An experienced assessor can look at a job and tell you it’ll grow. Project X-Ray puts that judgement on the claim itself, scored against your own resolved cases, in words that tell your team what to do about it.
The outlook gauge
One gauge, showing where the estimate is heading. The zone sets the verdict and the verdict sets the recommended action, so nobody has to work out what a number means for their next move.
This estimate looks solid. Similar jobs rarely grew once work started.
Proceed as estimated.
Mostly on track, but this one deserves a closer look at strip-down.
Proceed, but flag it before authorising the final cost.
Expect this estimate to grow once the vehicle is stripped down.
Scrutinise the strip-down and budget for a supplement.
This looks like a write-off, not a repair.
Route to the total-loss path. Do not authorise a repair.
A total loss overrides everything else. A written-off vehicle never reaches a garage, so the card drops the repair-stage risks instead of printing numbers your assessor would have to ignore.
What it changes
Most of what goes wrong on a claim is decided in the first hour and discovered in the third week. The value is in knowing where a job is heading while you can still do something about it.
Nine days of repair effort on a car that was always going to be written off costs you a courtesy vehicle, an engineer's time and a strip-down nobody needed. The card asks that question first, before anyone authorises anything.
Nobody has time to attend every strip-down. Once you know which jobs are likely to grow, you can point your scrutiny at those and let the solid ones run without the same friction.
Where costs did grow on jobs like this one, the card gives you the typical range up front. A supplement you saw coming is just a cost. One that turns up unannounced three weeks in is a reserve movement, an awkward phone call and a dent in your cycle time.
Your experienced assessors already know which jobs will grow, and that knowledge walks out of the building when they retire. Scoring against the firm's own resolved cases puts twenty years of pattern recognition in front of someone in their second month.
How it works
Retrieve
Before anything gets scored, the system pulls the cohort: the closed cases in your own history that most resemble the job in front of you. The card names how many, and it can show you them. You can always check its working.
Gate
The first question is whether the vehicle gets repaired at all. A write-off never reaches a strip-down, so every repair-stage risk is meaningless for it. When the headline outcome is a total loss, the card drops those signals entirely instead of printing numbers that can't apply.
Score
For claims heading to repair, the model scores how likely the estimate is to grow once the vehicle comes apart, using the intake facts and the contents of the estimate itself. The parts already listed are the fingerprint of the accident.
Present
The scores become a gauge position, a one-line verdict, a recommended action, and a typical cost range where one applies. Whatever the profile says should always be checked on a job like this gets listed too, whatever the scores say.
How it is built
Models are fitted to one customer's own resolved claims. No cross-tenant pooling, no shared industry corpus, and nothing of yours in anyone else's training set. Everything the card knows about your book, it learned from your book.
Assessors told us a percentage didn't tell them what to do with it. So every score reads as a natural frequency: about 6 in 10 similar jobs found hidden damage once stripped down. Then a one-line verdict, then what to do.
The card suggests a route and shows its working. Your assessor takes it or leaves it, and we record which. Nothing gets authorised, reserved or settled off the back of a score.
An Outcome Profile says which targets to score, which depend on which, and which value fields band the cost range. Adding one is something your admin does in an afternoon. You're not waiting on us to build you a bespoke model.
A scheduled job retrains on everything that closed since the last run and writes a fresh snapshot. The card tells you how many cases sit behind the score and when it last trained, so you always know how old the answer is.
A per-install setting and a per-user role decide who sees the card. Start with two assessors, widen it once you're happy, and nobody gets opted in by an upgrade.
Questions
It's running in pilot now and we're taking on more. We start by looking at your resolved claims to see which outcomes your history can support, then configure an Outcome Profile around them.
No. Every model is fitted per customer against that customer's own resolved cases, with no shared corpus and no cross-tenant pooling. It's what makes the scores worth reading: they describe how claims behave on your book.
Enough closed cases with recorded outcomes to assemble a cohort of genuinely similar jobs. Most claims operations of any age have far more than they think, because the value sits in outcomes already recorded on the case. The card reports how many cases are behind every score, so you can always see what it's reasoning from.
No, by design. It recommends a route and shows its reasoning. Authorisation, reserving and settlement stay with the people accountable for them. It's the same principle behind why our AI never agrees a total-loss value, and it's what makes the whole thing deployable in a regulated operation.
Every model and engine change gets scored against a backtest harness on a held-out evaluation database before it ships. Then the card records whether your assessors accepted or dismissed what it recommended, so you can see whether it's earning its place on the screen.
No. An Outcome Profile defines the targets, how they depend on each other, and the value fields that band the cost range, so the same engine handles a different kind of claim through configuration. Motor is simply where we built it first.
A pilot starts with your history rather than our demo. We look at what your resolved claims already tell you about the ones still open, and build the profile from there.
