What actually happens when you merge two duplicate customer records?
Two records become one, and something is always discarded. The rules that decide what survives — which name, which address, which lead source, which created date — are called survivorship rules, and most teams never write them down. Decide them in advance, prefer appending history over overwriting it, keep the losing record’s ID, and make every merge reversible before you let anything merge automatically.
The merge is a series of small decisions
Field by field, a merge asks the same question: when the two records disagree, which value wins? Different fields deserve different answers, and “most recent wins” is wrong surprisingly often.
Original lead source is the classic example. The oldest value is the true acquisition source; the newest is usually whatever channel triggered the most recent job. Letting recency win here quietly rewrites your marketing history and makes long-run channel analysis meaningless.
Sensible defaults per field type
- Created date and original source: oldest wins. These describe acquisition and should never move.
- Contact details: newest non-empty wins, with the older value retained as an alternate rather than deleted.
- Notes, calls, jobs, invoices, attachments: append everything. History is never a conflict; it is a union.
- Consent and opt-out status: the most restrictive wins. If either record says do not contact, the merged record says do not contact. This one is not a preference.
- Custom fields: decide individually. If nobody can articulate a rule for a field, that is a sign the field has no owner and probably no purpose.
Keep the tombstone
When record B merges into record A, something somewhere still refers to B — an old report, a synced record in another platform, a link in an email, an integration that cached the ID. Hard-deleting B breaks all of it silently.
The durable pattern is a tombstone: keep the old identifier with a pointer to the surviving record, so any lookup for B resolves to A. Every system in the chain should honor that redirect. This is one of the details that separates an integration built to last from one that works until the first cleanup, and it is a standard part of how an integration layer handles identity.
Reversibility and review
Before you enable automatic merging, answer one question: can you undo it? If the answer is no, run merges through a review queue and log the full before-and-after state of both records. Storage is cheap. Reconstructing a wrongly merged customer from memory is not.
Set thresholds accordingly — auto-merge only on very strong evidence, human review for the ambiguous band, ignore the rest. Then track how often the reviewers disagree with the system. That disagreement rate is the honest measure of whether your matching is ready for more autonomy, and it is the same discipline that governs any automated action against a system of record.
Topics: deduplication · merging · survivorship · data governance
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.