Score a signature against everything that signer has signed before
We take delivery notes signed on paper, scan them, and file them. Last month two of them came back with a signature that looked wrong to the person who has filed them for nine years — close enough to pass a glance, different enough to stop them. They were right, and the only reason we know that is that they happened to remember the earlier ones.
That is not a control. It is one person's memory, and it does not scale to 4,000 documents.
What we actually need
- A reference set per signer, built from the documents we already hold.
- Upload a document, and get a similarity score for its signature against that signer's earlier ones — a number, with the closest matches ranked.
- The reference beside the result, so we can see what the number is comparing.
We are not asking for a verdict. If a document scores far below everything else that person has signed, we want it flagged and looked at by a human. That is the whole job.
Why it matters
A forged delivery note is a delivered load nobody delivered. We find out weeks later, in a reconciliation, when the goods and the paperwork disagree and the person whose name is on the note says it was not them. By then the argument is about who remembers what.
Catching it at the point of filing turns a dispute into a question.
What we do today
Nothing systematic. Suspicious documents get pulled by hand, and someone opens three or four older files next to them on two monitors to compare. It happens when a person is paying attention, which is not every day, and never on the documents nobody had a reason to doubt.