Application Fraud Modelling
We build the models that identify misrepresentation and fraudulent applications at the point of sale, or validate the models you run.
Insurance
Fraud
Detection at two points: misrepresentation at the point of sale, and the fraudulent or inflated claim as it is notified.
Fraud Deliverables
Claims Fraud Modelling
We build the models that identify fraudulent and inflated claims as they are notified, reading the claim, policy and customer history that precedes them, with alerting into your investigation workflow.
Frequently Asked Questions
-
Fraud detection is judged by what reaches an investigator in time, so ask what the output does to their workload. At Gini we build predictive models that flag the suspicious claim before it is paid, and graph analytics that identify clusters of related claims, which is where organised fraud is visible and where claim-level scoring is not.
-
Yes. Gini tests detection rate at your operating false positive rate, stability over time, and whether the features still carry signal now that the behaviour has moved. Fraud models decay faster than most because the people they detect adapt, so validation looks hard at when the model was last refitted and what alert volume has done since.
-
At Gini we measure each rule’s and each feature’s contribution to confirmed outcomes against the alerts it generates, then retire or retune what produces volume without cases. Every change is evidenced against historical claims before it goes live, so the effect on detection is known rather than hoped for.
-
Relationships. Shared addresses, bank details, vehicles, repairers and contact points linking claims that each look ordinary. A cluster of eight related claims is obvious in the graph and invisible one claim at a time, which is why Gini runs the two approaches together rather than in competition.