There are probably very few of us who will seriously change our lives based on the predictions of a mysteriously veiled woman with big earrings who consults her crystal ball to determine the future. While such predictions may be entertaining or frightening, we tend to want something a bit more scientific on which to base important decisions.
Obviously, insurance is a business that is seriously concerned with predicting the future, and with making bets (e.g., selling coverage) that will in most cases be resolved to our advantage. In this age of computers and advanced statistics, the tool the industry uses to make its forecasts is known as predictive analytics.
As defined by AICPCU’s Predictive Analytics White Paper, predictive analytics is “a broad term describing a variety of statistical and analytical techniques used to develop models that predict future events or behaviors.” Most of these models generate a score (like a credit score), with a higher score indicating a higher likelihood of an event or behavior occurring.
Insurers use data mining to identify meaningful trends, patterns or relationships among the data. This information is then used to develop the predictive model that generates the all-important score. The white paper noted that the industry is using increasingly sophisticated statistical methods, including multivariate analysis techniques such as advanced regression or time-series models. If you don’t understand these terms, don’t worry; I actually had advanced statistics courses in graduate school and I can barely remember what these things mean. In the end, the paper stated, such techniques “enable organizations to determine trends and relationships that may not be readily apparent, but still enable them to better predict future events and behaviors.”
Such analyses are certainly important in developing and pricing products for potential customers, but the increasingly automated fashion in which they are applied can actually create problems on a human level. Let’s say that we have a product to sell, and that we determine that any potential customer who scores below 50 (out of 100) on our scale doesn’t get the coverage — or has to pay significantly more for it. That’s fine — it’s how insurance works, but let’s look a little closer.
It is certainly unlikely that we will want to assume the risk of those who score between 0 and 20 on our scale, but what about those who come in closer to the cutoff — say in the 40 to 49 range?
This group could include individuals who, perhaps because of a recent job loss or credit problem, may not make our grade today, but might potentially do so two or three years from now.
Unfortunately, our automated cutoff cannot consider such possibilities, and this could mean losing these customers down the road when and if they become more desirable from our scoring point of view.
The question is this: Can we accurately predict the point at which the rejected customer swears he or she will never buy coverage from us? And — in this extremely competitive insurance market — can we afford to totally ignore a potential group of customers for the future?
Predictive analytics is a fabulous technology for understanding and making decisions about potential customers today, but even with the help of a crystal ball, it has no power beyond that.
Consider the dynamics of human life, and don’t stop marketing to those whose scores are near misses.