An enrichment engine that cleans the CRM instead of polluting it
Enrichment can degrade a CRM: it overwrites good data with worse, burns credits re-processing the same records, and leaves reps on stale information. A governed HubSpot, Clay, and n8n pipeline enriches raw records, scores them against an ICP, and writes back under strict field-level rules, leaving the CRM cleaner on every run. Built and run live in HubSpot; every number here comes from the live run.
HubSpot
Clay
n8n
Enrichment usually makes a CRM worse, not better
Bad enrichment costs more than missing data: it overwrites good fields with worse ones and burns credits re-buying records you already have. The fix is governance, a system that knows what to write, what to protect, and what to skip, so every run leaves the CRM cleaner than it found it.
Three layers, clean separation of concerns
HubSpot
Owns contacts, companies, deals. The final authority on “truth.” Feeds sales and marketing execution. Never treated as a processing scratchpad.
Clay
A temporary working layer, not a database. It pulls raw records, enriches them (industry, headcount, revenue), validates, scores against the ICP, and decides what actually gets written back.
n8n (self-hosted)
Controls when enrichment runs, what gets written back, how often data refreshes, and what gets blocked. On a $0 stack, n8n also bridges Clay↔HubSpot via API, replacing both HubSpot’s paid workflows and Clay’s paid native sync.
Eight steps from raw record to routed account
Pre-enrichment validation skips already-enriched records, so the system does not burn credits re-processing data it already has.
What gets written, and what gets protected
Empty fieldExisting fieldCritical (domain, name)Derived fieldSignals12 accounts, enriched and scored A / B / C
Demo ICP: B2B software, roughly 50–2,000 employees, sales-led, real revenue traction. Scoring is rule-based and explainable on purpose.
| Company | Industry | Employees | Revenue | ICP |
|---|---|---|---|---|
| Apollo | Software | 504 | $500M–1B | A |
| Lemlist | Software | 247 | $75M–200M | A |
| Instantly | Software | 292 | $75M–200M | A |
| Clay | Software | 1,367 | $500M–1B | A |
| Salesloft | Software | 1,106 | $500M–1B | A |
| Outreach | Software | 1,608 | $500M–1B | A |
| Gong | Software | 2,556 | $500M–1B | A |
| Smartlead | Software | 138 | $5M–10M | B |
| Notion | Software | 6,420 | $1B–10B | B |
| Salesforce | Software | 83,794 | $10B–100B | B |
| Ramp | Financial Services | 2,469 | $1B–10B | C |
| Stripe | Tech / Internet | 15,871 | $1B–10B | C |
The model discriminates as designed: sales-SaaS in the sweet spot rises to A; under-revenue (Smartlead), over-size (Notion, Salesforce), and out-of-industry (Ramp, Stripe) fall to B/C and route to nurture instead of an SDR.
What I’d add, and why it is three systems in one
What I’d add for production
Error handling and retries with alerting · a compliance / suppression check (GDPR/CCPA, opt-outs) before enrichment · per-run credit caps · a closed loop that writes outcomes back to refine the ICP score over time · concurrency-safe locking so the scheduled, inbound, and outbound triggers cannot double-enrich a record.
Why it is three systems, not one
The same engine runs three ways by swapping the trigger: outbound (scraped and signal lists), inbound (form / webinar capture, routed instantly with async enrichment so hot leads do not go cold), and scheduled CRM hygiene (this build). One spine, three deliverables.
The data layer of the stack.
Governed enrichment and ICP scoring that keep a CRM trustworthy as it grows. Have a look at the rest of the stack, or get in touch.