enrichment

name	enrichment
description	Step 5 of the ICP playbook. Enriches the company list with firmographics, technographics, hiring patterns, funding, and growth signals, then finds the multi-attribute combinations that recur within each persona. Routes to Clay if available, otherwise Apollo, ZoomInfo, Cognism, Crustdata, Ocean, or Harmonic. Use after persona clustering and trigger tagging.

Step 5 — Enrich and find attribute combinations

The article's central claim about this step: single filters don't tell you much. Combinations are the signal.

Routing

Read ./icp-output/stack.json. Branch on stack.enrichment.tool:

• clay → load clay.md
• anything else → load fallbacks.md and pick the matching adapter

Unlike the call-intel step, no warning is needed when Clay is absent. The article recommends Clay but the kit treats every alternative as first-class as long as the user can supply API access or a connector.

What to enrich

For each named account that appears in ./icp-output/calls/ or ./icp-output/personas/:

• Industry (use both NAICS/SIC and a softer self-described industry — they often disagree)
• Headcount + 6-month / 12-month growth
• Stage + last funding round + amount + investors
• Geography (HQ + presence)
• Tech stack — CRM, sequencer, data warehouse, RevOps tooling, AI tools, sales enablement
• Hiring patterns — open roles by function, hiring spikes
• Recent exec changes — new C-level, new VP Sales/RevOps, new CMO
• Public signals — recent news, M&A, product launches
• Buyer-intent if available — visit/page-level signals from your enrichment provider

Save raw enrichment results to ./icp-output/enrichment/companies.csv (or .json). One row per company.

Find the combinations (the central deliverable)

For each persona, look at the 5–20 enriched accounts that map to it. Compute:

• Which 2–4 attribute combinations appear in ≥60% of accounts for that persona but in <25% of accounts outside that persona?
• Specifically test combinations like:
  • title × seniority × headcount
  • tech_stack × industry × funding_stage
  • hiring_signal × growth_rate × geography

Output

Write ./icp-output/enrichment/company-attributes.md:

# Per-persona attribute combinations

## <Persona name>

### Runnable filter (combinations)

- **Title:** "RevOps", "Revenue Operations", "Sales Operations", "Sales Ops", "GTM Operations"
- **Title exclude:** "Analyst", "Associate", "Coordinator", "Intern"
- **Seniority:** Director, VP, Head of, Senior Manager
- **Headcount:** 201–2,000
- **Industry:** SaaS, Software Development, IT Services, Cloud Computing
- **CRM:** Salesforce or HubSpot
- **Growth signals:** Headcount growth >15%, recent Series B/C/D, new CRO hire in last 6 months
- **Geography:** US primary, UK/EMEA secondary

### Why this combination

Recurs in <X>% of evidenced accounts in this persona, <Y>% in others. Pulled from <N> accounts.

### Single-filter false friends (do not use these alone)

- "RevOps" alone returns <bigger noisy set>
- "Series B+" alone returns <bigger noisy set>

### Sample evidenced accounts

- <Account> — slug, <call-id>, fit-score
- ...

Repeat per persona.

Failure modes

• A filter that includes only one industry but the actual buyer set crosses two — under-fit.
• A filter so broad it returns 50,000 accounts in Clay/Apollo — over-fit on intent, under-fit on combinations. Tighten.
• A filter that loses 80% of evidenced accounts when run — you optimized for breadth over recall. Widen, then re-tighten on a different axis.

Aim for filters that, when run, return 500–5,000 target accounts per pocket. That's a workable list for outbound.

State update

{
  "step": "5",
  "step_status": "completed",
  "personas_with_filters": <count>,
  "enrichment_coverage": <pct of accounts enriched>
}