Faced with the explosion of digital threats and the increase in climate risks, insurers are increasingly capitalizing on AI to better understand risks and fight against fraud.
But in an increasingly strict regulatory context, in particular with the GDPR and EU AI Act in Europe, as well as the action of the ACPR (the prudential and resolution control authority) in France, fluid access to actual data remains a major challenge. This is where synthetic data play a major role.
A response to data and data governance challenges
Did you think insurers had an abundance of data on their users and on the different risks they face? The reality is more complex. If the data exists, the stake lies in its governance, accessibility and exploitation. Indeed, despite major modernization investments undertaken for almost ten years, insurance companies must always compose with historical IT infrastructure (Legacy Systems) which slow down the fluid information circulation. In addition, European and French regulations, GDPR, right to be forgotten and specific requirements of ACPR, impose strong constraints, particularly on the use of sensitive health data. Fortunately, synthetic data allow you to get around these obstacles. Artificially generated from statistical models and machine learning, they faithfully reproduce the characteristics of real data without exposing sensitive information. By 2027, IDC provides that 40 % of the algorithms used in insurance will systematically integrate synthetic data (1).
Optimize fraud detection and anticipate the risks
With 695 million euros in total fraud identified in 2023 (2), insurance fraud is a constantly evolving scourge. According to the French Insurance Federation, it represents almost 10 % of compensation paid each year in the country (3). In order to improve the detection capacity of fraud in real time, a major issue to limit financial losses, synthetic data makes it possible to cause models of AI on massive volumes of scenarios, real or unpublished. Insurers can thus simulate millions of compensation requests. They are then able to identify abnormal schemes, assign risk scores and trigger in -depth analyzes, even before fraud is carried out.
Better understand the new climatic risks
Another concern for insurers, the climate crisis. Faced with this major issue, which upsets insurance portfolios, a total overhaul of risk assessment is essential. Droughts, floods, storms: extreme weather phenomena have been multiplied by 2 in 20 years according to the UN (4). Consequently, historical data are no longer enough to understand and anticipate these unpublished risks. However, the pooling of data between insurers – advocated by the Langreney report mandated by the French government – remains difficult to implement. Synthetic data here offer a precious alternative: they allow extreme events to be simulated in geographic areas where real wallets are still little exposed. This approach helps insurers better model the climatic impacts on their future commitments.
Precautions to be taken and success factors
If the advantages are numerous, several pitfalls should be avoided. Synthetic data remain dependent on the quality of the real data used upstream. Without prior work on biases correction, the risk is real to reproduce and amplify existing errors. It is the famous adage “Garbage in, Garbage Out”: an AI, even powerful, will only have value if it is powered by quality data. To succeed, it is crucial to adopt a structured approach around the “value, governance, technology” triptych. Each project must be based on clearly identified use cases (fight against fraud, climate modeling, improvement of customer experience, etc.) and ensure that the requirements of compliance, ethics and performance. Finally, democratize the use of synthetic data within all business functions – from subscription to compensation – will promote their adoption and maximize their impact. This is the reason why synthetic data generation platforms must be accessible not only to data scientists, but also to news, claims managers and business departments.
As the new data regulations, the requirements of transparency, ethics and sovereignty are increased. Insurers who have been able to take the turn of AI and synthetic data will be better armed to meet the challenges of tomorrow. More broadly, synthetic data has become an essential strategic lever for resilience, competitiveness and survival of the insurance sector.
(1) IDC Futurescape: Wordwlide Insurance 2024 Predictions – Asia/Pacific (Excluding Japan)
https://my.idc.com/getdoc.jsp?containerid=prap52003124&utm
(2) Alfa Study (Agence to Combat Insurance Fraud) for 2023
(3) French insurance federation, 2024 report on insurance fraud
https://www.ffa-assurance.fr
(4) UN, 2022 Report on Climate Disasters: https://www.un.org/en/climatechange
