Robert Clark is the founder and CEO of Cloverleaf Analytics, a leading provider of insurance intelligence solutions.
“Breaking news: This just in—another national insurer has been fined significantly after an independent industry audit of their AI models. The audit revealed that the third-party data the insurer purchased to build their world-renowned AI-powered insurance product was laden with tremendous levels of inaccuracy and bias impacting underwriting and claims.”
This example situation is unfortunately a foregone conclusion as 2024 is projected to see major increases in AI technology adoption by insurers that have a long-held practice of buying subpar data to run their business. With consumers and insurers salivating about the potential for AI, what pitfalls should insurers watch out for to ensure they procure quality data?
Potential Issues With Insurers Purchasing Third-Party Data
Insurers purchase data to build their products from a variety of sources, including credit and motor vehicle reports, geospatial data, social media monitoring data, information from smart home/connected car/IoT devices and other sources that can lead to inaccurate and biased data.
This third-party data is merged with existing customer data from disjointed internal insurer systems to create a subpar foundation for the products and services that consumers of all socioeconomic backgrounds purchase.
To avoid acquiring subpar data, insurers need to evaluate the data sources, analytics processes and data security controls of the third-party data vendors to ensure quality information. As part of this process, insurers should ask data vendors about their commitment to constantly monitor their information sources for errors or bias.
By not building a strong relationship with their data providers, insurers will inevitably run into consumer lawsuits and regulatory fines as bias becomes more prevalent while insurers race to show off their new AI-powered businesses.
A Powerful Yet Simple Question Insurers Should Ask
One simple question that insurers should ask themselves when purchasing data is: “Would I be happy to purchase insurance from an insurer that bought this data having known that the data source could have some issues?”
Even better, insurers should put this question to the data vendor as part of their evaluation process before purchasing the information.
The Insurance Data Butterfly Effect
The insurance industry is on the cusp of creating a real-world case study of how data purchases can have a butterfly effect across the entire insurance distribution lifecycle and other areas of consumer life. The butterfly effect suggests that a minor action can have a significant impact on a larger, more complex system.
Intelligent sensors, a growing and more diverse population, and the expanded use of AI have the potential to reshape the impact of one small occurrence of biased or flawed data on a scale like never before. Imagine if one of the third-party data providers that insurers rely on had an undiscovered data breach in which flawed information was stealthily inserted into their database.
This data is then purchased by a major insurance company that adjusts its AI models, underwriting and claims processes, resulting in a bias against a particular customer segment. Now let us say that the segment had an issue with managing a reasonable standard of life due to rising costs of living before this change.
The minor undiscovered change in the data lake of the third-party provider could affect auto insurance pricing so that a customer would pay $100 a month more for insurance due to perceived risks to the insurer based on this data error. That extra $100 could lead the driver to sacrifice other critical areas of their life like medicine because they need to drive to work, and their budget is already stretched. Alternatively, it could prompt the driver to operate an uninsured vehicle so they don’t miss work, and then get in an at-fault car accident, delivering an even bigger financial blow.
Think these types of attacks do not happen? Think again; they are called data injection attacks, which are when a hacker inserts flawed data to disrupt business operations to cause monetary loss for an organization or influence an organization’s decision-making. These attacks can also be used to gain unauthorized access to sensitive information in databases or to insert malicious code into databases for future, more devastating, attacks.
Another type of attack recently impacting GenAI is when artists use data poisoning tools like Nightshade to insert small changes to data sets to alter the image generation results in an attempt to protect their original artwork.
Encouraging A New Era Of Data Procurement
As this interesting recent blog post from analyst firm Celent discusses, insurers are entering a new paradigm of data modernization as they revamp internal core systems with emerging technologies.
The most concerning element of the race to AI in insurance and other industries is that companies and consumers alike are focused on the shiny new GenAI models, autonomous vehicles or other emerging technologies powered by AI without thinking about the foundation of these new products.
Insurers must do more than follow once-trusted data-purchasing practices if they want to maintain customer confidence. AI is exciting, but not at the cost of running blindly to implement innovations without knowing what type of data is powering the insurtech under the hood.
As I have said before, insurance is one of the oldest industries and has a long history of having treasures of customer data. The insurance industry should set the standard for quality data and data vendor monitoring best practices that power responsible, innovative AI products and services.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
