Contact cleaning
A routine that finds the same customer recorded twice, proposes merging the records and flags what was created by mistake, without deleting anything on its own.
- Domain
- AI for productivity
- Complexity
- Low
- Typical timeline
- 1 to 3 weeks
Contact cleaning finds the same customer recorded twice and merges the records into one, before the duplication spreads through the rest of the system.
What it is
A routine that reads the contact base looking for records that are the same entity written differently: the same tax number on two cards, the same email under swapped names, a company created once by the salesperson and once by the website form. It marks what it finds, proposes the merge, and separately flags what was created by mistake and has nothing to merge with.
It does not delete on its own. Each case is marked for a person to confirm, because merging two cards that turned out to be two different customers is far worse than holding a duplicate.
Where it pays off
It makes sense in any operation where contacts arrive through more than one door: the website, the sales team, the phone, an old import. Duplicates cost little individually and a lot collectively, because they split one customer's history in two, send the same message twice, and turn any count into a number nobody trusts.
How we build it
We start from the rules that decide what counts as the same: usually the tax number, then the email, then name plus phone, each with its own confidence. We set what the routine may merge on its own and what has to wait for a person, and run it read-only first to see how many cases surface and whether the rules are right. Only then does it run unattended.
Where we've built this
This workflow isn't theoretical. See the real project where we implemented it.
Frequently asked questions
Does it delete contacts?
No. It marks and proposes, and the merge is confirmed by a person. Merging two records that turned out to be two different customers is considerably worse than holding a duplicate one more day.
How does it know they are the same?
By rules with different confidence: the tax number is near certainty, the email is strong, name plus phone is an indication. What has high confidence can merge on its own if you set it that way; the rest waits.
And the history of the merged records?
It is merged rather than lost, which is precisely why this workflow exists: a customer whose history is split across two records looks like two small customers instead of one good one.
Investment in AI and automation may be eligible for public support in Portugal, subject to the calls in force. We don't handle applications and we're not a funding intermediary.