AI can treat customers as individuals

Data science and machine learning offer opportunities to provide better pricing and reduce fraud as well as offering other benefits, Airmic delegates heard at the ‘How can AI change the risk landscape for businesses and their insurers?’ workshop led by QBE at the Airmic conference.

Presenters said that natural language processing can help insurers extract relevant features from the text to detect fraud. This is a crucial advance for insurers as around 10 percent of claims are fraudulent. The technology can also analyse this data to find more cases of fraud by revealing links.

Machine learning can also pick up on trends and patterns that more traditional actuarial processes cannot. While actuaries tend to look at an aggregated level of data, machine learning is capable of a more granular analysis.

From an insured’s perspective this offers a better way to sell the importance of risk management internally as it can help to drill down into the information and offer the opportunity to treat customers as individuals.

A real-life example of this kind of artificial intelligence (AI) is Netflix. The service has a large data science team dedicated to keeping viewers on the platform for as long as possible by recommending things they like to watch.

Another example offered in the session was Google Vision API, which analyses data from images and can glean a huge amount of information from one photograph. From a photograph of a person it can identify emotions, gender, background setting and many other data points. It can also recognise foods, items, and even health issues and make recommendations.

AI can help insurers predict whether a company will be liable for a particular claim, potentially also speeding up the payment of a claim.

As a commercial insurer, QBE uses algorithms to support underwriters by making recommendations. For example, pricing might be adjusted favourably for clients who are good risk managers.

However, there are “dangers”, the workshop presenters warned. Beware spurious correlations, they said. They also discussed the example of a Microsoft chatbot that was allowed to interact with other users on the internet and ended up making sexist and racist comments.

The AI could not distinguish that it was inappropriate, so it had to be taken down.