November 20188What have we learned about AI?By Gary Anderberg, SVP, Claim Analytics, Gallagher-BassettIN MY OPINIONNew technologies can be very seductive. We all know the process. A new type of application begins making waves in some areas of financial services. It becomes the buzzword du jour. Articles about it proliferate. "Predictive analytics" went down this path a decade ago. The dazzling new tech of a few years back was Artificial Intelligence (AI). In this article we'll look at what we've learned about using AI in the real world of insurance claims handling, what problems it addresses successfully and where it has come up short.Real AI is complex and expensive, even now. It is not a magic solution for every problem, so the first question is--do we really need to use AI to address the problem we are trying to solve? What AI does well is to build models based on sophisticated pattern and correlation recognition which can then be applied to identify probable outcomes, given a new universe of similar initial conditions. The obvious applications are in anticipating claim costs and other outcomes based on the initial data set for a new claim. Every new claim is a puzzle. How will it develop? Will it require some form of intervention? What will it cost?In our experience, AI, after considerable development and tinkering, has proven to be a much better tool for estimating claim reserves and predicting the need for specialist interventions than the classic rules engines it has replaced. In our world AI is most effective in the middle of the bell curve and less useful with very small dollar claims and with cat claims. Even with an extremely large data base, we have
< Page 7 | Page 9 >