Mine your won-deal notes for the patterns — firmographics, buying triggers, decision paths, objections that got overcome — and compress them into an ideal-customer-profile one-pager that sales and marketing both actually use, grounded in deals you won rather than customers you wish you had.
Tier: Core · Hands-on time: ~35 minutes · Tools: Claude, ChatGPT, Google Docs
Most ICP documents are written forward — someone describes the customer they'd like — and quietly ignored. The credible version is written backward from evidence: the deals that actually closed, what those companies had in common, what triggered them to buy, and what nearly stopped them. That analysis is a pattern-extraction job over messy notes, which is exactly the work a model does well under supervision.
And when not to:
Assemble the won-deal corpus (8 minutes). For each won deal from the last 12–18 months, gather what exists: company name, size, industry, deal size, sales cycle length, plus every scrap of notes — discovery summaries, objection records, close notes. Paste into one working doc, one deal per block, numbered. Include two or three prominent lost deals at the end, labelled LOST — contrast sharpens patterns.
Run the pattern-extraction pass. Paste the corpus with this prompt:
You are a revenue analyst. Below are numbered notes from our WON
deals (and a few labelled LOST for contrast). Extract patterns —
only what the notes support:
1. Firmographics: what the won companies share — size range,
industry, structure, tooling. Cite deal numbers for each claim.
2. Buying triggers: the events that started each purchase (new
leader, audit, growth, tool failure...). Group and count them.
3. Champion & decision path: who championed, who signed, who else
was involved — as a pattern with counts.
4. Objections overcome: recurring concerns and, where notes show
it, what answer worked.
5. Won-vs-lost contrast: what the LOST deals had that wins didn't,
or lacked that wins had.
6. Anti-patterns: any signals that appear only in losses.
Rules:
- Every claim cites deal numbers (e.g. "8 of 12 wins: #1, #3...").
- If the notes are too thin to support a section, say "insufficient
evidence" — do not fill gaps with general B2B wisdom.
- Flag any pattern driven by fewer than 3 deals as WEAK.
DEAL NOTES: [PASTE]
A good output reads like this:
Buying triggers (12 wins)
- New finance leader in first 6 months: 5 deals (#2,#4,#7,#9,#12)
- Failed audit or near-miss: 4 deals (#1,#3,#8,#10)
- Outgrew spreadsheets at ~150 employees: 2 deals (#5,#11) — WEAK
- Unclassified: 1 deal (#6 — notes don't state a trigger)
Won-vs-lost contrast
- Both LOST deals had no internal champion below the CFO; 11 of
12 wins had an ops-level champion first.
Audit the citations (7 minutes). Spot-check the claims: open the cited deals and confirm the notes actually say what the pattern claims. Kill anything the notes don't support — the one-pager's authority rests entirely on "this is what our deals show", so one stretched claim poisons the page. Respect the WEAK flags: keep them, labelled, or cut them.
Compress into the one-pager. Same conversation:
Turn the verified patterns into a one-page ICP document:
- Profile: 3-4 lines describing the company we win (firmographics).
- Buying triggers: top 3, with counts, ordered by frequency.
- Who to find: the champion profile and typical decision path.
- Objections to expect: top 3, each with the answer that worked.
- Disqualifiers: signals from losses that predict a bad fit.
- Evidence line: "Based on N won deals, [date range]."
One page hard limit. Plain language a new rep understands day one.
Move it into Google Docs, and keep the pattern-extraction output as a linked appendix — the one-pager makes claims; the appendix holds receipts.
Pressure-test with the people who closed the deals. Walk it past two salespeople who won some of those deals: "What's wrong on this page?" Their corrections — a trigger the notes never captured, a champion pattern that's really one heroic rep — are the last mile of accuracy. Then put it to work: pb-25's {{trigger}} and {{pain_hypothesis}} columns should be filled from this page's triggers list, which is the moment the ICP stops being a document and starts being targeting.
Won-deal notes contain customer names, deal sizes, and commercial terms. Use your organisation's approved AI tenant, anonymise customer names in the corpus where the analysis doesn't need them, and keep the one-pager internal — a document listing your customers' triggers and objections is competitive intelligence about you.
Date-stamp the ICP and re-run the mining every quarter as wins accumulate. The most valuable output over time isn't the profile — it's watching the triggers list change, because that drift is your market telling you where it's moving next.