Insurance
The file has been open eleven days because the adjuster's report arrived as a scanned PDF nobody has keyed in, and the policyholder has called three times. Inferana reads those reports inside your perimeter and logs every query.
What each regulation in your sector requires, and which part of that the deployment solves.
Digital operational resilience and ICT third-party risk management for insurance undertakings.
Auditable EU provider, dedicated or on-premise deployment and OpenAI-compatible API.
Health data and third-party data inside a claim file demand minimisation and reinforced confidentiality.
Zero logs by design: prompts and responses are never stored or used to train models, with access control per user and team.
Risk assessment and pricing in life and health insurance is a high-risk system (Annex III).
Open models with published model cards and licences, access controls and a record of every query to document human oversight.
Advice duties, pre-contractual information and traceability of what was advised in insurance distribution.
An auditable record of every query and of the answer the intermediary received, exportable when the customer complains.
The same ones you already use, specific to your sector.
Nobody approved it, but claim forms and adjuster reports already pass through outside tools to close files sooner. What follows is always the same:
A file takes twelve days and eight are spent waiting for someone to read the report.
Head of claimsEvery adjuster sends their report in a different format and we key it in by hand.
Claims back officeThe policyholder rings about their claim and we take ages to say where it stands.
Policyholder servicesUnderwriting a large risk is three days of reading reports.
UnderwritingThe team pastes claim reports into an outside chat to move faster.
Technical managementThere is health data and third-party data in that file I cannot control.
Data protection officerLoss adjusters and brokers work outside, with their own tools.
ComplianceWherever you decide: on European infrastructure we manage, or in your own data centre, on-premise. In both cases prompts and responses are never stored or used to train models, so the file leaves no copy outside.
Technically yes: we store no prompts or responses and the deployment can stay inside your perimeter. The legal basis, the impact assessment and the decision on Article 9 special categories belong to the insurer, which remains the controller.
The service and deployment documentation: which functions it supports, where they run and with which dependencies. Filing it in your register of contractual arrangements and keeping it current remains the insurer's job.
That use falls under Annex III of the AI Act. Classifying the system and taking on the high-risk obligations is the insurer's responsibility. We provide open models with a published model card and licence, access controls and a record of every interaction, which is what you need to document traceability and human oversight.
Through access per user and per team, with every query logged. An external adjuster gets their own scope and is traced like any internal handler. Revoking access stops the queries, and the history of what they asked stays with you.
It keeps who asked what and which answer came back, timestamped, and it exports. That is the traceability the IDD expects you to hold over advice. Whether that answer reached the customer unchanged is evidenced by your own sales process.
Every adjuster hands over their own thing: native PDF, a scan of a signed report, captioned photos and a separate repair quote. Damage, cause and amount are pulled out of that material, and anything doubtful is flagged for the handler to look at. You define the field template, and the document never leaves your perimeter.
Yes, with the same key. There the query does go out to the model provider, so a file with health data is handled with the open models inside your perimeter. It is configured per team.
The models are open and the API is OpenAI-compatible, so your exit plan does not depend on us: you can take the weights with you and point the code elsewhere by changing a key. Your data and your indexes are exportable.
In the demo we go through your case: which models you need, where they run and what it takes to meet the regulation that applies to you.
We reply within one business day.
