The costliest mistake in investigative work is not missing something. It is attributing something to the wrong person — a company directorship, a conviction, a social media post — because two people share a name.
It happens constantly, it is easy to miss in review, and it is the error most likely to end in a correction letter or a lawsuit.
Why it happens so easily
Names repeat. In some name populations, a full name plus a city still describes dozens of people.
Aggregators merge. Databases join records on weak keys, and a merged profile arrives looking like a single coherent person.
Confirmation feels good. Finding a match after two hours of nothing is psychologically rewarding, and the impulse is to keep it.
Nobody re-checks the first step. The identification made on day one is rarely revisited on day five, even as conclusions pile on top of it.
The rule: a name is a lead
Treat a name match as a hypothesis requiring an exact link before anything is attributed. Exact links look like:
- A unique identifier consistent across records.
- A record that ties the identifier to the person (a filing, a registry entry, an official document).
- A dated image or record placing the same individual in both contexts.
- A chain of corroborating details that would be implausible for a different person, stated as such rather than assumed.
Weak links: a shared employer, a city, an approximate age, a similar photograph, a common email handle. Useful as support, never sufficient alone.
Keep the candidates separate
When several people plausibly match, hold them as separate candidates and attach each piece of evidence to the one it actually supports. Collapsing them early is what produces a monster profile that fits nobody.
If the evidence never resolves, that is the finding: the subject could not be distinguished from N others with the same name. That sentence protects everyone, including you.
What never counts as evidence
Inferring origin, nationality or ethnicity from a surname is not identification. It is guesswork wearing a lab coat, and it produces exactly the errors that damage people who were never part of the case.
How a workspace should help
By refusing to merge identities silently. Proposed matches stay proposals until a person accepts them; candidates stay distinguishable; and when an identification is accepted, the case file keeps what it rested on. See accepted and rejected evidence and Engramite Graph.
Frequently asked questions
How much corroboration is enough? Enough that you would be comfortable defending it to the person you would be wrong about.
What if the client insists it is the same person? Record the evidence and the limitation. Their certainty is not your finding.
Can automation solve this? It can find candidates and flag collisions. It cannot decide, and a tool that decides quietly is worse than none.
Read next: accepted and rejected evidence or provenance and retention.