Assign every company a numeric score from observable signals, then bucket into tiers. ColdIQ models run 80–100 points across 4–6 signal groups. The math is local; this skill notes which ColdIQ endpoint provides each scored data point so the model runs on fresh data.
ColdIQ Marketplace Endpoints (data sources for scored fields)
| Scored data point | Method | Path | Credits | Endpoint ID | Notes |
|---|---|---|---|---|---|
| Firmographics (size, revenue, industry, funding) | POST | /v1/limadata/enrich/company |
1 | limadata.enrich.company |
One call for most groups |
| Large-scale company list | POST | /v1/ai-ark/companies |
per result | ai_ark.companies.search |
Build the list to score |
| Tech-stack signal | POST | /v1/builtwith/domain |
flat | builtwith.domain |
Complexity/specialization group |
| Funding / PE backing | GET | /v1/signalbase/funding-signals |
unknown | signalbase.funding_signals |
Ownership/bonus groups |
Score from enriched data — don't pay to enrich rows you'll DQ. Filter obvious DQs first, then:
→ POST /v1/limadata/enrich/company · 1 cr · limadata.enrich.company
Standard tier thresholds
| Tier | Score | Action |
|---|---|---|
| Tier 1 | 65+ | Top priority, send first, best personalization |
| Tier 2 | 50–64 | Strong fit, second wave |
| Tier 3 | 35–49 | Marginal, volume plays only |
| DQ | <35 | Do not send |
Common signal groups
- Scale (10–25): 10k+ → 25, 5k–9,999 → 20, 2k–4,999 → 15, 500–1,999 → 10, 50–499 → 5, <50 → 0.
- Revenue (15–20): $50M–$500M → 20 (sweet spot), $25–50M → 15, $500M–1B → 10, $10–25M → 8, $1B+ → 5.
- Industry (10–15): core → 15, adjacent → 10, stretch → 5, unknown → 5, excluded → −100 (auto-DQ).
- Complexity / specialization (15–30): the client-specific signal (multi-location count, market count, competitor/complementary tech, digital maturity).
- Ownership / PE (10–20): PE-backed → 20, PE subsidiary → 15, VC/growth → 10, unknown → 5, founder/non-profit → 0.
- Bonus (5–10): franchise HQ, cash-flow keywords, rapid growth, recent funding, relevant hiring.
Data cleanup before scoring
BOGUS_NAMES = {"local my business", "google ai plugin", "auto-entrepreneur"}
def is_bogus(name):
l = name.strip().lower()
return l in BOGUS_NAMES or any(s in l for s in ["follow us", "test account"])
# Remove slug franchises (keller-williams-realty-dpr), flag extreme values, dedup by name (keep top score).
Python scorer pattern
def score_company(row):
total, breakdown = 0, []
for label, fn, field in [("Scale", score_scale, "# Employees"),
("Revenue", score_revenue, "Annual Revenue"),
("Industry", score_industry, "Industry"),
("Custom", score_custom, None)]:
pts, reason = fn(row if field is None else parse_number(row.get(field, 0)))
total += pts; breakdown.append(f"{label}: {pts} ({reason})")
if any("EXCLUDED" in b for b in breakdown): return -1, "DQ", breakdown
tier = "Tier 1" if total>=65 else "Tier 2" if total>=50 else "Tier 3" if total>=35 else "DQ"
return total, tier, breakdown
# load → clean(is_bogus) → score → dedup by name(keep highest) → sort desc → export CSV with Score/Tier/Breakdown
How to build a new model
- Pick 4–6 signal groups summing to 80–100. 2. Set thresholds. 3. Write scoring rules per signal.
- Clean data. 5. Run on the enriched export. 6. Generate per-tier CSVs. 7. Send a sample to the client for approval. 8. Re-run when new enrichment data arrives. Dedup first (list-dedup) so you never score a company twice.