The next frontier in fraud prevention: collaboration

Fraud is a team sport. The people who commit it share tools, accounts, identities and playbooks, and they move between institutions faster than any one of those institutions can react. The defence, by contrast, is mostly played alone.

Every institution sees one frame of the film

A bank sees a new business account receiving money. A telecom sees that the phone number on that account changed SIM last week. A utility knows there is no active service at the address on the application. Each fact on its own is unremarkable. Together they tell a story, but no one is in a position to put them together.

The obvious fix, pooling everyone’s data in one place, is the wrong one. It creates a single, very attractive target, it strains every privacy law on the books, and it asks institutions to give away the thing they are most responsible for protecting. Unsurprisingly, most decline.

Share the answer, not the data

There is a narrower path. An investigator rarely needs another institution’s records. They need an answer: is this account established, did this transfer arrive, has this SIM just changed. A question like that can be answered by the custodian, inside its own walls, and only the answer needs to travel.

One answer. No records. The custodian stays in control.

That shift changes what collaboration costs. The provider exposes nothing it has not agreed to. The requester gets a fact it can act on, with a note of where it came from and how fresh it is. The customer’s data stays with the institution they gave it to.

Collaboration has to be fair to the innocent

There is a second reason to work this way. A model that only looks for bad news will find it. Good research asks whether the evidence supports a legitimate explanation just as seriously as a suspicious one, and is willing to say “we do not know”. An anomaly is not proof of wrongdoing. A real company can be a victim. A refusal to consent is not a confession.

Collaboration that respects those limits does more than catch criminals sooner. It lets institutions close unfounded suspicions quickly, which is its own kind of justice.

Where to start

• Pick one workflow where an outside fact would change the decision.
• Name the single question, and the one institution best placed to answer it.
• Agree the purpose, the permission route and the evidence standard before anything goes live.

The next gains in fraud prevention will not come from a cleverer model looking at the same data. They will come from institutions being able, safely, to ask each other questions.