughly the same size. Ninety wins against fifteen losses makes the loss-side percentages meaningless.
• Under 50 total wins? Run it anyway, label the result directional, and re-run each quarter as deals accumulate. A rough test on 30 real deals still beats an untested guess.

The Fields Your CRM Never Had (and How to Fill Them in Chat)

Here is where most teams stall: the export from your CRM has a company name, a domain, a deal amount, and an outcome, and your profile is written in traits none of those columns contain. You cannot test "500 to 2,000 employees, running a modern cloud stack, hiring in engineering" against a spreadsheet that only knows the company's name. Filling in those missing traits, what the data world calls enrichment, is the step every ICP guide skips and the whole reason profiles go untested.

The Traits Worth Filling In

Trait in your profile	What to ask for, in plain words	In your CRM today?
Recent company activity	Hiring pushes, funding news, expansion moves	No
Tools they run	Their CRM, cloud platform, analytics setup	Almost never
Company size and money	Employee range, revenue range, industry	Sometimes, usually stale
Funding history	Last round, total raised, acquisitions	No
Team growth	Headcount trend, which departments are growing	No

Vibe Prospecting fills in every row of that table from a single connection in Claude or ChatGPT, drawing on 150M+ company profiles and 18 categories of recent company activity. Powered by Explorium Enterprise Business Data, a messy export still resolves to the right companies at 97.8%+ match accuracy, misspellings and renames included. If you want to sanity-check the source first, run a side by side data provider comparison on a slice of your own deals.

Run the Whole Test in a Chat Window

The fit test used to require an engineer, an API contract, and a script. In chat it is three messages. Connect Vibe Prospecting in Claude or ChatGPT and walk through it:

• Message 1, upload and match: "Here is a CSV of 120 companies with a won or lost column. Match each one to its real company record and tell me which ones you could not resolve."
• Message 2, fill in the traits: "For every company, add employee range, revenue range, industry, the main tools they run, funding history, and whether they are hiring in engineering. Show me 5 sample rows before doing the full list."
• Message 3, score it: "My ideal customer is 100 to 1,000 employees, $10M to $100M revenue, software or logistics, hiring in engineering. For each trait, what share of won deals matches and what share of lost deals matches? Give me the gap per trait."

The preview step in message 2 matters: you see sample records and a cost estimate before anything is charged, so you audit quality before committing. And because up to 1,000 records per request are handled server side, the full export survives intact instead of being quietly trimmed to whatever fits in a chat context.

Connect It First

Add Vibe Prospecting from the Claude or ChatGPT Connectors Directory, or start in the web app. Teams building this into a Claude Code workflow should install the Vibe Prospecting Plugin or add the connection manually:

{
  "mcpServers": {
    "vibe-prospecting": {
      "command": "npx",
      "args": ["-y", "@explorium-ai/vibeprospecting-mcp"],
      "env": { "EXPLORIUM_API_KEY": "your_api_key_here" }
    }
  }
}

Reading the Scoreboard: The Two Numbers That Decide It

Two thresholds settle the verdict. First, if fewer than 60% of your won deals match the profile, the profile is wrong outright. Second, if lost deals match within 20 points of the won rate, the profile is not separating anything. A 75% win match feels great until you notice losses match at 70%. That profile describes your market, and a market is not a customer.

The Verdict Table

What you see	What it means	What to do
Under 60% of wins match	The profile is wrong	Rebuild it from the traits your wins actually share
Losses match within 20 points of wins	It describes the market, not the ideal customer	Rebuild from the per-trait gaps
60 to 80% of wins match, gap of 20 to 40 points	Partly right	Delete the traits with no gap, re-score
80%+ of wins match, losses under 40%	Valid and separating	Keep it, re-test quarterly

Check the Gap Per Trait, Not Just Overall

One strong trait can hide three dead ones inside an overall score. Ask for the win-loss gap on every trait separately: gaps above 20 points are your real profile, gaps near zero are noise dressed up as strategy.

Rewrite the Profile Your Wins Actually Describe

The rewrite is mechanical once the gap table exists: keep every trait with a 20-point or larger gap, delete every trait both piles share, and add the high-gap traits your old profile never mentioned.

What Goes In

• The biggest-gap traits, written in checkable terms: "$10M to $100M revenue," not "mid-market."
• Traits about tools and activity that separated the piles, like a specific cloud platform or an engineering hiring push.
• Walk-away criteria: the traits that over-index in your fit losses.

What Comes Out

• Region and industry lines that matched wins and losses equally.
• Aspirational segments with zero won deals behind them.
• Anything written in words no data source can check.

Before You Hand Targeting to an AI Agent

An AI sales agent does not fix a wrong profile. It scales it. An agent sourcing and messaging thousands of accounts a week off an untested profile produces wrong outreach at a volume no human team could match, and the damage shows up as quietly sagging reply rates that everyone blames on the copy.

• Gate any agent rollout on a profile that passed the fit test with a 20-point-plus gap.
• Feed the agent the same filled-in traits the test used, from the same connection, so "good fit" means one thing everywhere.
• When a quarterly re-test shows a trait decaying, update the agent's targeting the same week.

The 20 Minute Re-Check Every Quarter

Teams that refresh their profile quarterly report a 9.7% higher pipeline creation rate than teams that revisit it once a year or less. The second pass takes a fraction of the effort: your earlier deals keep their filled-in traits, so each quarter you only add the new wins and losses, re-ask for the gap table, and adjust whatever fell below the 20-point bar. If a trait decays two quarters in a row, treat it as market news worth a conversation, not just a settings change.

Your First Fit Test This Week

• Day 1: Export 50 to 100 won deals and a similar pile of lost ones from the last 12 to 18 months, with name, domain, and outcome. Tag price and timing losses so you can set them aside.
• Day 2: Connect Vibe Prospecting in Claude or ChatGPT with a free account and match the export to real company records.
• Day 3: Fill in the traits your profile is written in, previewing 5 sample rows and the cost first.
• Day 4: Ask for match rates on both piles and the gap per trait.
• Day 5: Rewrite the profile around the 20-point-plus traits and put the quarterly re-check on the calendar.

The stakes are simple: every list you build, every sequence you send, and every agent you deploy inherits the profile underneath it. Test it once, and everything downstream gets sharper.

Stop prospecting off a guess. Start free with Vibe Prospecting and run your first fit test this week.