n8n is ideal for teams that want full API control, custom scoring math, bulk processing, and strict overwrite rules.
What you’ll build
A production pipeline with:
- HTTP Request to Lusha Company/Person endpoints
- Function nodes for scoring and normalization
- Conditional overwrite (only write to CRM if the field is empty, or Lusha confidence ≥ 95)
- Bulk processing via list chunking
- Signals webhook that applies a temporary score multiplier
Nodes overview
Trigger: Webhook (real time) or Cron (batch).
HTTP Request → Lusha /v2/company (domain).
HTTP Request → Lusha /v2/person (email/name + company).
Function: normalize & score
Map industries to canonical values
Seniority map (IC/Manager/Director/VP/C)
Score calculation (fit + trust + timing)
IF: email_confidence ≥ 90 → proceed; else → write to “validation_queue” (BigQuery/Sheet).
IF: field overwrite rules (only overwrite if empty, or confidence ≥ 95).
CRM write (Salesforce/HubSpot) + Slack alert.
Signals Webhook: on funding/job change → update score_boost field with an expiry date.
Example function (pseudocode)
Bulk strategy
- Ingest CSV → Split In Batches → call Lusha Bulk Person/Company (if available) or loop single requests with Wait for rate-limit safety.
- Use Error Trigger and retry logic on 429/5xx.
Conditional overwrite policy
- If
crm.emailis empty → write Lusha email (any confidence ≥ 90). - If
crm.emailexists → overwrite only iflusha_confidence ≥ 95. - For phone fields, keep primary stable; write new numbers into secondary fields with a
verified_source=lusha.
Observability & audit
- Write a compact audit log row per record:
{lead_id, score, confidence, timing_boost, overwrite_fields[], ts}to a warehouse or log table.
Rollout checklist
- Dry-run against a sandbox org.
- Freeze picklists; deploy normalization first.
- Start with read-only CRM creds for a week, flip to write once QA passes.
“n8n gave us the control we needed: real scoring logic in a Function node, safe overwrites, and a signals webhook that bumps scores with an expiry. It feels like a proper RevOps service, not a bunch of disconnected zaps.”
