TL;DR
Identity resolution decides whether two records refer to the same real-world entity. Golden DB uses deterministic matching first (exact email match), then probabilistic matching (fuzzy name + domain + similar fields). Confidence scores accompany every match so you can audit edge cases.
The two-tier approach
- Tier 1 — deterministic matching: exact match on email address (for people) or domain (for companies). High confidence; runs first; resolves ~80% of cases.
- Tier 2 — probabilistic matching: fuzzy match on combinations of name, domain, employer, location, and role. Used when deterministic fails.
What deterministic matching catches
- Same email — different sources providing the same email = same person.
- Same domain (for companies) — same primary domain = same company.
- Same external ID — sources sharing a common ID (e.g. LinkedIn URL, Salesforce ID).
What probabilistic matching catches
- Email variation — work email at company A, current email at company B = same person who moved jobs.
- Name + employer — same first/last name at same employer = likely same person.
- Domain variations — acme.com vs acmeinc.com vs acme.io for the same company.
- Subsidiary mapping — parent company vs subsidiary names.
Confidence scores
Every match has a confidence score (0–100):
- 95+ — high confidence, auto-merge.
- 70–95 — medium confidence, merged but flagged for review.
- <70 — low confidence, held for human review.
Adjust the cutoffs in Settings → Identity Resolution → Thresholds.
Reviewing borderline matches
Data → Match Review. Shows matches with confidence below 95, with side-by-side records and the agent's reasoning. Approve, reject, or merge with notes. Most teams review weekly; volume is low after initial setup.
Did this answer your question?
