One of the most exciting areas in artificial intelligence or machine learning (AI/ML) for insurance is the development of dynamic of “living” processes. These use AI to make decisions and continually adjust to produce the right outcomes. To better see this in practice, let’s consider a simple case.
For most commercial insurance business, underwriters receive more submissions than they have time to process. Many carriers use a triage process to help the underwriter know which submissions should be pursued first, based on their chance to successfully write the business. Traditionally, carriers have used analytic tools, rules or template-based criteria sourced from the carrier’s experience.
While these solutions are effective, they are static and cannot adjust to changing market conditions without being refreshed. The living process solution is to use a machine-learning solution to determine a positive outcome probability. The solution would initially come based on the carrier’s experience, but by feeding back the most recent win/loss experience, the solution can constantly adjust to changing market conditions.
These type of living processes offer tremendous potential for insurers to move beyond the static and complex rules most carriers still use today. These changes can then offer from two to 10-times the benefits. That said, as we have seen from recent news AI and machine-learning solutions also come with some risk.
For example, ahead of Hurricane Irma, we have seen many cases where dynamic pricing hasn’t worked as well as it should have. The straightforward task where machine learning generated optimized pricing saw rapid price increases to ridiculous levels. These new AI tools were like eager beavers in the quest to optimize price without fully understanding other needs.
As carriers increasingly make greater use of AI and other intelligent processes, they need to ensure they have the proper controls, guardrails and governance so that these intelligent solutions aren’t eager beavers that inadvertently do more harm than good.