Cinimex implemented a ML project for Rosgosstrakh
Cinimex together with Rosgosstrakh implemented a complex project using latest data assessment methods. Data analysis specialists developed machine learning-based services enabling an insurer to assess risks, forecast severe losses and evaluate portfolios of agents, brokers and partners more accurately.
The Cinimex team developed a geographic segmentation model. The algorithm enables risk optimization through handling selected client segments and geozones. The risk assessment involves historical data regarding losses and territorial features from open sources. Clustering is sensitive to free zones where an insurance company is poorly represented and has the potential for increase in sales. In the long run, the analysis can be extended by including demographic and other statistics.
Artificial intelligence enables efficient portfolio evaluation of agents, brokers and other partners dealing with the Company. For example, a change in agent’s behavior or execution of nonstandard contracts can indicate an increased risk. To forecast the loss ratio, the system assesses and forecasts potential sales for different periods.
“A key feature of this project is the segmentation itself, both by numerical input data (many sources) and by connection of each client to a geolocation (geosegmentation). Each of subtasks is covered by different machine learning models, which shall efficiently operate not only in standalone mode, but also in gear,” comments Rodion Martynov, Project Manager in Cinimex.
“The choice of the technology was influenced by the highly competitive environment and permanently expanding assortment of financial products. A contributing factor was the increase in the number of clients and data growth concerning each of them. We asked Cinimex for development from scratch because we were facing goals that cannot be achieved just by using available boxed solutions. Most boxed solutions do not take into account all data and cannot be adapted to the unique features of our company (for example, predictive analysis based on cartographic information and sequences). Within this project, what we needed was a project design initially based on the company’s data and particular business rules. Also, we ascertained that the decision to use mostly open source software and independent developments in the project was correct, taking into account the import substitution trend,” says Olga Veresova, Head of Analysis and Control, Rosgosstrakh Insurance Company.