Big data for insurers: The 'apps' that matter
Insurers with apps that better align price with risk are more likely to find competitive advantage.
Remember when an application — better known in today’s tech shorthand as an ‘app’ — was a sheet of paper used by insurers to consider a consumer’s qualifications for a policy/?
Then smartphones took hold, and the app’s definition suddenly evolved to include digital applications used to review coverage terms or submit a claim. These apps have become popular among millennials and others who value quick, decisive outcomes in their insurance transactions.
Related: 16 of the coolest P&C insurer mobile apps
The app’s story doesn’t end there
As the definition continues to evolve, the word app also now refers to almost any business application of insights derived from data.
The bad news for insurers is that the value of many apps has been fleeting, regardless of definition. Abandonment is well known when talking about apps for coverage, with many consumers opting out before binding.
In addition, the average smartphone app sheds approximately 80% of its users within 90 days, and insurance-themed apps are on the lower end of retention compared with “core apps” such as calling or texting.
Further, apps of data-driven insights in insurance and elsewhere have struggled to pick up steam, with many big data projects falling short of expectations.
In data, one size does not fit all
Yet, there’s a bright side found in apps that fall into two key categories: prediction and exposure modification. While the most extensive uses have been predictive exercises such as pricing, history suggests exposure modification may offer many equally potent apps as well.
When considering the evolution of data volume, velocity, and variety (the three Vs often used to describe big data) in insurance over time, velocity may be the most powerful decision point that informs project selection.
But an ideal strategy may be better drawn from whether an occurrence is frequent, infrequent, or somewhere in between. At either extreme — infrequently (e.g., insurer insolvency) or frequently (e.g., driving behavior) — data science may be deployed toward exposure modification.
In contrast, the remaining middle ground of frequency (such as in claims) is more likely fertile ground for prediction.
In visual terms, Figure 1 (at right) displays various data sources that have been introduced over the years, along with estimates of potential value from a prediction or exposure modification perspective.
Any such decision is grounded in sound exposure modification principles. Excessive rates can cause policyholder attrition, whereas inaccurate ones can drain surplus. Either one of these conditions could trigger something that’s rarely observed and never welcome — that is, insurer insolvency.
Related: Buying insurance? There’s an app for that
Made for prediction
As data has gotten ‘bigger’ over time, its use in the industry has evolved away from exposure modification toward more predictive apps. Everyday occurrences such as claims are observed frequently and are often predictable. There’s perceived fairness and business sense to pricing and underwriting based on expected losses. Insurers with apps that better align price with risk are more likely to find competitive advantage.
Carriers that, for instance, choose not to surcharge policyholders having prior claims will likely become inundated with business from the claims-prone population and experience profitability deterioration, which is the definition of adverse selection. These lagging insurers are forced to refine their predictions in kind, after which, leading insurers seek new and larger sources of external information to retain an advantage, setting in motion a data spiral.
Data such as Public Protection Classifications (PPC®) and vehicle symbols hearken back to the origins of apps’ exposure modification by conferring incentives to improve fire safety practices or purchase safer vehicles.
Others, such as credit history, have clear and significant prediction potential but little or no obvious loss control incentives. These analyses rely on prior information — that is, “given condition X, outcome Y is expected to occur.”
The opportunity to benefit relies on attaining a balance between having enough information to derive a reasonable expectation and enough time to put a related decision into practice.
The surge in exposure management
Consider two events that occur relatively infrequently and lend well to common apps of data that may modify or manage exposures.
First, the graph above indicates that insurers have long conducted rate reviews using aggregated premium and loss data to project redundancies and shortfalls in revenue, which then helps them modify pricing accordingly. Finding a proper rate level reduces this risk for the insurer and any policyholder’s resulting risk of not being indemnified after an unfortunate event.
A second app geared toward infrequent events and exposure modification is catastrophe modeling. Catastrophes (also known as “cats”) can cause insurer insolvencies, but extreme cats of any variety generally aren’t observed more often than every thousand years. For example, no one truly knew what a Hurricane Katrina and its price track, landfall, wind speeds — and destruction — looked like until they’d witnessed one. So, looking at data derived from the immediate past 100 years provides only minimal insight.
Cat models address a lack of historical data by simulating millions of possible events using meteorological, geological, and structural data. This data can be diverse and deep. And although models can’t predict something inherently unpredictable, they can allow insurers to more judiciously manage concentration and surplus in recognition of the possible.
While frequency supports better-informed predictive apps, the more frequently something occurs may actually lead to less value in predictive apps: Why predict something and, for better or worse, reflect it on someone’s renewal bill six months down the road when you could potentially address it more effectively in the moment? This realization has led to a surge in apps that leverage big data’s strengths for exposure modification.
Internet of Things (IoT) technology aptly illustrates this premise. Insurers have made increasing use of IoT data over the past two decades. Usage-based auto insurance (UBI), for example, offers discounts to policyholders who provide data that shows they operate their vehicles safely and at appropriate speeds relative to road conditions.
With pay-per-mile UBI, a driver could reportedly reduce premiums 50% or more by operating a vehicle sparingly. This effort essentially turns each mile into a decision for a policyholder, creating frequent opportunities to modify exposure. An insurer could also conceivably incentivize for policyholders to relocate vehicles when sensors detect an imminent weather catastrophe. While price competition is one motivating factor behind IoT usage in insurance, greater potential may yet reside in these apps that help policyholders reduce their risk.
Availability of sensor data may, of course, lead down a path to exposure modification overkill, just as today some insurers may be arguably over-investing on prediction to the detriment of exposure modification. Autonomous vehicles (AVs, as seen in Figure 1) in one sense represent UBI in its perfect state, because they crunch sensor data to manage the risk of accidents down to near zero—in the process potentially eliminating insurers from the equation, should the legislative landscape ever permit that. But they also present new exposures — such as cyber risk— which require management or modification.
To find a true calling with big data, insurers would benefit from exploring apps that hearken back to insurance’s early exposure modification focus while remaining aware of new exposures that may come with the territory.
Jim Weiss, FCAS, MAAA, CPCU, is director of analytic solutions at ISO, a Verisk (Nasdaq:VRSK) business. Su Wash is a senior actuarial associate with ISO. To contact these contributors, please send email to ContactMe@Verisk.com.
The opinions expressed here are the authors’ own.
See also: