Skip to main content

ICP playbook

How to Validate Your ICP Using Closed-Won Deal Data

TR
Tom Regan·8 min read·Updated
Quick Answer
Validate your ICP with closed-won data by tiering your last 20 to 40 wins, extracting the firmographics, trigger events, and buying-committee patterns they share, then confirming those same patterns are rare in your closed-lost deals. A pattern that appears in wins and losses alike is noise, not a predictor of who will actually buy.

Most ICPs are written before the company has enough wins to know who it actually sells to. They are aspirational: the revenue band the founders want, the logos that would look good on the site, the industries the board keeps mentioning. Then a year of deals closes, and nobody goes back to check whether the profile survived contact with the market. Validating your ICP with closed-won data is that check. It replaces the ICP you assumed with the one your revenue proves, then feeds the result into the Pain-First ICP scoring framework so pain signals carry the weight and firmographics stay a filter.

Why validate your ICP against closed-won data?

Because an unvalidated ICP quietly sends your best reps after the wrong accounts. The profile feels right, so it never gets questioned, and every downstream decision, from targeting to scoring to routing, inherits its errors. Closed-won data is the only unbiased record of who your product actually earns money from. When B2B buyers now define their own solutions and 61% prefer a rep-free buying experience, the accounts that reach a closed-won stage are self-selecting on real fit. That makes your win list the highest-signal dataset you own for who to chase next.

Cite This

73% of B2B buyers actively avoid suppliers who send irrelevant outreach, and 61% prefer a rep-free buying experience. Targeting built on a closed-won-validated ICP is how a team stays relevant to the accounts that actually buy, instead of spraying a firmographic list the buyer has already learned to ignore.

Gartner B2B buyer survey, as reported by Demand Gen Report

How do you validate an ICP with closed-won data?

Four steps: tier your wins, extract the shared pattern, test it against your losses, then re-score your target list. The whole loop is a weekend of focused work if your CRM data is clean, and it is the cheapest high-leverage change most GTM teams can make.

1. Tier your closed-won deals

Pull your last 20 to 40 closed-won deals and sort them into tiers by the outcomes you actually want more of: fast to close, high ACV, low churn risk, strong expansion. Do not average all wins together. A deal that took nine months, discounted 40%, and churned in year one is not the customer you want to clone, even though it counts as a win. Validation starts by deciding which wins are worth repeating.

2. Extract the shared pattern

For your top tier, record the firmographics (size, stage, industry, tech stack), the trigger event that preceded the deal (funding, a new executive hire, a tooling migration, a metric downturn), and the buying-committee shape (who championed, who signed). Use Amplemarket to enrich every closed-won account in one pass, so the firmographic and technographic attributes surface without hand-collating CRM fields. Official Artemis GTM partner. Affiliate link. The pattern you are looking for is the set of attributes that shows up again and again across the top tier.

3. Test the pattern against closed-lost deals

This is the step teams skip, and it is the one that makes validation real. Take each attribute that looked like a qualifier in your wins and check how often it also appears in your closed-lost deals. Keep only the attributes that are common in wins and rare in losses. Everything else is descriptive, not predictive, and treating it as a qualifier is how a plausible ICP still fills the pipeline with deals that never close.

4. Re-score and re-route your target list

Rebuild your account and lead scores on the validated attributes, then re-rank your active target list. Most teams discover a meaningful slice of their current pipeline never matched the real pattern, and an unworked segment did. Feed the validated profile into your Pain-First ICP score so pain signals carry the weight and firmographics act as the filter, not the other way around.

What patterns in closed-won deals actually predict a win?

The predictive patterns are almost always about timing and pain, not static company attributes. Firmographics tell you an account could buy; trigger events tell you it is ready to. When you tier and compare, the same three pattern types tend to separate winners from look-alike losers.

Pattern typeWhat to recordWhy it predicts
Trigger eventFunding, exec hire, migration, metric downturn in the 90 days before the dealMarks the moment the pain became urgent and budgeted
Buying-committee shapeWho championed, who blocked, who signedRepeatable committees mean a repeatable sales motion
Acute pain matchThe specific problem the buyer named in discoveryNamed pain, not assumed fit, is what closed the deal

Static firmographics still matter, but as a disqualifying filter rather than the primary signal. Two identical-looking accounts convert differently because one is in the trigger window and one is not. As Harvard Business Review argued in The End of Solution Sales, buyers who define their own solutions reward sellers who understand the specific problem they are in, not sellers who lead with a generic company-fit pitch. Your closed-won data is where that specific problem is recorded.

How do you turn a validated ICP into a working score?

A validated pattern is only useful once it runs automatically against every new account. That is a scoring and routing job. The Lead Scoring agent ($349) turns a validated closed-won pattern into a live two-model score: it pulls your closed-won and closed-lost data, fits a firmographic model against your real wins, layers behavioral intent on top, sets thresholds against your conversion data, and wires the BDR queue routing so the validated ICP operates inside your CRM instead of sitting in a deck. The deeper build underneath it, mining win patterns and running customer interviews to define the profile from scratch, is the ICP Definition agent.

Related reading: firmographic vs pain-based ICP shows why fit scoring alone caps conversion, and how to build a lead scoring model covers the fit-plus-intent scoring the validated profile feeds.

Frequently asked questions

What does it mean to validate an ICP with closed-won data?

It means checking your Ideal Customer Profile against the accounts that actually bought, rather than the accounts you wish would buy. You tier your closed-won deals, extract the firmographics, trigger events, and buying-committee shape they share, and then confirm those patterns are absent from your closed-lost deals. A criterion that appears in both wins and losses is not a qualifier; it is noise your reps have been treating as signal.

How many closed-won deals do you need to validate an ICP?

Enough to see a repeated pattern rather than a coincidence. In practice, tiering your last 20 to 40 wins is usually enough to spot the firmographic and trigger clusters that recur, and to notice which supposed qualifiers do not. If you have fewer than 20 wins, weight recent deals more heavily and treat the output as directional until more data accrues; if you have hundreds, sample your best-fit and highest-value cohort rather than averaging everything together.

Why test the ICP against closed-lost deals, not just closed-won?

Because closed-won data alone tells you what your winners look like, not what separates them from the accounts that looked identical and still walked. If Series B fintechs appear in 70% of your wins but also 70% of your losses, company stage is not predicting anything. The signal lives in the attributes that are common in wins and rare in losses. Skipping the loss side is the single most common reason a validated-looking ICP still misfires in market.

How do I turn a validated ICP into a working lead score?

Once your closed-won pattern is validated, the Lead Scoring agent ($349, at /agents/lead-scoring/) builds it into a two-model score: it pulls your closed-won and closed-lost data, fits a firmographic model against your real wins, layers behavioral intent on top, sets thresholds against your conversion data, and wires BDR queue routing so the validated ICP runs inside your CRM instead of living on a slide. The deeper win-pattern and interview build is the ICP Definition agent (/agents/icp-definition/) underneath it.

How often should you re-validate your ICP against closed-won data?

Re-validate quarterly, and immediately when one of four signals appears: sales cycles lengthening without cause, win rate against a specific competitor declining, churn concentrating in one segment, or average deal size dropping. Win patterns drift as your product, pricing, and market move, so an ICP validated 18 months ago is describing customers you may no longer win. A quarterly re-tier keeps the profile honest.

Sources & references

Run this play in your own stack

Read the guide, then install the engine.

The Artemis AI GTM Engineer prices this leak in dollars before it recommends anything, then builds the fix with you inside your own Claude. See how an agent installs and buys, or start with the free audit that prices all seven leaks.

Your go-to-market needs real systems.

Install the free AI GTM Engineer and get a full GTM audit in one session.

✓ Free, no account✓ Agents from $349✓ Run by Claude
Cookie categories