Fraud detection in insurance
Detect fraudulent providers in insurance sector
Challenge
Most insurance contracts include a breakdown assistance clause for their insureds. In the event of a breakdown or an accident, companies may offer several services such as a taxi, a rental car or a towing service. These services are provided by external service providers and are relatively expensive for the companies.
As with all services, there are abuses: either on the part of the insured or on the part of the external companies that perform the services for the insurer.
Insurance fraud on the part of breakdown service companies, replacement car hire companies and taxis amounts to hundreds of millions of euros per year.
These frauds can take various forms: modified mileage, modified route, modified surcharge, etc.
How to detect frauds upstream and gain in profitability?
Solution
Three major steps are necessary before applying our AI model to detect fraud: Data Collection, Data Preparation and Data Enrichment. These three steps will be applied for 3 datasets:
Taxi assistance, car rental assistance and towing assistance. At each step, a Dataviz synthesizes the intermediate result.
The data from the different service providers and the insurer’s operators are all ingested in a Data Lake. An intermediate DataViz allows the identification of the proportion of missing and erroneous data. The real challenge is therefore to qualify all these data in order to obtain the best possible result.
The data is then cleaned and standardized to ensure its quality. Poor quality data would result in too many false positives and noise due to the use of these poor quality data. Once the data has been prepared, we enrich it to improve the results and to limit, for example, collusion between the operators’ data and that of the insurer’s suppliers. The data is enriched with external data from Open Data, geocoded data… Our enrichment model allows us to add more than 1000 additional features to further qualify the data.
These enriched data are then submitted to our fraud detection model. This is a supervised ML model based on scorecards.
The results are obtained in real time and can be shared from the tool to management, the network team, the middle office and the leakage team.
After six months of use, the model has saved the insurer money :
Towing: €1 million // Taxi: 700K€ // Car rental: 500K€
Why Octopeek?
The insurance industry needs to move from ad-hoc experimentation with AI to enable everybody within the organization to use it to improve both operational efficiency and deliver business value thanks to true business apps.
Octopeek aim at becoming the platform that democratizes AI in the enterprise: it makes possible a dynamic creation and provisioning of full-fledged enterprise AI apps, customizable by business analysts.
