AI, ML, and modeling serve to augment human knowledge and free people to use their time more effectively. (Photo: everything possible/Shutterstock)
Thankfully, 2020 is behind us. No year in recent memory has been nearly as disruptive. This has caused many insurers to take stock of their business models and the processes that underpin them.
Many carriers are trying to reconcile how new ideas in artificial intelligence (AI) and machine learning (ML) can support their goals in tangible and beneficial ways. They realize that modernization is necessary but often don’t know when or how to start the process or what projects will achieve a positive return on investment (ROI). They also can’t strike the balance of where automation ends and human expertise begins.
Insurers collect a great deal of data during the underwriting process and put it to good use to inform pricing and risk selection. They also collect a lot of data during the claims process but are more challenged in using this data efficiently. In the not too distant past, one of us (Dylan Whitehead) was leading a claims organization at a major insurer. Often, when the underwriters asked their supervisors extremely reasonable questions about loss trends, they rarely got direct answers quickly. It wasn’t because of anyone’s ineptitude or a lack of understanding about the claims, but rather due to our inability to aggregate loss data from the disparate sources and systems and convert them into a succinct and actionable analysis. For example, if someone asked why a certain cause of loss had caused a spike in severity in a specific geographic area, the typical answer would be, “Give me a few days.” The supervisor would then spend many hours reading and analyzing unstructured data from various sources to come up with an answer while simultaneously sacrificing focus from other pressing tasks.
AI, ML, and modeling can enable us to eliminate “Give me a few days,” or better yet, can proactively provide the analysis before the questions are even raised. One of us (Eugene Kolker) for years practiced developing and leveraging AI, ML, and modeling in diverse industrial and R&D settings. This experience resulted in a recent introduction of a new MIA DAMATM approach, which stands for Management, Innovation, and Action (MIA) for Data, AI, ML, and Analytics (DAMA). This approach enables organizations to conduct proactive analyses, such as what the underwriters desired in the above use case.
Effectively using data to improve ROI
Whether or not you find yourself nodding sympathetically with Dylan’s experience, the chances are that you feel some level of frustration with your ability to effectively utilize your collected data in a real and tangible way. There is nothing more concrete to an insurer’s balance sheet than loss reserves. They can be a source of consternation and controversy, and many actuarial or claims managers have had to answer for inaccuracies that resulted in real consequences to the business. Despite proper reserving being a key to ROI and profitability, plenty of organizations continue to rely on relatively unsophisticated methods for statistical reserving. Typically, these methods are based on historical averages of paid claims, time from report to final payment, or some combination of the two. This is better than nothing, but companies can achieve so much more through AI, ML, and modeling.
Today, in many setups, only a few data points will inform the methodology used for most statistical reserves. They are probably calculated once every (couple of) quarter(s) and modified, as necessary. Spirited discussions take place over whether it will result in redundancy or deficiency. At the end of the day, much time is wasted, and usually, nobody is better off for it. But what if you can use tailored statistical models, augmented by ML, and supplemented by AI, to provide a better, more sensitive, robust, and nuanced answer? And it is not as complex or costly as it might sound, so ROI can become very significant.
Let us consider auto claims, in general, and collision claims in particular. You might know how much you paid on average for collision claims last year. You also might break this down further by how long claims remain open, and you might think you have a good idea of what your final cost might be. But does your reserving methodology consider the year, make, and model of the vehicle? How about the zip code where it is being repaired? Or whether a preferred or non-preferred shop is chosen? Or what are the driving patterns of the parties involved? Similarly, in the case of bodily injury claims, how does the age of the injured party impact your exposure? The hospital where they received treatment? Even the specific attorney they hired. Nowadays, the limits are only what you decide them to be.
Yes, the claim might be settled and paid in a few weeks or months, but in high-frequency lines of business, an extra week or month of being able to accurately project your costs can make a huge difference to a quarter’s performance. It can reduce volatility and create an accurate snapshot of a book’s performance down to the day! A statistical reserve supported by AI, ML, and modeling, enabled by MIA DAMATM, can soften curves and create an unbiased process that is fully supported by data and analysis to know sooner and more accurately how much claims will cost. Importantly, none of this replaces the wisdom of an expert claims’ handler putting in a manual reserve. AI, ML, and modeling, like other great technologies, serve to augment human knowledge and free people to use their time more effectively.
This sort of easy application of AI, ML, and modeling will make many lives easier now. When business leaders realize how simple and cost-effective this kind of modeling can be, they will be utilizing it much sooner. What about you? Are you ready to try AI and ML to significantly improve your ROI?
Dylan Whitehead is CEO of Alloy Claims in Dallas. Eugene (Gene) Kolker, Ph.D., is executive vice president, Global Enterprise Services & co-director of the AI/ML Center of Excellence at DataArt. Kolker also serves as a professor in the Technology Management & Innovation Department at New York University’s Tandon School of Engineering. The opinions expressed here are the authors’ own.