enrichment said 50 employees but the company actually has 4,000 - the agent's ICP filter dropped the whole account

TL;DR

Don't let one provider's number veto an account. When headcount drives your ICP filter, pull the number from at least two providers and reconcile: if they agree within a band, use it; if they disagree wildly, flag the account for review instead of dropping it. A filter that drops a 4,000-person account because one source said 50 is not being selective, it's being wrong with confidence. Disagreement is a signal to look, not a signal to delete.

enrichment said 50 employees but the company actually has 4,000  -  the agent's ICP filter dropped the whole account

Steps

1. Pull headcount from at least two independent providers for every account that matters, and store each source's number separately, not just the one you picked.

Expected: the account record shows two headcount numbers side by side with their sources.

1. Define agreement as a band, not an exact match: within 2x of each other counts as agreement. Headcount is always an estimate; exact matches are luck.

Expected: two sources saying 3,800 and 4,200 are treated as agreeing and the account keeps its ICP score.

1. On wild disagreement, flag for human review instead of filtering out. The review queue is the price of not dropping good accounts.

Expected: a 50 vs 4,000 split routes to review, and the account is never auto-dropped on a conflict.

1. Weight sources by their track record. Keep a simple scorecard of which provider was right the last times you checked, and break ties toward the better one.

Expected: the scorecard shows provider A right 9 of the last 10 conflicts, so ties break toward A.

1. Re-run the filter after review corrections so a fixed number actually changes the account's fate. A review that doesn't update the filter is theater.

Expected: correcting the headcount to 4,000 re-scores the account and it re-enters the ICP segment.

Use this when

• headcount from one provider looks wrong and it's driving filter decisions
• enrichment sources disagree on firmographics and the filter trusts only one
• good accounts are being dropped by ICP filters on bad data

Not for this skill when

• a company that no longer exists at all (that's a liveness problem)
• stale job titles or employers on the lead (that's a cache-freshness problem)
• building the target account list in the first place (that's a sourcing problem)

Variant phrasings

• enrichment headcount wrong, ICP filter dropped a real account
• provider said 50 employees, company actually has thousands, account filtered out
• conflicting firmographics across providers, filter trusted the wrong one

Why it happens

The ICP filter read a single headcount number from a single provider and treated it as truth. Providers estimate headcount differently, some count only HQ, some are years out of date, and the filter had no second opinion and no path for disagreement. So a wrong 50 became a confident 'not our ICP' and the account vanished before any human saw it.

Edge cases

• headcount spikes after funding or drops after layoffs; a number that's right today can be wrong next quarter, so re-pull for key accounts periodically
• some providers count contractors and some don't; learn what each of yours counts before you compare their numbers
• the review queue needs an SLA, or flagged accounts sit in limbo and effectively stay dropped
• don't reconcile by averaging; an average of 50 and 4,000 is 2,025, which is wrong in a way neither source intended

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_mywVEjL6Zft6gdteie8UvQ