lead enrichment returned a company that shut down in 2023 - the data provider's cache was 2 years stale and the agent...

TL;DR

Never trust a single enrichment pull on whether a company is alive. Cross-check liveness with at least one independent signal before the lead enters a sequence: is the website up, is there any recent news or funding, does the domain still resolve. If the signals disagree or the data is old, quarantine the lead instead of emailing it. Congratulating a dead company on its bright future is the fastest way to tell a prospect your data is garbage.

lead enrichment returned a company that shut down in 2023  -  the data provider's cache was 2 years stale and the agent didn't cross-check

Steps

1. Check the provider's data vintage on every pull: when was this record last verified. Treat any company record older than your threshold as unverified, not as fact.

Expected: a record last verified 2 years ago is flagged unverified before it reaches any template.

1. Run at least one independent liveness check: fetch the company homepage and confirm it loads, or check for news or funding in the last year. One live signal beats a stale database row.

Expected: a company whose homepage is gone and has no news since 2022 is quarantined, not sequenced.

1. Quarantine leads that fail liveness instead of dropping them silently. Someone should review the quarantine and confirm the shutdown, because providers do get this wrong in both directions.

Expected: the quarantine list holds the flagged leads with the reason and the evidence, ready for a human glance.

1. Track liveness failures per provider. If one source keeps serving dead companies, its weight in your enrichment stack goes down or it gets replaced.

Expected: the provider scorecard shows which source produced how many dead-company flags this month.

1. Re-verify liveness at send time for high-value accounts, not just at enrichment time. Companies die between list build and send day.

Expected: a same-week re-check catches the account that shut down after the list was built.

Use this when

• enrichment data might be stale and nobody checks whether the company still exists
• you need a liveness gate between enrichment and sequencing
• a provider's cache age is unknown and you're treating its rows as current

Not for this skill when

• headcount numbers that disagree across providers (that's a reconciliation problem)
• a lead's job change making the cached employer stale (that's a cache-freshness problem)
• picking which enrichment provider to buy (that's a vendor-evaluation problem)

Variant phrasings

• enrichment returned a company that shut down years ago, provider cache stale
• agent prospected a dead company, no liveness check before outreach
• data provider serving 2-year-old company records as current

Why it happens

The provider's cache hadn't been refreshed in two years, and the agent treated the pull as current fact. Nothing in the pipeline asked the obvious question, does this company still exist, because liveness was assumed rather than verified. Stale data doesn't announce itself; it just sits in the payload looking exactly like fresh data.

Edge cases

• a company in stealth or pre-launch can look dead by these signals; allow an explicit override for known stealth accounts
• acquisitions and rebrands look like shutdowns to naive checks; a homepage redirect to an acquirer is a signal to update the record, not quarantine it
• liveness checks cost time and requests; run them on the shortlist entering sequences, not on the raw imported list of 100,000
• some providers refresh on a schedule you can learn; if yours refreshes quarterly, your vintage threshold should know that

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_j-8OVnPcE2BeQkq3MnZlMw