Rebuilding the claim form people meet after a crash
The old claim form asked for everything at once, and quietly filed the claim before the customer had finished. I rebuilt it as a short sequence that adapts to what actually happened and commits to nothing until the last button.
Nobody fills in a car insurance claim in a good mood. They have just had a crash — they're on a roadside, or at a kitchen table that evening — and now a form wants document numbers, photographs and bank details.
The old one asked for all of it at once, on a single long page. I rebuilt it as a short sequence of questions that adapts to what actually happened, and that doesn't file anything until the person says they're done.
A form you meet on the worst day of your year
A single long form has a particular cruelty to it: it shows you everything you don't have to hand, all at once, before you've answered a thing.
It also asks for things that don't apply to you. Whether you need a police decision or a self-completed accident report depends entirely on how the crash was recorded. Whether you're claiming as the injured party or the one at fault changes half the questions. One page has to show every case at once, so everybody sees a form that is mostly not for them.
Asking only the questions that apply
The new flow is eight short steps, and it branches early:
- Who are you in this? Injured party or at fault — this decides what comes next.
- How was it recorded? A self-completed report or a traffic police decision, each needing a different set of documents.
- Which policy? Checked against our records first, so a policy that isn't ours ends the flow right there with an explanation, instead of ten more steps and a rejection.
- Who are you, really? Identity confirmed through the national digital ID, so there's no passport to type in.
Then the documents, the photographs of the damage, and the bank details for the payout. Each step holds one decision, and when something is wrong the error appears on the step that owns it — not at the bottom of a page of forty fields.
Nothing is filed until the last button
This was the fix that mattered most. In the first version, the early steps quietly created the claim as you went. Anyone who opened the form to look around, or left to find a document and never came back, left a real claim sitting in the claims department's queue for staff to chase.
Now all three requests — create the claim, send the accident report, upload the documents — go out together when the last button is pressed. That raised two problems worth solving properly:
- Photographs are enormous. A crash claim can carry a dozen phone photos and a video. One upload would blow past the size limit, so the files are packed into batches that fit, with headroom left for everything else in the request.
- A retry must not file twice. If the upload half fails on a weak connection, pressing "try again" has to resume — not lodge a second claim. The claim already created is remembered, so a retry picks up where it stopped and the person lands back on the step that actually failed.
Finding out where people give up
Eight steps means eight places to lose someone, so the flow now reports its own progress: which step was reached, where sign-in happened, where payment began, where people stopped.
One detail took a while to get right. Sign-ins that complete in the same browser tab weren't being counted at all, which made the digital ID step look far worse than it actually was.
Why I'm proud of it
The flow is finished and sitting behind a release switch, waiting on the business side to turn it on. I'd rather say that plainly than imply otherwise — it's built, it works end to end, and it is deliberately not live yet.
The hardest part was never the eight screens. It was the boring promise underneath them: filling in a form should never commit you to anything until you say so. Keeping that promise meant reworking how the whole thing talks to the claims system, which the customer never sees and is the entire point.