Enrichment Vendor Rotation and Field-Level Provenance (QA)
A checklist with 8 steps: Freeze the client's current CRM export and label it as the baseline before any append runs.
By InnovaAI ResearchPublished
What are the steps?
Enrichment Vendor Rotation and Field-Level Provenance (QA)
- 01
Freeze the client's current CRM export and label it as the baseline before any append runs
Timestamp the file and store it outside the enrichment workspace so a bad batch can be rolled back without reconstructing history from activity logs.
- 02
Score every field by how much a wrong value costs the client
A stale job title wastes a personalization token; a wrong direct dial burns a rep's afternoon and can breach a do-not-call commitment. Rank fields into high, medium, and low consequence tiers.
- 03
Assign each high-consequence field to at least two independent providers
Waterfall the lookup so the second source only fills gaps the first left empty, then flag any record where the two disagree rather than silently picking a winner. Clay's multi-provider waterfall and Cleanlist's aggregation of 15+ sources both support this pattern.
- 04
Run a 200-record control sample through the full stack before touching the live list
Hand-verify the sample against company sites and LinkedIn, then record the observed match rate per field. If direct dial accuracy lands under 80%, renegotiate the field mix before scaling.
- 05
Stamp each enriched field with source name and append date
Field-level provenance is what lets you answer a client's 'where did this number come from' question in seconds instead of an afternoon. It also tells you which vendor to drop when decay shows up.
- 06
Re-verify any record older than 90 days before it enters a new sequence
Job changes and company moves are the main decay drivers. Cognism and Dropcontact both lean on verification passes for this reason; schedule the re-check as a recurring job, not a manual scramble.
- 07
Log the batch outcome and vendor hit rate in the client's QA file
Track match rate, disagreement rate, and bounce rate per vendor per month. Two consecutive months of decline on one source is the trigger to rotate it out.
- 08
Escalate any record that fails verification twice to a human reviewer
Do not let an automated retry loop keep guessing. A named person decides whether the record is suppressed, manually corrected, or returned to the client as unverifiable.