Both of our products rest on the same idea: a photograph taken on an ordinary device contains enough information to answer a claims question, provided you control the capture and you understand what the image is made of. One product uses that to measure damage. The other uses it to decide whether the image can be trusted.

Fixed hail scanning installations produce good results and impose a logistics problem: the vehicle has to come to the equipment. After a regional storm that constraint, rather than the measurement itself, is what sets the pace.

We take the opposite approach. Detection and sizing run on the device in the operator's hand. Scale comes either from the LiDAR sensor on recent hardware, which gives the best experience, or from a printed or magnetic scale marker placed on the panel, which produces dent sizes on devices without a depth sensor. That choice is what allows the same product to work on an assessor's managed tablet, a repairer's phone and a policyholder's own device.

Accuracy in the field is determined more by capture than by the model. On-screen masks and overlays position the operator and the reference marker, so the system receives the images it was trained to interpret rather than whatever the operator thought to take. This is also what makes results comparable between a trained assessor and an untrained policyholder.

Detection produces dents. Claims need a number that can be defended. Between the two sits the panel zone model: every dent is assigned to a zone of the vehicle, counted and grouped by size band, so two assessments of two different vehicles are expressed in the same structure.

Twenty two calculation matrices are built in, twelve reflecting established industry standards and ten built to individual customer specifications, covering both metric size ranges and coin based sizing for the North American market. A matrix does not have to carry pricing: in dent count only mode the output stops at a standardised count and the commercial step stays in the customer's estimating system.

Reference standards include the BVAT calculation specifications and hail tables used across the German-speaking market.

Image processing runs at roughly one second per image, which is what makes a full vehicle assessment in around fifteen minutes possible rather than theoretical. Detection runs on the device, so field work does not depend on connectivity at the vehicle.

The same reasoning applied to a different question. Photographs submitted with a claim can be generated, edited or reused, and a measurement taken from a manipulated image is worse than no measurement at all. Our second product, currently at MVP stage, examines the evidence itself: pixel level analysis for signs of manipulation, metadata forensics covering camera, location and timestamp consistency, and vehicle identity validation, combined into a single score with every point traceable to a specific finding.

It is built to flag claims for human review, never to decline them, and personal data is removed before any analysis takes place.