Deduplicate the numbers, not the spreadsheet rows
Normalise every number to international format before you deduplicate, because the same person written three ways looks like three people to a spreadsheet. Once numbers are consistent, match on the number rather than the name, and decide up front which record wins when two disagree.
Duplicates cost real money. You pay per message, so the same person appearing twice is charged twice, and receiving the same campaign twice is exactly the behaviour that pushes a quality rating down.
The reason most deduplication fails is that it runs in the wrong order.
Deduplicate in the right order
Normalise the numbers first
9876543210, +91 98765 43210 and 098765-43210 are one person and three strings. Any tool matching raw text sees three contacts.
Strip everything that isn't a digit, add the country code, and store one consistent format.
Match on the number, not the name
Names are unreliable. Priya Sharma, P. Sharma and PRIYA S are the same person, and two genuinely different people can share a name.
The phone number is the identity in a messaging list. Treat it that way.
Decide which record wins before you merge
Two records for one person will disagree about region, source or stage. Pick a rule now: most recent activity usually wins, or the record with more fields filled in.
Whatever you choose, apply it consistently rather than case by case.
Keep the history from both
Merging shouldn't lose the conversation attached to the record you discard. If your tool drops it, you've traded one problem for a worse one.
Check this before you merge anything in bulk.
Handle one person with two numbers separately
This is the case deduplication can't solve on its own, because both numbers are valid and reachable.
Decide whether they're one contact with two numbers or two contacts, and be consistent. Most businesses want one.
What counts as a duplicate
| Situation | Treat it as |
|---|---|
| Same number, different formatting | One contact. Normalise and merge |
| Same number across four uploaded files | One contact. This is the common case |
| Same name, different numbers | Two contacts until you have evidence otherwise |
| One person, personal and work numbers | One contact with two numbers, if your tool allows it |
| Same number, two different people | Rare but real, usually a reassigned number. Trust the recent activity |
| Duplicate email, different phone | Depends on your business. Pick a rule and stick to it |
Stop creating duplicates in the first place
Stop keeping the master list in spreadsheets. Duplicates are a symptom of the same person being stored in four places. Fix where the truth lives and the symptom fades.
Normalise at the point of capture. Forms, imports and integrations should all write the same format, so nothing has to be repaired later.
Deduplicate on import rather than before a send. Doing it under time pressure the night before a campaign is when mistakes happen.
Some duplicates are judgement calls, and no tool should silently decide for you. One person with two numbers is a business decision, not a data problem.
Merging is hard to undo. Export a copy before a bulk merge, always. This is advice we'd give about any platform, ours included.
Merge them on the way in
Amplifeed normalises and merges duplicates when you import, across every file you've ever uploaded, and shows you what it merged before anything sends.
Upload your messiest sheetRelated questions
Why does Excel miss duplicate phone numbers?
Because it matches text. Two spellings of the same number are two different strings, so they never match until you normalise the format first.
Should I match on name or number?
Number. Names are inconsistent and shared; in a messaging list the number is the identity.
Do duplicates cost me money?
Yes. Messaging bills per message, so the same person twice is charged twice, and it also raises your risk of being reported.
What if one person has two phone numbers?
Decide whether that's one contact or two, then apply it consistently. Most businesses treat it as one contact holding both numbers.