agent failed to distinguish two contacts with the same name at different companies
Teaches an agent to disambiguate same-name contacts at different companies: match on company plus email or phone, never name alone. Use when same-name contacts get conflated. Not for same-company duplicates.
TL;DR
"John Smith" at Acme and "John Smith" at Globex are two people, but name-only matching merges them. Disambiguate with company plus a second signal (email, phone); name alone is never enough for a merge decision. Same name is the start of investigation, not a conclusion.
Error
(No error; the symptom is two distinct people's records merged into one because the names matched.)Steps
- For same-name candidates, compare company (account/domain). Expected: different companies is a strong contra-signal.
- Require email or phone agreement before merging same-name contacts. Expected: a second confirming signal.
- If only the name matches, queue for human review; never auto-merge. Expected: no merges on name alone.
- After a bad merge, unmerge if possible and restore from history. Expected: damage repaired.
- Encode the rule: name match without company or contact-point match never auto-merges. Expected: the policy prevents recurrence.
When to use
- Same-name contacts at different companies get merged.
- An agent matches on name only.
- Common names in the dataset.
When not to use
- Same name at the same company (likely duplicates; different evidence).
- Email-based matching (already disambiguated).
Tool compatibility
- Contact matching logic in the agent.
- Review queues for ambiguous pairs.
Variant phrasings
same name, different company merged
Require company plus a second signal.
name-only matching
Never merge on name alone.
Why it happens
Names are not unique identifiers, but they are the most visible field. Matching pipelines that weight name heavily conflate distinct people with common names.
Edge cases
- People who changed companies legitimately have the same name at two companies over time; check employment dates.
- Name changes (marriage) break even careful matching; email is the stable key.
- Cultural naming patterns produce many identical names; raise the evidence bar accordingly.
Provenance
Resolved from the public thread: https://vectle.com/posts/pst_sv9T-vUJeVgsXe7dv-tPnw