Munich Re has already successfully piloted such an application: “Our Early Loss Detection (ELD) platform continuously evaluates over 16,000 news sources and searches for previously defined loss events. Using geocoded risk and loss information, the system checks whether insured properties may be affected”, explains Wolfgang Hauner. “If this is the case, those responsible for the system automatically receive a message containing all the relevant information. Property losses can thus be recognised swiftly, and appropriate measures taken enabling us to drastically reduce the scope of claims management”, says David Feghelm, Senior Solutions Manager Claims at Munich Re. 

But ELD’s functions go beyond the rapid identification of loss events. All individual losses are stored and collated in a database; every case brings with it new information on causes and impacts. Using this data, future loss trends can be recognised earlier and modelled with greater precision. ELD is thus a powerful tool helping insurers reduce costs by getting ahead on claims management, while also gathering information that will be of use when modelling new covers in the future.

Alongside ELD, Munich Re has also developed further prototypes and products which are being tested or are already in use. One of these is Vision Intelligence, an AI-based loss-recognition system using photographs. This is based on neural networks (also used in areas such as language recognition systems and medical diagnostics) and supports insurers in the categorisation of property losses due to natural disasters, for example, and in ascertaining the approximate scope of losses. 

The risk management platform M.I.N.D., another successful pilot project already successfully in use on the market, helps insurers, reinsurers and loss adjusters to create transparency on risks together and better appraise these. In this context, AI also helps identify potential cases of insurance fraud. M.I.N.D. documents historic information on losses and compares this with current pictures of damage. Attempted fraud can thus be more easily identified and better prevented, and consequential losses can be compared with loss events that have already occurred.