{"id":1346,"date":"2023-01-18T16:35:36","date_gmt":"2023-01-18T16:35:36","guid":{"rendered":"https:\/\/tst418075918.wordpress.com\/2020\/06\/23\/how-will-insurance-regulators-evaluate-ai-black-box-underwriting-decisions\/"},"modified":"2023-01-18T16:35:37","modified_gmt":"2023-01-18T16:35:37","slug":"how-will-insurance-regulators-evaluate-ai-black-box-underwriting-decisions","status":"publish","type":"post","link":"https:\/\/test.veracityid.com\/new\/how-will-insurance-regulators-evaluate-ai-black-box-underwriting-decisions\/","title":{"rendered":"How will insurance regulators evaluate AI\/Black Box underwriting decisions?"},"content":{"rendered":"
<\/div>\n

There is quite a bit of auto insurance industry activity and discussion around using AI machine learning models to identify bad risks during the underwriting process. The idea is that AI tools can tease out subtle variations in customer status and behavior that make a significant financial difference. But there’s a catch.<\/p>\n

The Legal and Regulatory Risks of AI Machine Learning<\/strong>
\nGiven that insurance rates are tightly regulated, using AI machine learning to evaluate customer risk means AI models will be used to deny coverage to customers that do not ‘measure up’. Which is a problem. Because customers expect carriers to have a specific, rational reason for why they are being rejected. And regulators and courts become concerned when certain groups, for example minority customers, are rejected at higher rates than others. Unfortunately, AI machine learning models cannot explain why<\/em><\/strong>\u00a0they make a particular decision\u00a0because they are ‘black box’ solutions. Their creators literally cannot explain how the machine arrives at its decisions, they can only judge the results.<\/p>\n

This means relying on AI models to sort the ‘good’ customers from the ‘bad’ ones carries large potential regulatory and legal risks should its decisions result or even just be suspected of resulting in disparate impact for a disadvantaged class.<\/p>\n

How VeracityID Uses AI<\/strong>
\nWe use AI models to help carriers sort between good and bad customers too. But we only use them as a ‘Tip and Lead’, as an indicator that a given customer may pose a greater than normal fraud risk. We then use our other tools such as graph analytics, business rules, mobile documentation apps and real time automated interaction with the customer to identify and resolve specific real risks. The result is that when a carrier using VeracityID tools declines to cover a customer they can give a specific reason that\u00a0makes sense to both customers and regulators:<\/p>\n