Florence illustrates continuing data gap at NFIP and private insurers

Register now

Now that Hurricane Florence has pummeled the southeastern coastline with dozens of inches of rain, lashing winds, and flooded rivers, homeowners and insurers are scrambling to survey the damage to properties. Today, technology companies and insurers are working hard to enable a future in which insurers and their customers can not only better withstand events like hurricanes, but better understand the risks to their property before a weather event like Florence. This is a future wherein insurers and first responders leverage machine learning, imagery, and data to reactively address crises and rebuild more quickly, and also to proactively position homeowners in advance of a catastrophe.

Some of it is already happening.

Hurricane Florence broke all-time rain records in both North and South Carolina and many flooded homes are unlikely to have had flood insurance. 335,000 homeowners have flood insurance across the Carolinas. However, there are 1.8 million homes in coastal counties alone and it is estimated that only 10 to 20 percent of those homes are covered by flood insurance.
In the example of Hurricane Harvey, which occurred in 2017, this coverage gap was on full display. Federal flood insurance based on sorely outdated flood maps exacerbates this situation – more than half of the flooding in Houston was outside of all defined flood zones. Insurance is about helping families and communities manage risk so they can get back on their feet faster, but home insurance policies with complicated policy exclusions make it hard for consumers to know what coverage they are buying.

Kin and Lemonade are examples of new companies using machine learning and user-oriented design to make it easier for homeowners to understand, choose, and receive the right insurance policy. Many of today’s leading insurance carriers also have their own efforts underway. Their focus is on making complex policies and potential risks easy to digest so homeowners can understand their true financial exposure.

Newly available data, together with machine learning, can help us better refine our risk prediction. The National Flood Insurance Program should be overhauled, and private options that leverage the best technology should be created to give homeowners greater information about their risk, and wider choice in coverage. In Florida, the major risk from a storm like Irma comes from wind rather than flood, and structures can be built or retrofitted to lower the chance of catastrophic damage.

Technologists can now leverage geospatial imagery and machine learning to automatically detect these critical risk factors based on location, and track changes in these factors over time. The below geospatial imagery was captured after Hurricane Harvey, by Nearmap. In these images from Houston, you can see the house on the upper left was severely damaged by wind, whereas the house on the lower right suffered minimal roof damage and remained largely intact. In this case, roof shape (hip versus gable) was likely a key factor.
Roof shape, covering material, and “large missiles” like trees are broadly accepted as major risk factors, and we can now track these attributes at scale. The power of machine learning, together with large-scale imaging capabilities, and massive computing capabilities has opened the door to analyzing much higher quantities of near real-time imagery.

In addition, easy mitigation steps are often missed. Beyond having the right coverage, many homeowners can take simple steps to mitigate risk. But homeowners are busy and have lots to worry about, and it’s up to the carriers to dispense proper advice and policies in ways that encourage these behaviors. By leveraging emerging data and analytics, insurers can get much smarter about providing helpful, targeted, timely advice, rather than long policy documents that no one can remember. For example, insurers might observe risks that change the life of a policy and incentivize risk-mitigating actions, such as:

· Identifying when trees on the parcel have grown to the stage of presenting a risk, and recommend tree trimming services to the homeowner (perhaps with a discount on the service, or insurance policy)

· Identifying when a roof has neared end of useful life and offering a financial incentive to replace the roof (as well as guidance on what roof material is appropriate for the location, and perhaps recommended local contractors)

These are not minor improvements -- they can be the difference between a house surviving a hurricane or being completely destroyed. Prior to future disasters, we’d like to see policyholders proactively targeted with specific, timely advice based on such data.

Currently, it takes way too long to assess claims, and initiate payouts following a catastrophe. There are few material crises worse than someone’s home being destroyed. Waiting six weeks, twelve weeks, or more for financial reparations is terribly stressful – insurance carriers are striving to do better, and tech companies can help them.

Nearmap, for example, captures high-resolution as soon as an event has passed. Kespry is one of several tech companies who deploys drones and captures detailed imagery, which insurers like Farmers can then use to pay out claims in a fraction of the time. Doing any of this at scale, across a broad area, requires machine learning.

Going forward, the tech community has a major opportunity to help build this better future. Homeowners can be presented the correct, personalized coverage, in simple language. Insurers can know more about structures, and how they’ve changed over time, in order to better assess and underwrite risk. Policyholders can receive relevant, timely advice and aid to reduce the risk of catastrophic damage. Finally, post-disaster imagery and change detection will allow insurers to pay claims before they’re even submitted.

This is not just technology for technology’s sake – these innovations can have a measurable impact on decreasing human suffering after events such as Florence and Harvey.

For reprint and licensing requests for this article, click here.
Analytics Geospatial data Claims