Marketing Data Can Reduce AI Bias in the Insurance Industry

Just a decade ago, the idea that artificial intelligence (AI) might play a role in everyday insurance practices may have sounded far-fetched. As a legacy industry with longstanding ways of doing things, many insurers kept to traditional practices even as digitization grew up around it — by 2017, only 1.33% of insurance companies were investing in AI capabilities.

Though the winds of change have blown from many directions, two major events helped pave the way for AI in insurance: the insurtech boom and the pandemic.

The former brought much needed digital-disruption and fresh-faced competition to this legacy industry, while the latter catalyzed a global trend towards digital processes and remote functionality. The result is that AI plays an increasingly larger role in the insurance toolkit — for identifying prospective customers, improving the claims experience, determining premiums that more accurately reflect individual needs, and more.

But AI is only as good as the data that informs it, and pairing future-forward processes with outdated actuarial data can come with serious implications — first and foremost, magnifying biases that unintentionally disadvantage certain groups of people. Only by pursuing cutting edge, relevant marketing data and thoughtfully integrating it into AI processes can the insurance industry reduce the bias that currently exists in many of their algorithms — otherwise, they run the risk of getting bogged down by historical biases that do a disservice to themselves and their clients.

AI Bias in Insurance

AI does not generate bias on its own — only through flawed development or input does it result in skewed outputs. To address such AI bias, therefore, one must turn to the source: the collection of actuarial data and the way employees interact with it.

Actuarial data sets are compiled over years, so the biases therein become deeply rooted, making them difficult to spot in real time. Additionally, the humans who input this data can create yet another layer of bias through their selection or rejection of specific data sets. For individual agencies, carriers may use the data they are inputting on their clients, however, the potential for bias to creep in becomes much larger if there isn’t a clearly centralized method for collecting data.

This means different regions could present different biases from their data collection and input methods. AI algorithms input whatever data they are given, biases and all, thus amplifying these trends.

However, it is not enough for insurers just to “keep an eye out for bias” while collating data. The prevalence of bias in AI has become so ingrained that according to estimates, about 38% of all “facts” used by AI have bias within them. In insurance specifically, AI has perpetuated biases that generate discrimination based on class, race and gender. For example, minorities are much more likely to be denied conventional mortgages and in some cases, women wound up paying about $100 more on average than men for auto insurance.

A Marketing Data Makeover

For an industry like insurance, with its deep roots in commerce and vast stores of historical data, it is no easy task to mitigate and ultimately eliminate these biases. But though the problem persists, a promising solution is rising in its wake.

Though it may be more convenient to fall back on historical data, it will ultimately benefit insurers in the long-term to collect real-time, firsthand data from customers. This data should span everything from customer acquisition information to email correspondence with the company; from web behavior to purchase intent; transaction history, claims data, and more. This also extends to the data that is collected by individual insurance agencies, with the possibility to leverage regional-specific data and additional insights that they may have on their customers that major carriers may not keep track of.

One of the biggest advantages of using marketing side data is the ability to generate a large amount of data in a relatively short time. As claims data (and other deep performance metrics) may take years to compile, conversion and purchase data can be generated immediately.

Not only will the amount of up-to-date data quickly surpass traditional actuarial data, it also will yield insights more relevant to the needs of the insurance industry and customers alike. In doing so, insurers will eventually be able to phase out their old, biased data sets altogether.

Additionally, first-party data equips insurers with the tools and customer insights needed to generate targeted marketing campaigns in order to attract new policyholders, while also giving existing policyholders a far more personalized experience. The increased flexibility of this marketing data also allows insurers to provide algorithms that actively account for and correct AI biases, allowing for data parameters that ultimately reduce discrimination without compromising accuracy or relevance.

Insuring a Better World

The inherent bias in current AI algorithms can lead to real life consequences for groups historically affected by discriminating policies or pricing. By embracing improved marketing practices, leveraging real-time data collection, and ultimately phasing out outdated, bias-ridden data sets, the industry can create a more equal playing field for policyholders.

Everyone deserves an equal chance to feel safe and secure — it’s the job of the insurance industry to create a future where that is unequivocally true.