Insurance and mortgage files: from paperwork to decision
A file is not a form: it is a stack of documents arriving at different moments, from different people, in formats you did not choose. And until they are all there and consistent, nothing gets decided. This is how that stack became a process. Names and identifying details are withheld: some of these projects were commissioned by third parties.
01 Where it started
ID documents, payslips, land registry extracts, surveys, declarations: every file gathers a dozen of them, and each arrives when it arrives. The real work was not deciding — that takes expertise — but knowing where each file stood, what was missing and who was waiting for whom. Information that lived in emails, in folders and in people's heads.
The cost was not only time. A file stalled over a document nobody chased is a customer waiting and, in a market where speed is part of the offer, often a customer lost.
02 What was built
A system built around the life cycle of a file: explicit states, what is needed to move to the next one, who is waiting for whom, deadlines that speak up before they expire. Not a document archive, but something that can say at any moment where you are and what is blocking the next step.
Here too, no generalist product: the life of a mortgage application looks nothing like a sales pipeline, and forcing it into a CRM means spending the rest of your life explaining to users why the field is called what it is called.
03 What it does today, with AI
AI reads the documents that make up the file and extracts the data that used to be retyped: names, amounts, dates, references. It pre-fills the file and — the part that matters most — flags inconsistencies: a value that does not match across two documents, a missing attachment, an expiry date that makes a certificate unusable.
The decision stays where it was: with the case handler. What changes is the starting point. Instead of opening ten PDFs to find out whether the file is complete, you open one screen that tells you — and spend the time you saved on the files that genuinely need an expert eye.
The number: processing hours were cut in half. 50% of the time previously spent reading, retyping and checking is now automated — over a year that is tens of thousands of euros of resources freed up, which is the right way to read it: not fewer people, but the same people on work worth doing.
| Task | Who does it today |
|---|---|
| Reading documents and extracting data | AI, with the handler confirming |
| Working out what the file is missing | AI, flagged in plain sight |
| Judging edge cases and deciding | The person, always |
| Chasing the customer at the right moment | The system, on real deadlines |
04 If this sounds like you
This applies to any work where "the file" is the unit of measure: insurance, mortgages, lending, tenders, litigation, permits. The symptom is always the same: someone opening a folder and reading everything to find out where things stand.
The number to measure before any project is how long passes between a document arriving and someone noticing it. If the answer is "it depends", there is room.
Is customer data safe if AI reads the documents?
Yes, and that is an architectural choice rather than a promise: data is not used to train models, processing can stay inside the European Union and every access is logged. For files with particularly sensitive data we consider processing on dedicated infrastructure. We cover this on the data and GDPR page.
What happens if the AI extracts the wrong value?
The system is built so the error is visible: extracted values sit next to the source document and confirmation is a human action. Automation that decides silently would be irresponsible in this field — and it is one of the cases where we recommend not automating the final step at all.