Starting point: a spreadsheet list and two ways in
The HR department of the plant receives a spreadsheet from an external company with several dozen people to enter before work starts. Some of them were already in the system: one was registered at the gate, another scanned a QR code from a poster and created an account on a phone. The spreadsheet does not know this, and the person entering the rows has no practical way to notice it across several dozen entries.
A typical starting point at a plant looks similar: work permits on paper and in Excel sheets, companies invited by different people, training validity tracked by hand. In such an environment the same person easily appears in several places under slightly different spellings of the name.
Production event: two accounts in six minutes
In the production event, one person had two accounts. The first was created at the gate, without an e-mail address. The second was created six minutes later from a QR code poster. The system treated them as two different people, so both received a training task and the person completed it twice.
The loss was not limited to lost minutes. Two accounts mean two training histories, two attendance records, and the risk that a work permit is issued to the account that does not have valid training. During an audit, someone then has to explain which account is the correct one.
- first account: registration at the gate, no e-mail,
- second account: QR poster registration six minutes later,
- effect: training completed twice by one person.
Trainings for a list of 100 people: with and without the check
How the system matches people: strong and weak matching
The duplicate check distinguishes two levels of certainty. Strong matching uses an e-mail address, a phone number in E.164 format, an employee ID or an external login identifier. Weak matching uses first and last name, ignoring letter case and diacritics, so 'Jose Garcia' and 'José García' are treated as the same pair.
The distinction matters because an e-mail address clearly points to one person, while a name does not. Two different people can have the same name, so weak matching never merges accounts automatically.
Strong matching by e-mail
If the registration e-mail matches an active account that has no password and no login identifier, the system links the registration to the existing account. This is exactly the case of the person entered earlier at the gate. Other strong matches, such as an account with a password, a pending invitation or a deleted account, end with response 409, and a deleted account is never restored.
Weak matching by first and last name
With weak matching the decision is made by a person. The system shows at most 3 candidates with masked e-mail and phone numbers, and the decision token is valid for 780 seconds. Masking protects the data of people whom the person entering the data does not need to see in full.
Matching by phone number only
Phone numbers are not verified, so matching by phone alone from a public form changes nothing on the account. The registration goes to approval by an administrator. This way someone else's number typed into the form cannot be used to take over an account.
Which entry paths the check covers
The duplicate check covers the invitation from the panel and registration from a QR code, which are the two paths through which lists of external companies usually arrive. Gate entry, quick adding of an employee, API, card swipe and tablet registration keep their existing checks. This matters when planning: not every entry into the system goes through the same filter, so for a large list it is worth choosing a path covered by the check.
The check works from the moment of deployment, not retroactively. Duplicates that already exist in production must be merged separately, and merging is a data change that requires a separate decision.
What to do with the spreadsheet before entry
The cheapest way to avoid duplicates lies on the spreadsheet side. Each row should have an e-mail address or a phone number in E.164 format, because these are the data on which matching is strong. Rows without such data go to the weak matching path and require a human decision.
A sensible sequence for a large list has four steps. First, we tidy up the spreadsheet: remove repetitions in the file itself and unify the spelling of names. Then we enter the list through a path covered by the duplicate check. Next we review the queue of decisions, that is the weak matches and phone registrations waiting for approval. Finally we check whether the number of active accounts matches the number of people on the list.
- unify phone numbers to E.164 format,
- complete e-mail addresses where the person has one,
- remove repetitions within the spreadsheet,
- resolve candidates in the queue before training starts.
Result and conclusions for the plant
After the check is introduced, the same person does not get a second account if the strong data agree with an existing account. Training is completed once, the history is single, and the work permit is linked to the correct account. In an example calculation for a list of 100 people, of whom 10 have a duplicate, this means avoiding 10 repeated trainings, that is 10 unnecessary entries into the training module, assuming one training per person.
Practical conclusion: duplicates are a problem of input data, not of handling. Whoever cleans up e-mail addresses and phone numbers before the import moves most decisions to automatic matching and leaves the human only with the truly ambiguous cases.
Video
Frequently asked questions
Will the system merge accounts automatically based only on the surname?
No. Matching by first and last name is weak, so a person chooses from at most 3 candidates with masked data.
What happens when an e-mail matches a deleted account?
The registration ends with response 409, and the deleted account is not restored. The administrator makes the next decision.
Will the check remove the duplicates we already have?
No, it works from the moment of deployment. Merging existing duplicates is a data change that requires a separate decision.
Have a list of people to enter?
We will show on sample data how the system matches accounts and what the decision queue looks like. Book a talk or a demo.
Let's talk