Skip to main content
Business transformation

Behind the patent: A method for curbing insurance fraud

Article 21-Jun-2023 Read time: min
By: Clea Zolotow

Predictive systems can be very helpful in many situations, including detecting errors and automating steps of repetitive tasks. Deploying predictive systems in complex environments, however, has proven exceedingly difficult. One recent patent, developed by my fellow engineers and I, addresses this problem.  Cognitive Systematic Review (CSR) for Smarter Cognitive Solutions” teaches techniques to determine the accuracy of certain events in even the most complex situations–such as risk mitigation and fraud.

What insurance fraud in complex environments may look like

Take insurance fraud, for example. Insurance fraud affects both insurers and their customers. In 2022 alone insurance fraud is estimated to have cost US insurers $308 billion.1

Let’s say I have a friend, who has a house on an island off the southeast coast of the US.

This island experiences seasonal, hurricane-force storms that can cause property damage. My friend’s house, however, is inland enough to be out of the flood zone–and was constructed more recently, under new building codes, providing a sturdier defense against extreme weather.

Despite the relative insulation, my friend’s property insurance premium recently spiked due to an increase in destructive weather events over the past few years. Rising premiums are, of course, sometimes inevitable. But what frustrated my friend was that her premium increased just as much as premiums for homes much nearer the ocean, as well as homes that aren’t up to the new building code.

Now: What my friend doesn’t know is that some number of her neighbors–also inland, also with recently constructed houses–have been making fraudulent claims (as detected by the insurance companies after significant manual review and investigation), hiking up the premiums for all houses in their vicinity.

How technology can help detect deceptive actions

Identifying deceptive actions can be a highly manual and time-consuming process, which can also be easily biased. Many professionals within the insurance industry don’t have the time, access, or resources to gather and process all the information available about a particular event, area, or claim. Or, specifically when it comes to property insurance, tend to group homes together by county, regardless of microclimates or variations in terrain. This translates to less rigorous and accurate predictive models, and ultimately an increased likelihood that red flags slip, unnoticed, beneath the surface.

Specifically, smarter technology could give businesses a tool that would supply their predictive models with a robust ecosystem of evidence-based predictor variables.

What my colleagues and I determined was that smarter technology could be deployed to improve the cognitive and predictive systems used by insurance companies, and other companies facing similar challenges. Specifically, smarter technology could give businesses a tool that would supply their predictive models with a robust ecosystem of evidence-based predictor variables. This in turn would help these teams improve their predictions, and therefore customer and client insight, over time.

The engines behind this technology are Hidden Markov Chains–one of my personal favorite methods for determining commonalities of behavior.

Put simply, Hidden Markov Chains are mathematical models that are used to make predictions about the likelihood of future events. What makes these models particularly useful here is that the patterns that make them run are based on past events that occur in sequence–some of which are hidden or unknown. What’s more: Hidden Markov Chains haven’t yet been used to detect many types of insurance fraud.

What makes Hidden Markov Chain models particularly useful is that the patterns they run on are based on past events that occur in sequence–some of which are hidden or unknown.

Why this predictive model can be a game changer

Let’s get back to the hypothetical island home example. After a tropical storm sweeps the coastline, my friend’s neighbors see a chance to take advantage of the situation. Though their property has not sustained any damage during the storm itself, over the years, their roof has fallen into disrepair. And they decide to file a claim for damages.

Now, while the storm did occur, and there was damage to other properties, the damage was concentrated within the flood zone area. Using our new approach, an insurance company could feed their model with a broad sweep of publicly available data–from news reports and studies to controlled trials and even social media.

For instance, the company might evaluate the claimant’s exact location on the island, with detailed data from open sources such as Google Maps; weather data, on where the worst wind gusts and rain fall were concentrated to; and risk profile, through government records about build years, building codes, and recent storm damages to other nearby structures. So that when an adjustor is sent out to evaluate the claim, for example, they will be better informed about the likelihood of the roofing damage in question having occurred during the storm itself.

Using this method, this company could then more accurately evaluate the veracity and likelihood of the neighbor’s claim–curbing the fraud itself and mitigating risk, but also potentially protecting nearby properties, like my friend’s, from increased costs.

Opening the door to more rigorous and accurate predictive models

If this method were to be realized as a product or service, it holds the potential to streamline and optimize the predictive systems of insurers across coverage areas. This could save these businesses the financial investment required for the current manual and time-intensive systems in place. Meanwhile, it could also lead to more rigorous, defensible, and accurate predictive models over time–and therefore, better performance and business outcomes.

But the value of this new method truly lies in its versatility.

My colleagues and I originally designed the approach for fraud use cases. But more generally it also shows the potential to assess risk and future life events, resulting in more accurate, personalized premiums for customers. This more precise decision making eventually may also help insurers in their case management, training, and marketing processes.

And these are just a few of the possibilities created by the magic of probability.


Clea Zolotow is Vice President, Enterprise Architecture for Kyndryl's office of the CTO.

  1. Facts + Statistics: Fraud, Insurance Information Institute, Inc.