Engineer your ICP from win/loss data

Founder / head of growth / PMM / RevOps · Assemble in ~20 min; first lift-scored signal report in <30 min once your CRM export is ready.

GTMEdit flowon X◆ new (unverified) 89

step 1connector

Export CRM win/loss data

⚡ CRM (HubSpot / Salesforce / Sheet)

step 2ready

Discover differential signals between won and lost accounts

niche-signal-discovery12 alts

step 3partial

Enrich signal report with competitive landscape

competitive-teardown

step 4ready

Build the ICP definition

sales-icp1 alts

step 5ready

Build scored synthetic buyer personas

buyer-persona1 alts

step 6ready

Score the account pool against the ICP

ai-sales-intelligence-and-outreach12 alts

step 7ready

Generate and validate synthetic persona test cases

buyer-persona-generator1 alts

step 8sub-recipe

Publish ICP + personas to outbound

↘ ai-sdr

Did this recipe work for you?

What you get

▸Lift-scored differential signal report (won vs. lost accounts) with ranked signals by statistical strength
▸Structured ICP definition document with firmographic criteria, triggers, objections, and messaging angles
▸3-5 scored synthetic buyer personas ready to load into outbound or product workflows
▸Scored account list ranked by ICP fit, ready to hand to an SDR or outbound campaign

tools: CRM (HubSpot free tier or existing Salesforce seat) — no additional SaaS required for the analysis itself

tokens: ~$5-15 per 100-account win/loss batch (signal scraping + persona generation)

≈ $10-30/mo at one ICP refresh cycle per month

What stays yours

· Pull and clean the CRM export (field mapping, deduplication) — the agent cannot access live CRM without an integration
· Validate the top 3-5 lift signals against your own memory of why deals were won or lost — the model cannot distinguish a spurious correlation from a causal signal without your context
· Approve the ICP definition and persona set before publishing to outbound — a bad ICP at this stage poisons every downstream step

The problem

When I need to know who actually buys and churns, I want an agent to engineer and continuously refine my ICP from closed-won/closed-lost CRM data plus external signals and build scored synthetic personas, so I target and qualify the right accounts on data instead of a stale gut-feel deck.

Cybernetic layers

Memorypartial

The ICP definition and scored signals should be persisted as versioned files (e.g., icp-v1.md, lift-signals-v1.json) so future runs diff against prior versions and you see what changed. No harvested memory artifact automates this — use a local folder or Notion page. Re-run step 2 monthly or after 20+ new closed deals.

Background-schedulingbyo

The recipe is batch-shaped (one session per ICP refresh cycle — recommended monthly or quarterly). True continuous refresh (auto-trigger on CRM close events) requires a webhook + scheduler not yet in the library. A Claude Code cron scheduled task can approximate it.

Harnessready

Claude Code is the runtime. Install niche-signal-discovery, buyer-persona, sales-icp, ai-sales-intelligence-and-outreach, competitive-teardown, and buyer-persona-generator into ~/.claude/skills/ and invoke in order.

Click any step to see what fills it — and the ranked alternatives you can swap in.