Written specification · August 2026
List building and enrichment specification
The exact pipeline I use to turn a directory scale source into a clean, deduplicated and verified list of target accounts, and why the step order is the whole game: filter before you enrich, so every credit you spend lands on a row that survives to the export.
The pipeline, in execution order
SOURCE pull wide from the ICP source (directory, Apollo search, web set)
|
NORMALIZE casing, domains without www, phone and geo formats, title synonyms
|
DEDUPE stable keys BEFORE enrichment:
company = root domain + normalized name
person = email when present, else name + company key
location = one firm with three branches is ONE partner record
|
FILTER TO ICP drop out of scope rows NOW, while they are still cheap
|
ENRICH waterfall only on survivors:
Apollo first, Clearbit style fallback second,
each field records WHICH provider filled it
|
VERIFY email verification pass; catch alls and role accounts flagged,
hard bounces removed, verification status kept as a column
|
QA GATE sampling review against acceptance thresholds (below)
|
DELIVER live Clay table + CSV + Google Sheet, column dictionary included
Table design
| Column group | Fields | Source |
|---|---|---|
| Identity | Company, root domain, person, normalized title, seniority | Source plus normalization |
| Firmographic | Employee band, industry, location by city, region and country | Enrichment waterfall |
| Contact | Verified email, verification status, LinkedIn URL, phone where licensed | Waterfall plus verifier |
| Provenance | Source of each enriched field, enrichment date, dedupe key | Generated |
| ICP flags | Match results, inclusion reason, exclusion reason for near misses | Filter logic |
Provenance columns matter more than they look. When a field is inaccurate you know which provider to stop trusting, and when the list is refreshed next quarter the table runs again instead of being rebuilt.
Acceptance thresholds I hold the list to
- Zero duplicate company keys and zero duplicate person keys in the export.
- Every row carries a verified or flagged email status; hard bounces do not ship.
- Titles normalized to a controlled vocabulary, so filters and mail merge work.
- Every enriched field traceable to its provider.
- A random sample check documented before handover, with the sample and its results shared.
Where the same discipline already runs
pg-importer is a public repository with 17 automated tests that applies the same rules to Excel and CSV imports into PostgreSQL: files are staged, only rows that changed are updated, and a file already imported is skipped unless the run is forced. List hygiene is not a Clay feature; it is a working method that Clay makes fast.
Specification written in August 2026; the pg-importer reference was added in September 2026. Criteria for any project plug into the filter and ICP flag stages without changing the architecture.