As in so many other industries, disruption is affecting insurance, and machine learning is both the cause of and the cure for that disruption.
Industry leaders have seen success result from pairing insurance and machine learning. As these leaders become more adept at extracting value from unstructured and real-time data, they have realized a competitive advantage and profitable growth. Their success exerts pressure on their competitors, who face a choice: lose business or figure out how to use machine learning to their own advantage.
Let’s consider a real-world example. An insurer noticed its subrogation numbers were not aligning with its historical data. The difference was not huge, but it was noticeable. The firm decided to apply text analytics to the unstructured data in its claims notes and diaries to understand why the discrepancy was occurring. As a result, it found $1 million in unidentified subrogation.
While the initial business problem was small, the results were substantial and captured the attention of the company’s president. The $1 million discovered in that first project helped fund their next round of use cases, to which they are applying advanced analytics and machine learning. The key is to understand that any business problem you face may be used as a business lever. Even solving a small problem can lead to big gains.
The value of the partnership between insurance and machine learning is the untapped richness contained in new data sets. Telematics, social media, geolocation, emails, texts, sensors, video, photos—these and so many other unstructured data sources provide a wealth of information that can add accuracy and clarity to existing structured sources.
Far too many insurance firms find themselves unable to penetrate this data using their current structured data systems and architecture. That leaves data stuck inside the source and unable to deliver value. Machine learning allows insurers to work more efficiently and finally tap the unstructured data sitting in claims notes, diaries, or other real-time sources.
Claims are a top priority for insurers, because that is where money is being spent. Insurers should view unstructured data as a way to create a more complete picture surrounding an incident claim.
Consider an auto claim that also results in property damage and medical payments. A machine learning model could take advantage of all data sources to help with the specific claim, and also to strengthen predictive modeling for future claims. Through machine learning, the carrier could access a host of sources related to the accident scene, including weather conditions, dash cam video, photos provided by witnesses, traffic information gathered by the city’s smart transportation system, vehicle sensor data from the automobiles involved in the accident itself, and even surrounding vehicles.
With this information, the carrier could be alerted early to possible fraud, or be able to more quickly validate the claim and settle faster. After the claim has been evaluated, machine learning could aid in claims payment optimization to reduce the chance of claims leakage by the carrier. In turn, all this data could be fed into the carrier’s predictive model to identify the types of claims that leak versus those that don’t.
Insurers face pressure to better identify good and bad risks. The information mined during the claims process could refine the carrier’s risk appetite and underwriting analysis, leading to better risk selection and pricing, as well as identify positive and negative trends that can be shared with underwriters to modify behaviors that improve business results.
Carriers that embrace machine learning have much to gain. As they become more skilled at using machine learning, they can drive down claims costs, bring down premiums for less risky users, and make smarter underwriting decisions. Companies can more effectively apply precision pricing by more accurately evaluating pricing risks within their risk appetite, and customers gain a better understanding of what their policies cover.
If you face the question of how to marry insurance and machine learning, where do you begin? First, identify a use case that will make an impact. As in the subrogation example above, you don’t need to start with the easiest task. Instead, start with one that will make a noticeable difference. Understand the business pain you want to solve and don’t be afraid to pursue that goal—even if it proves to be more complex. When you identify the use case, work backward to find the data sources that can illuminate the problem and fill in your knowledge gaps. Learn from each project and build on each success. The more you and your team learns, the more expertise you’ll have to apply to bigger business problems.
Find out more about how the insurance industry can use connected data platforms to leverage and analyze data in real time.