The problem is intake, not assessment
A motor claim arrives as an unstructured bundle: a set of photographs of varying quality, a registration document, sometimes a police report, sometimes a repair quote. Before anyone can assess anything, someone has to establish which vehicle this is, whether the documents belong to it, and whether the photographs are usable at all.
That triage work is entirely mechanical, and it is where claim cycle time actually goes.
What the pipeline resolves
Identity. CarScope reads the plate, the VIN and windshield markings directly from the claim photographs, and profiles make, model, year, body type, colour and trim from a single image. The output is standardised, so it can be matched against the policy record without a human transcription step.
Capture quality. ImageScope grades each photograph for environment, weather, lighting, blur and reflectivity, and reports camera angle, proximity and which part of the vehicle is visible. A claim whose photographs cannot support an assessment is identified at intake rather than after assignment.
Documents. ArchDoc classifies the paperwork across 30+ document types and extracts the fields that matter, rather than returning a page of OCR text. New document types can be onboarded with a custom schema in minutes.
Damage. DamageScope segments every visible damage area at pixel level, classifies type and severity, identifies the parts involved, and produces a repair-versus-replace recommendation with estimated labour hours.
Assessment. ExpertArch reasons over all of the above and produces a natural-language expert summary, a risk model and a cost prediction — the document an adjuster would otherwise have written from scratch.
Where the decision boundary sits
Every extracted field carries a confidence score. That is what makes selective automation possible: a straightforward claim with high-confidence extractions across the board can be routed automatically, while a claim with a document mismatch, an unreadable plate or a borderline severity classification is escalated with the specific reason attached.
The alternative — a single quality score for the whole claim — forces a choice between automating everything and automating nothing.
Integration shape
The workflows are independently callable, so a first deployment does not have to be the whole pipeline. Insurers typically start with intake quality and vehicle identification, where the operational gain is immediate and the risk is lowest, then extend to damage assessment once the confidence thresholds are calibrated against their own historical files.