PredictLeads Lookalikes (account discovery)

Pulls up to 50 companies similar to a seed domain via PredictLeads' similar_companies endpoint. Returns ranked results with a similarity score (0–1) and (when available) a sentence explaining why each is similar. Cheapest possible discovery: 1 credit per seed domain.

When to use

• Expanding a target list from a known anchor account ("find more like Stripe")
• Sanity-checking that a prospect fits an existing client's pattern
• Building a watchlist of competitors for monitoring
• First step of prospect-discovery-pipeline (but use that skill if outreach is downstream)

Don't use when: you need contacts/CMOs at the lookalikes (use prospect-discovery-pipeline); you want signal data on a single known company (use predictleads-signals).

Quick reference

# 50 lookalikes from a seed domain (1 credit)
npx tsx src/cli/index.ts signals:similar --domain stripe.com --limit 50

# Tighter list (still 1 credit per call)
npx tsx src/cli/index.ts signals:similar --domain linear.app --limit 20

# Read back from local SQLite (no API call)
npx tsx src/cli/index.ts signals:show --domain stripe.com --type similar

Cost

Always 1 credit per seed domain regardless of --limit. Result rows land in company_signals with signal_type='similar_company' and stay cached for 7 days.

Output shape

Each result is a domain + score + reason:

deel.com (score=0.85)
remote.com (score=0.849)
oysterhr.com (score=0.848)  — Both provide global EOR services for distributed teams.

Reasons are populated for some seed domains and not others (PredictLeads inconsistent). Score above 0.80 typically means strong match in their model.

Common pitfalls

• Megacaps in results: PredictLeads returns market peers including SAP, Microsoft, Google for many B2B SaaS seeds. Filter by hand or via Crustdata company_identify before going further.
• Multiple seed merge: if you run two seeds (stripe.com and linear.app), results are stored separately by source domain. Dedupe in code or use prospect-discovery-pipeline which handles the merge.
• similar_company rows have no event date: they're a lookup, not an event. Don't sort by event_date; sort by score (in payload) or position.

Required env

PREDICTLEADS_API_KEY + PREDICTLEADS_API_TOKEN in ~/.gtm-os/.env. See TEAM_SETUP.md.