Underwriting and anti-fraud
An application is checked by rules and scoring models. If there are no rules or the model is uncertain, the application goes to the underwriting queue instead of being approved automatically. The same models assess the fraud risk of applications and claims.
Plan Rules on all plans, anti-fraud scoring from the Professional plan.
What is inside
- Five model types: rules, logistic regression, ONNX models, an external model and a hybrid
- Models, rules and features are configured in the interface
- Underwriting queue: a manual decision on deferred applications
- Statutory limits for compulsory lines and a check of the agent’s authority before issuance
- Fraud scoring of applications and claims
- Claim notification parsing by a language model: Claude, ChatGPT or your own model. The model sets priorities, and the decision stays with a person
- A closed list of fields that are sent to the external model
- Scoring of recovery candidates
Conditions
Claim notification parsing by a language model is off by default and is switched on by the insurer, who chooses the provider and the list of fields passed.
Frequently asked questions
What happens if a product has no underwriting rules configured?
The application goes to manual review. The system does not approve a contract automatically if it has nothing to check it with.
Can we connect our own model?
Yes. A model in ONNX format is uploaded to the system, and an external model is connected as a service. They can be combined with rules in a hybrid scheme.
What data goes to the language model?
Only fields from a closed list approved by the insurer. The model makes no payout decisions, it only helps set priorities in the queue.
Other features
Shall we show the system on your products?
We will build your product in the builder, take a policy from quote to claim and estimate the implementation cost for your lines of business and countries.