ICP discovery from closed-won data: a 30-day rebuild for revenue teams

Your ICP was defined in a board meeting 18 months ago. A founder said "mid-market SaaS, VP of Sales, 200-500 employees." It went on a slide. SDRs started dialing. Nobody checked whether the companies you actually close look anything like that profile. Two years later, the team is still using the original ICP, and the slide deck is the reason your outbound is hitting the wrong accounts.

1. W1
   WEEK 1 Data prep

   Pull 18-24 months of closed-won. Reconstruct the trigger field for 60-80% of deals. The trigger work is what makes the rebuild work.

   • Industry & size
   • Buyer title & tenure
   • Tech stack at purchase
   • Sales cycle
   • Trigger
2. W2
   WEEK 2 Pattern analysis

   Look for clustering in size, vertical, buyer tenure, tech signals, and triggers. Most rebuilds surface 2-3 real segments, not the 4-6 the legacy ICP claimed.

   • Segment clustering
   • Trigger frequency
   • Cycle correlation
   • Buyer tenure
3. W3
   WEEK 3 Lookalike disqualification

   Identify what almost-fits but never closes. The disqualification list is what stops outbound from chasing accounts that look right but never convert.

   • Lost-deal demographics
   • Missing triggers
   • Disqualification rules
4. W4
   WEEK 4 Validation & rollout

   Test against lost deals (should predict losses), open pipeline (should track success), and a peer sample. Roll into CLAUDE.md + scoring rubric.

   • Lost-deal backtest
   • Open pipeline correlation
   • CLAUDE.md update
   • Scoring rubric live

Most aspirational ICPs target one tier above the actual buyer. The rebuild surfaces the real segment, the real title, and the trigger signal nobody had named. Output feeds the CLAUDE.md scoring rubric directly.

Real ICPs come from real win data, not founder intuition. This post is the 30-day rebuild we run for revenue teams whose outbound is underperforming, whose AE conversion is uneven, or whose sales cycle keeps stretching for reasons no one can name. The output is a working ICP plus a scoring rubric that feeds the outbound subagent's signal scoring.

Why the legacy ICP is almost always wrong

Three structural reasons.

It was aspirational. The company wanted to sell to enterprise, so the ICP was written for enterprise. The actual buyers turned out to be mid-market with strong product-market fit, but the marketing positioning still talks to enterprise, the SDR team still chases enterprise, and the close rate at enterprise is 4% while mid-market is 22%. The ICP shaped the team's effort against the wrong target.

It wasn't validated. The ICP was set, the team executed against it, but no one ever checked whether the companies in the closed-won column looked like the ICP. After 50 closes, the list of buyers usually looks substantially different from what the ICP slide claims.

It missed the trigger. Every real buyer has a reason to buy now. The ICP captured demographics (size, vertical, title) but not the behavioral or contextual trigger that distinguishes "this prospect will buy in 90 days" from "this prospect looks like a buyer but never closes." The trigger is the highest-leverage piece of the rebuild.

The 30-day sequence

Week 1: data preparation

Pull every closed-won deal from the last 18-24 months. For each, capture:

• Company name, industry, size at purchase, geography
• Buyer title and tenure (how long had they been in the role)
• Tech stack at time of purchase (BuiltWith snapshot or self-reported)
• Source: how did they enter the funnel
• Sales cycle length
• Deal size
• Trigger: what changed in their world that opened the buying window

The trigger field is usually missing from CRM. Reconstruct it from won-deal notes, post-sale conversations, or a quick Slack to the AE who closed it. The trigger is what makes the rebuild work; don't skip it.

Most teams discover the trigger field for 60-80% of closes; the remainder get tagged "unknown trigger" and excluded from the trigger analysis (still useful for demographic patterns).

Week 2: pattern analysis

Look for clustering. The questions that tend to surface real patterns:

• What size company tends to close fastest? Slowest?
• Which industry verticals are over-represented in closed-won versus the overall pipeline?
• What buyer tenure correlates with close rate? (New-to-role buyers often close faster than long-tenured ones)
• What tech stack signals predict fit? (Companies running specific tools often have the pain we solve)
• What triggers appear most often? Hiring patterns, leadership changes, funding rounds, regulatory changes, replacement of a competitor

The output is a draft ICP with named segments and triggers. Usually 2-3 segments emerge as the real ICP, with one being dominant. The legacy ICP usually claimed 4-6 segments; the real data shows 2-3.

Week 3: lookalike disqualification

Equally important: what almost-fits but doesn't actually close. These lookalikes are where the team wastes the most outbound spend.

Look at lost deals and stalled opportunities. What demographic profile do they share with closed-won? What signal is missing? Often the difference between "buyer" and "lookalike" is one specific trigger or one specific stack signal that the team has been treating as optional.

The disqualification list is the deliverable. "Companies in segment X that lack signal Y are not ICP, regardless of how qualified they look on title and size." This list is what stops the outbound team from chasing accounts that look right but never close.

Week 4: validation and rollout

Test the new ICP against:

• The last 12 months of lost deals. The new model should predict losses (lost deals should score out-of-ICP). If lost deals score as in-ICP, the model is too generous and needs refinement.
• Current open pipeline. Deals scoring in-ICP should be progressing well. Deals scoring out-of-ICP should be the ones reps are struggling with. If the correlation breaks, the model needs work.
• A blind sample of 20 known-lost deals from peers in your category. If your model would have warned you off these (predicted out-of-ICP), it's tracking real signal.

After validation, the new ICP rolls into the team CLAUDE.md, the lead scoring rules, and the outbound subagent's signal scoring rubric. We use the structure described in the CLAUDE.md template.

What changes after the rebuild

Three operational shifts within 90 days of rolling out the new ICP.

Outbound efficiency rises. The same team produces more meetings because they're targeting accounts that actually convert. Forward rate on outbound usually doubles in the first 60 days because the trigger anchoring is sharper.

Sales cycle compresses. In-ICP deals close faster than the previous baseline because the trigger work means you're reaching prospects in their buying window, not before it. Average cycle length drops 15-30%.

Lookalike spend stops. The team stops working accounts that look qualified but never close. The freed capacity goes to in-ICP accounts. Win rate rises noticeably even though total pipeline volume drops.

Where this fits in a broader rebuild

ICP discovery is the second phase of the 90-day PE portco rebuild. Pipeline truth-finding comes first (you need to know what's actually in the pipeline before you redefine the target), ICP discovery comes second, system buildout comes third. The order matters; building outbound subagents against the wrong ICP wastes the build.

For non-PE contexts, the same sequence applies. Audit before you redefine. Redefine before you automate.

The honest read

Most ICPs are wrong. The teams running the same ICP for two years without a refresh are almost certainly mistargeting. The 30-day rebuild is short enough that there's no operational excuse to skip it, and the downstream impact (forward rate, sales cycle, win rate) shows up within a quarter.

The piece teams resist is the trigger work. Adding a "what changed in their world" field to closed-won analysis is unsexy. Reconstructing the trigger for 60+ deals takes time. Skip the trigger work and the ICP rebuild is just demographic clustering, which is half the value.

We run ICP discovery as a fixed-fee 30-day engagement, with the deliverable being the ICP document, the scoring rubric, the disqualification list, and the trigger taxonomy. The output goes into your CLAUDE.md so the outbound subagent can score against it the day after the engagement closes.

What surprises teams during the rebuild

Three patterns show up consistently across rebuilds that surprise the founders or CROs commissioning them.

The aspirational segment is dragging down win rate. Teams chasing enterprise often discover their close rate at enterprise is 3-5% while their close rate at mid-market is 18-25%. The aspirational segment is consuming sales capacity and producing the worst returns. The fix is unsexy: stop chasing it. Most teams resist this because the marketing positioning is built around the aspirational segment.

The trigger isn't what marketing thinks it is. Marketing usually claims the trigger is a content asset, an inbound lead form, or a webinar. The real trigger from closed-won data is almost always something else: a hire, a regulatory change, a leadership transition, or a competitor's product issue. The assets that marketing produced may have been adjacent to the buying moment but rarely caused it.

The buyer title is wrong by one or two layers. The legacy ICP says "VP Sales." The closed-won data shows the actual decision-maker was the CRO or the Director of Sales Operations. The "VP Sales" sat in the conversation but didn't move the deal. Re-targeting the right title compresses sales cycles by 20-40% on the affected segments.