After more than two years, the global pandemic has brought both challenges and opportunities for health insurers. It has accelerated trends that are now reshaping the way insurance is underwritten, distributed and managed. At the same time, some of the problems that have challenged the industry for years still exist.
To peel back the onion on the insurance industry, we are going to examine previous, current and future assumptions and how they actually played out in the real world. We will also look at using artificial intelligence (AI) to challenge future assumptions as the industry and our world continue to evolve.
Making Assumptions
Actuaries and underwriters make tons of assumptions every day, so much so that we forget that we’re even making them. Three key assumptions make up the backbone of our work in health insurance:
- Law of Large Numbers – large data sets tend to behave in a predictable way
- Claims Trend – healthcare costs are always rising
- Past Is Predictive – the underlying stochastic processes that influence health outcomes move slowly from period to period
When calculating the renewal of an insurance group or forecasting claims into the next 12 months, we almost always rely on the most recent couple of years’ experience as the basis of our calculations—typically applying more weight to the most recent year. Under normal circumstances, this is the prudent and best practice. But we are not living in normal circumstances (don’t worry, I’m not going to call it a new normal), and the COVID-19 pandemic threw our ability to make assumptions for a loop.
COVID Realities - Medical Care
During the COVID pandemic, especially in the first half of 2020, the pattern of healthcare use across the country changed dramatically. For one, providers rushed to adapt to an influx of COVID cases. In doing so, they diverted people and resources from normal operations and tried to figure out how to charge for the new services and care protocols that didn’t exist just a few weeks prior.
Simultaneously, people abruptly stopped going to the doctor for routine and preventive care, either because their doctors were discouraging it or they wanted to avoid exposing themselves to the virus unless absolutely necessary. Additionally, patients with acute issues thought twice about going to the ER.
In just the first seven weeks following COVID shutdowns, surgical volume dropped 48%, although by July 2020 that volume was only 10% less than we saw in 2019. Overall, millions of non-acute procedures were postponed indefinitely or canceled. And because different regions addressed the pandemic in different ways, insurance agencies lost the ability to make blanket statements about the way their clients would be affected.
To continue providing necessary care to their patients during this time, providers turned to online and virtual tools to help diagnose and treat patients. And while telehealth existed before the pandemic, it wasn’t widely used in the same way it is now, meaning many healthcare facilities had to learn how to bill appropriately for it.
See also: Optimizing Insurance's Role in the Pandemic
COVID Realities - Insurance Industry
Post-pandemic, the insurance industry held its collective breath, waiting to learn about the financial impact. As the claims started rolling in (or not rolling in, as was actually the case), it became clear that something unprecedented was indeed happening.
Claims submitted for March and April 2020 were far below historic or forecasted levels. Payers and insurance company risk managers were happy the pandemic didn’t spell doom for the insurance industry.
As the pandemic continued deeper into 2020, those of us in the industry continued to watch the claims feed, expecting there to be some sort of a correction. We wondered when the claims would just return to “normal” levels or even run high as the market over-corrected for the low claims period. By the time it came to calculate plan renewals and year-end reserves, the overcorrection never came. The assumptions weren’t working as they had in the past.
How to Address
The common practice for actuaries and underwriters was to go ahead and use the prior year’s claims in renewal forecasts but apply a simple adjustment (usually 4%) to account for the dip in utilization based on guidance given by the Society of Actuaries.
On the surface, the experience from March 2020 through the present day has been such a large departure from the past that it appears to be unusable in traditional actuarial models. But is that really the case? And is applying a simple load to the experience really the best we can do to make the data usable?
While the historical data doesn’t quite fit the industry’s current needs, AI can improve actuarial models to help insurance providers make more accurate estimates. Artificial intelligence gives underwriters a more complete picture of risk, separating good risks from bad risks and helping them price policies accordingly. Additionally, AI can make these predictions faster and draw deeper insights from the data, further improving business decisions.
AI Works Where Assumptions Can’t
Traditionally, underwriters have used decades of historical information to develop rules and guidelines to assess risks. However, if the relevance of historical data diminishes over time, it may not accurately predict future trends and exposures, resulting in poor risk assessment and inaccurate pricing. For example, relying on historical loss experience to write natural catastrophe risks used to be considered adequate. But it may be insufficient in the future; changing climate, urbanization and increased asset concentration in climate-exposed areas could significantly alter risk patterns.
The insurance industry relies on accurate underwriting to remain profitable, and underwriters have always relied on data to make decisions. When the normal layers of data aren’t there, underwriters have to be able to turn to artificial intelligence to fill in the gaps. Artificial intelligence can identify trends and outside factors faster than humans, allowing it to provide more accurate predictions when historical data doesn’t fit current patterns.
