Skip to main content
A reliable B2B enrichment waterfall is a controlled sequence of providers, validation rules, and fallback paths. It should return a field only when the field meets an explicit quality rule, retain the supplying provider and cost, and send failures to a review queue instead of silently writing weak data into the CRM.

​
What an email enrichment waterfall does

An email enrichment waterfall looks up a person’s work email across declared sources and validates candidate results. It does not send email. Sending infrastructure, inbox placement and permission to contact someone are separate decisions. Follow the Claude Code implementation guide to run a checked Play, or start with the quickstart. Define a usable record before comparing costs; the enrichment evaluation guide includes a worked acceptance example.

​
Key takeaways

  • Rank providers with a labeled sample from the target segment, not a vendor’s aggregate match-rate claim.
  • Stop only when a field passes its acceptance rule. Preserve the provider, evidence, validation result, and complete cost.
  • Separate missing, invalid, conflicting, and provider-error results so each state has a defined retry or review path.

​
Who this is for

RevOps and GTM engineering teams that need better contact or company coverage without accepting lower accuracy.

​
Decision guide

​
Operating method

  1. Define the record type, required fields, acceptable values, and a clear owner for rejected results.
  2. Normalize names, domains, CRM IDs, and country before calling any provider. Deduplicate first.
  3. Call the lowest-cost appropriate provider, validate the response, then use the next provider only for missing or rejected fields.
  4. Store provider, request time, response status, cost, validation result, and workflow version with every proposed field.
  5. Write only approved changes, then read the CRM record back and compare it with the requested update.

​
Implementation details

  • Normalize the input before calling a provider: lowercase domains, preserve the original record ID, and separate person and company identifiers.
  • Stop the waterfall when the acceptance rule passes, not merely when a provider returns a value. Record which provider supplied the accepted result.
  • Treat missing, invalid, conflicting, and provider-error results as different states. Each state needs its own retry or review rule.
  • Measure cost per accepted field and coverage on a fixed holdout set. Provider response count is not a quality metric.

​
Common failure modes

  • Treating an API response as correct without validation.
  • Overwriting a known-good CRM value with a lower-confidence value.
  • Counting missing results as quality failures without separating provider coverage from validation failure.
  • Running the whole waterfall again when one field changed.

​
Questions teams ask

​
Which provider should be first?

Use a held-out sample from the target segment. The first provider should have the best accepted-field coverage at the lowest complete cost for that field and geography.

​
How do we avoid duplicate contacts?

Resolve identity before enrichment with stable CRM IDs, normalized email and domain, and a documented match policy. Send ambiguous matches to review.

​
Can we bring our own provider keys?

Yes. Keep the provider contract, request receipt, cost, and credential ownership separate from the orchestration logic.

​
What should happen when providers disagree?

Do not select a value silently. Apply a documented precedence rule or send the field to an exception queue with the evidence.

​
How often should we rerun enrichment?

Use field-specific freshness rules. Job changes and phone data need different refresh intervals from company firmographics.

​
Methodology and sources

This guide is based on public product information and a production-workflow model: an input, a validated output, an accountable owner, a failure path, and a measurable unit cost. Recheck product and pricing claims before purchase. Apollo integration · People Data Labs integration · Crustdata integration · CLI installation guide GTM provider consolidation · HubSpot enrichment workflows · BYOK data enrichment

​
Start with one workflow

Take the workflow you’d least like to do by hand. Run it with examples of what good looks like. Let the agent iterate to the optimal solution.