WHITE PAPER

A Simple Explanation of How AI Can Be Used in Insurance

AI in Insurance: Practice Makes Perfect

As the saying goes, “practice makes perfect.” In other words, people learn from experience. According to the Oxford Dictionary, “regular exercise of an activity or skill is the way to become proficient in it, especially when encouraging someone to become persistent in it.” While practice cannot truly make someone perfect in any industry or activity, persistence does usually lead to increased proficiency, and seasoned practitioners typically outperform those who are just starting out.

In the insurance world, this concept applies as well, for both underwriting and claims management. A new underwriter gets better at predicting the cost of a new policy by seeing more policy applications, evaluating them, pricing them, and then seeing whether the price quoted results in a favorable or unfavorable loss ratio at the end of the policy term. Similarly, a new claims manager gets better at identifying and managing high-cost claims by seeing more new claims, triaging them, managing them, and then seeing whether the ultimate incurred is more or less than the amount of the reserve set.

Data science leverages this concept of repetitive practice leading to improved performance in the form of Artificial Intelligence or “AI.” At a high level, AI replicates the kind of human learning described above, but an AI model can “learn” from far more policies and claims than any person could ever hope to.

Where an experienced underwriter might see tens of thousands of applications in a career, an AI model can consider tens of millions of expired policies and the closed claims made against those policies. Further, the AI model can “remember” every fact in every one of those policies and claims and can compare those facts to every fact in every other policy and claim it’s “seen.”

This training process provides the model with the “knowledge” it needs to assess new underwriting applications and claims going forward, providing the organization with powerful capabilities to enhance its traditional operating practices.

Complicated Science; Understandable Principles

Understandably, before an insurance organization can adopt AI – or any technology – it must be understood before it can be implemented, and its value fully realized. While the science behind AI itself is complicated, the principles are not and can be understood easily with a few explanations and illustrations.

Just like a person learning to predict what’s going to happen when presented with a new set of facts, an AI model looks at an application or claim and takes stock of what characteristics make it similar to, and different from, all the policies or claims it’s seen in past.

Unlike a person, however, whose experience may be limited or biased, the model then evaluates those similarities and differences using a purely mathematical algorithm to predict what’s going to happen and determines the probability of a similar outcome.

Measuring the Accuracy of AI Models

To assess the model’s ability to predict the outcomes of policies or claims that it hasn’t seen in training, it first needs to be tested. This is typically done by showing it the facts presented in expired policies or closed claims without showing it the outcomes. That is, without showing it the claims actually made against those policies or the actual costs of those claims. The model is then asked to predict outcomes, and those predictions are measured against the actual outcomes.