Snapsheet

Strip the models out and a claims system still runs from first notice of loss through settlement, which is what carriers buy it for. Virtual appraisal started with remote human appraisers reviewing photos, so automation improved a workflow that already worked.

Agents either fire automatically inside event based workflows or wait for an adjuster in a side panel copilot, and the carrier directs their actions in plain language instructions. That makes the customer the author of the guardrails, and no approval class, referral threshold or default limit is published.

Every plain language prompt change alters the behavior of a system making evaluation determinations, and nothing published describes version control, pre launch testing or approval for those changes. No accuracy, benchmark against adjusters or drift monitoring is published for any agent.

More than 4.3 million claims and $15.3 billion of indemnity processed size the money moving through the platform, though both figures come from a 2023 report not restated since. IAT Insurance Group is the clearest named customer, and no accuracy, automation rate or expense saving is credited to the AI.

A carrier designates which data an agent may use when writing its prompt, so the boundary around sensitive claim material sits with whoever configures it, and no default restriction is described. Whether claim data trains models is unaddressed, and a third party report that models learned from millions of past claims finds no confirmation in company material.

Claimant photographs, adjuster notes and disbursement instructions combine personal, evidentiary and financial data about people who are not the customer. A privacy policy is linked site wide, but its contents are not reflected here, so no retention or claimant data commitment is confirmed.

A Trust Center linked from every page gives a security team a named destination, a better structure than a request queue. Its contents are not reflected here, so no SOC report, certification or penetration testing statement is confirmed.

Claims handling is examined at state level for timeliness and fair settlement, and the company once employed appraisers, an activity several states license. Yet no audit trail capability, state requirement support or position on the commissioners' AI guidance is described.

Liability check agents assess fault, which sets whether a person is paid or pursued, and fault drawn from claim narratives is where language models absorb historical handling patterns. The company publishes no fairness testing, bias evaluation or model card.

Communications agents write to claimants who sit outside the contract, and nothing published tells them an agent shaped the figure they were offered or how to contest it. Because the carrier configures what the vendor built, responsibility for a defective output divides between them, and nothing allocates it.

Writing agent personas and actions in plain language is the interface of a general purpose language model, yet no provider, family or version is named. Floatbot is disclosed as the conversational agent partner, which leaves the claims agents themselves unattributed.

The platform runs claims through to payment disbursement, so the vendor sits inside the settlement transaction rather than handing off before it. A partner network above 140 and a joint stack with Socotra extend it, but policy administration, billing and underwriting are outside its scope.

A typical installation takes 45 days, about 30 of them spent on business rules and training, an honest breakdown from an interview several years old. Hosting provider, regions and tenancy are all unpublished, and no private deployment path appears.

The early appraisal business charged per estimate, the only known basis signal, and nothing says whether the platform kept that transaction shape or moved to a license. Whether agents cost extra, and how payment disbursement is priced, are both unaddressed.

Carriers, third party administrators and self insured organizations all buy it, a fit for a market where the party handling a claim often is not the one carrying the risk. Auto is where depth sits, extending through property and commercial claims, and no presence outside the United States appears despite stated expansion plans.