The market has already seen extensive lists outlining the myriad ways AI-powered digital solutions can improve data discovery, strengthen analytics, and ultimately speed and simplify underwriting and onboarding processes.
New data mining and analytical capabilities as well as new technologies are also bringing to light a broader range of personal metrics categories that are proving to be effective indicators of mortality and morbidity risk. Insurers are using these metrics in innovative new ways to refine underwriting, pricing, and product development.
It is important to be aware, however, that although incorporating these new tools will be an exciting step forward, maintaining appropriate risk mitigation frameworks remains an ongoing challenge. Insurers can turn to existing model risk management processes to evaluate the risks inherent in expanding the use of data via AI and other digital tools.
Optical character recognition (OCR) and natural language processing (NLP) technologies have significantly improved over the past two decades. Software now exists that can efficiently sift through and mine pertinent data in hundreds of pages of medical records information and produce a cogent report for underwriters – a process that until recently had to be done by humans. The enhanced contextual intelligence capabilities now available can also correctly correlate information in the report, even if, for example, the mention of a particular drug is many pages distant from that of the condition it treats.
These improved capabilities are enabling new alternative underwriting data categories, which are proving effective predictors of mortality and morbidity risk. The traditional underwriting factors of health status, financial status, occupations, avocations, and travel habits have now expanded to include education, residency and work locations, marital status, physical activity, premium payment frequency, past claims and credit information, and more.
Besides standard wearables data, which have been factored into risk assessment for years, underwriters are also incorporating new technologies. For example, remote photoplethysmography (rPPG) using a phone video camera can detect subcutaneous blood volume changes and through that, measure heart rate, heart rate variability, blood pressure, and more. Coupled with AI and data models, rPPG could derive more blood parameters, such as blood sugar, blood cholesterol levels, and more. Much of this is still under development and will need time to reach the accuracy levels necessary to be part of the insurance onboarding process.
Additionally, AI tools are making possible more precise population segmentation using the many new alternative data points. So not just product design and underwriting, but onboarding as well, can be customized for multiple cohorts of target consumers.
With new capabilities, of course, come new challenges. For example, AI programs must be trained to ensure the right information is gleaned from sources used (e.g., medical, prior claims, credit) so that the program can address pertinent underwriting needs. As this training can only be provided by humans, full program optimization will take time.
A second challenge is that new data categories can increase the risk of underwriting bias and unfair discrimination against certain groups (e.g., marrieds vs. singles). Regardless of the amount of new data, its value still depends on how well it reflects reality, and data processed by AI tools is susceptible to inaccuracies. Insurers need to be vigilant in identifying such errors to ensure unfair discrimination does not occur.
The governments of many Asian countries have been developing insurance industry databases that insurers can use to both to check a person’s health history and analyze market data. New automated underwriting products that leverage this new information are currently in development in several countries.
The ability to access and use larger and wider datasets that include several new types of data, enabled by significant advances in AI and ML technologies, is opening new opportunities for insurers. At this juncture, insurers must be sure they have in place a strong oversight framework. Rules need to govern what is and is not permissible, and must include factors such as customer consent, user access limitations, storage security measures, and secure data transfer capabilities.
Advanced digital tools and techniques can enable analyses, predictions, and recommendations, or even guide decisions, but as capabilities advance, so must the understanding that being able to do something does not mean it should be done. Technology can create and amplify asymmetry between available data and the ability to manage it, resulting in unfair bias. The industry must therefore pursue innovation in an ethical and transparent manner, accounting for differences in markets, regions, companies, and other relevant factors.
Also, as with any technology, AI-enabled tools bring a risk of overreliance. Insurers must implement processes to guard against oversimplification and balance AI’s imposed simplicity with data accuracy. Are the tools contributing to customers being treated fairly and equitably? Are the interpretations provided by digital platforms being properly analyzed and reviewed by on-staff experts?
Bottom line: Digitization can provide a range of benefits to insurers – automating repetitive tasks, speeding data analytics and discovery, simplifying information mining to providing a broader range of products and services, and more. However, insurers must also keep in mind that digitization is fundamentally a tool, and one that should augment, but not replace, human expertise, experience, and judgment.
