Automating the first pass on vehicle insurance claims
On-device photo guidance plus damage extraction, deployed across insurers in the US and India.
- Insurance
We build the first pass.
Your experts make the call.
You hand over the work that has to happen before anyone can decide: the packages to be checked against a hundred regulations, the encounters to be audited, the meters to be inspected, the collection to be catalogued. Work that is mechanical, high in volume, and consequential when it is done badly.
Illustrative
First-pass review queue4,182 packages
| Package image | Package | Checks flagged | Confidence | Flagged for review |
|---|---|---|---|---|
 | PKG-8841-C | 14 checks flagged | conf. 0.94 | Flagged |
| PKG-8839-A | 9 checks flagged | conf. 0.91 | Flagged | |
| PKG-8846-F | 3 checks flagged | conf. 0.62 | ||
| PKG-8850-B | 1 check flagged | conf. 0.41 | ||
 | PKG-8852-D | 0 checks flagged | conf. 0.12 |
We hand back a system that does that first pass inside your environment, validated against your own historical data and deployed into the workflow your team already uses — along with the training to retrain it yourselves. The system ranks what to look at. Your people determine what it means.
- 3 hours → 20 min
- regulatory package review
- 2 weeks → 2 days
- first-pass collection indexing
- 4.21%
- baseline error rate the audit system was measured against
- ~€12,000/mo
- recovered, first test cycle
Machines should take the work that is repetitive, mundane, and consequential — the work where attention runs out before the task does. Not the judgment on top of it. Most of this industry has that backwards.
How engagements run
01
Discovery
Two to four weeks. We work against your historical data and establish what the current process actually costs.
02
Pilot
Four to eight weeks, with a target agreed in advance. The pilot runs on your data, in a form you can inspect.
03
Production and handover
Deployment into the workflow your team already uses, then training so they can run and retrain it without us.
If the data does not support a system that beats the current process, this is where we say so.
Selected work
All 6 case studies
Cutting regulatory package review from three hours to twenty minutes
A production assistive system for federal compliance inspectors, won through competitive RFP.
- Government / Regulatory compliance
Tagging a million-image art collection
A tagging pipeline that took first-pass indexing from ten people for two weeks to three researchers for two days.
- Cultural heritage
Turning a compliance system into a working pipeline queue
Lifecycle modelling and live dashboards for a mortgage brokerage that had the records but couldn't get answers out of them.
- Financial services / Mortgage origination
What we don't do
- Systems that make the determination
- Work we can't measure
- Staff augmentation
- Anything that requires us permanently