Despite the attention the industry pays to predictive mortality models, many still lack a clear understanding of how best to take advantage of them in the life underwriting process. In this post, we dig into the advantages and disadvantages of rules engines and predictive models and how to effectively use them to expedite the underwriting process and improve the customer experience (CX).
For life insurance underwriters who assess mortality risk, the process used to be fairly consistent across the industry. Underwriters would gather as much information as they could and decide whether an applicant was worth the risk. Guidelines were developed in consultation with medical experts and actuaries and they were codified in life underwriting manuals.
To automate underwriting, carriers had to get information from paper to digital formats. Early automation replaced simple manual tasks like ordering evidence and routing cases to the most appropriate underwriter. As more and more applications were digitized, workflow rules evolved and rules engines were needed to manage the increasingly complex environment.
Workflow rules evolved in complexity and rules engines were needed to manage the increasingly complex environment.
Today’s rules engines are essential to automating, managing and tracking underwriting programs. They undeniably make underwriters’ jobs easier and provide more consistent risk assessment. And since the rules typically carry out logic that underwriters have designed, rules and rules engines are well-understood and trusted tools.
With huge amounts of data being created, data scientists were hired to derive the maximum value. They went to work using advanced techniques like machine learning and artificial intelligence to build models in hopes of giving companies an edge in the underwriting process. In some cases, when they had the right data in sufficient volume, they succeeded.
Rules engines have been around for a long time because they have lots of upsides. Their biggest benefit is their simplicity; they’re easy to explain and easy to understand: “if this, then that.” They’re also consistent in terms of how they process applicants. Each carrier creates its own rules and fine-tunes them to suit its unique underwriting, pricing, CX and operational goals.
The industry used rules to take on complicated challenges like improving the CX, shrinking the uninsured gap and improving the speed and quality of automated underwriting.
The biggest challenge with a set of rules is complexity. At its simplest, a rule may determine that if an applicant has a BMI of 36, then the applicant may receive 50 debits. The number of debits may be modified by credits based on other factors, such as age, sex, resting heart rate, blood pressure or the presence of certain medical conditions. There may be a rule for every combination of these factors.
Suddenly, there's a massive collection of extremely complex rules.
Suddenly, there’s a massive collection of extremely complex rules crafted by numerous new data sources such as prescription history, electronic health records, etc. Managing them can be tedious, time-consuming and costly. Plus, there’s a practical limit to the complexity that can be managed before redundancies or gaps creep in.
At some point, the cost of adding rule complexity outweighs the cost of the underwriter’s time and the impact of a delayed decision. At a minimum, rules excel at identifying the less-complicated cases and passing the rest to an underwriter for a more thorough evaluation.
Predictive models can handle complexity. In building predictive models, modern techniques are able to evaluate dozens of variables simultaneously and tease out the interactions between those variables. In addition to replicating an underwriter’s judgment and applying guidelines, predictive models have the potential to improve risk analysis because they can identify patterns in the data that are simply not possible for the life underwriter.
Predictive models can greatly reduce the number of rules and help keep more cases in an automated process while still managing risk.
Like rules engines, predictive models can process data fast, which means a better CX for applicants. But when rules meet their practical limits and punt to a person, predictive models can take over. In fact, predictive models can greatly reduce the number of rules that organizations need to manage. They can help keep more cases in an automated process while still managing risk. (For those who still need convincing, read If the Proof Is in the Pudding, Grab a Spoon.)
Predictive models can solve for a number of use cases. Our models estimate mortality risk based on dozens or hundreds of known data points. We’ve also created models that can narrowly focus and assess the risk of a specific medical condition based on prescription history or medical claims codes.
It’s easy to see how the advantages of predictive models start adding up. Though they require an initial investment to build, many options are available on a cost-effective pay-as-you-use basis.
Models need to be used where they have a real advantage over rules or human judgment. And we still need to overcome some obstacles to their widespread adoption.
One important drawback is that predictive models require structured data. Any unstructured data, such as physicians’ notes, etc., must be converted or structured so predictive models can use it. And that can be a chore.
Also, the often-opaque nature of predictive models is another drawback to their widespread adoption. Explainability is the reason we created the Score Report, to help underwriters understand exactly what drives the overall risk assessment and model results.
Another drawback is trust or the lack thereof. Humans tend to trust other humans more than they trust computers. (Thank you, Stanley Kubrick.) And underwriters and actuaries, being human, tend to trust solutions created by other humans.
Let us be clear: Rules engines and the platforms that help us manage them are here to stay. However, certain rules or sets of rules could be replaced and/or enhanced by models, thus simplifying the process, reducing costs, expediting underwriting and improving the CX. These benefits alone are enough for you to look at where predictive models can fit into your underwriting process and create a plan to get there.
When you identify the proper mix of rules and models, you can derive maximum value from your underwriting efforts. Contact us if you’d like to learn more.
MM202602-304488