CRM Deduplication: Choose the Record You Are Willing to Trust

Two contact records may represent one person using a work and personal email, two people sharing an inbox, or one address that was reassigned. Two company records may be subsidiaries, locations, a rebrand, or a true duplicate. Identity rules belong to the object and the business context.

HubSpot documents email as a primary automatic deduplication signal for contacts and company domain for companies; record IDs and custom unique-value properties support other operations. Preserve the platform record ID as an internal identity even when the business also uses an external customer, account, or ERP identifier.

Generate possible pairs with exact identifiers, normalized phone or address, domains, names, associations, and other approved fields. Then score or review the evidence. Similar names should surface a candidate, not authorize a merge. Missing data is uncertainty, not agreement.

Create confidence bands. High-confidence pairs can be merged automatically only when the merge is recoverable and no protected difference exists. Medium-confidence pairs need a reviewer. Low-confidence pairs remain separate. Add explicit blocks for active opportunities owned by different teams, conflicting consent, legal entities, and regulated records.

Choose which record survives and how each field resolves: most recently verified, system of record, non-empty trusted source, or manual choice. Preserve original source, consent and suppression state, lifecycle timestamps, owner, activities, deals, tickets, and association labels. Never let a newer but untrusted import overwrite a verified customer fact simply because its timestamp is later.

Record the candidate IDs, evidence, decision, field changes, surviving ID, affected external IDs, reviewer, and timestamp. Notify or reconcile downstream systems that still reference the retired record. A merge that looks clean inside one CRM can break invoicing, product access, or attribution elsewhere.

Report new duplicate candidates by source: forms, imports, events, sales creation, API integrations, enrichment, or sync. Track review queue age, automatic merge rate, reversals, conflicting consent cases, and records that return after cleanup. The source distribution points to the prevention work.

Use upsert or lookup patterns with durable identifiers, normalize approved fields at entry, and test concurrent requests. Review integrations after schema or vendor changes. The objective is not a perfectly empty duplicate queue; it is a trustworthy customer history and fewer people receiving contradictory service from fragmented records.