A customer calls and complains: "you called yesterday, you are calling today — how many times do I have to explain?" You check the records and find the same person exists twice. One record came from the web form, the other from a phone call.
How duplicates form
There are four main sources:
- Number formatting. "0532 111 22 33" and "+905321112233" are the same number written differently. The system sees two.
- Different channels. The same person messages on WhatsApp, then fills in a web form. Two channels, two records.
- Bulk imports. An Excel file uploaded a second time.
- Missing lookup. A new record is created without searching for an existing one, or the search only matches on name.
Prevention is cheaper than cleanup
Two controls at the point of entry stop most duplicate creation:
- Normalise the number. Before a record is created the number should be converted to one form: country code, no spaces. That single step closes the largest source of duplicates.
- Search before creating. Every flow that creates records — forms, webhooks, imports — should search by number first and update when it finds a match.
How to find them
Duplicate detection rests on three criteria, in this order of reliability:
- Phone number. The most reliable. The same number is nearly always the same person.
- Email. Reliable, but company addresses get shared (
info@). - Full name. The weakest. Two different people can share a name; it should never justify a merge on its own.
A rule that works in practice: a number match generates an automatic merge suggestion, while a name match only lands on a review list.
What happens in a merge
Merging is hard to undo, so what becomes of what should be known in advance:
- Conversations, calls and notes combine; nothing is lost.
- Tags combine.
- Conflicting fields need a rule: which record is the "primary"? Usually the older or the more complete one.
- Conflicting owners require a human decision.
Pre-campaign check
Running a duplicate scan before a large campaign lowers cost and prevents complaints. Calling the same person twice is a campaign's most visible error and leaves the impression of a disorganised company.
Cases that must not be merged
Some records look similar but should stay separate: two people at the same company, family members sharing a landline, two departments using the same info@ address.
So automatic merging should happen only on a high-confidence number match; everything else belongs behind human approval.
Tomorrow we look at bringing lists in from outside: from Excel to campaign.