While trying to stay warm during this cold winter break, my family decided to bundle up on the couch and watch some movies. One of the movies we picked was Captain America: The Winter Soldier from 2014. Despite the name, the movie didn’t have much “winter” in it, but it did get me thinking.
Be aware there are some spoilers ahead. In the movie, the antagonists are trying to achieve world peace by taking out potential threats before they ever happen. They built an algorithm that mined all the available big data and came up with 20 million people who they expected to threaten their organization in the future, including “a TV anchor in Cairo, the Undersecretary of Defense, a high school valedictorian in Iowa City.…” The top secret Project Insight was to partner the algorithm with some big gun ships that would take out all the potential threats at once. I’m not sure about the technology for the big gun ships, but we do have the technology for the algorithm. Using predictive modeling to weed out threats is starting to feel rather realistic.
Unfortunately for us numbers geeks, the movie didn’t get into the details of the predictive model. Did they use R, Python, SAS, or something new? How were they able to validate their data? Was the model supervised or unsupervised? Nevertheless, it did get me thinking about how all that data could be brought together. Can we use data to protect our group insurance organizations?
The group life and disability insurance world has been slower to adopt predictive analytics than other branches of insurance. One of the main reasons we are behind is that we often have limited information on who we are insuring. However, there are still many ways to incorporate predictive modeling technology to improve our results.
Pricing models can be improved by using predictive models to look at interactions between variables that we are already using in our pricing. For example, in disability insurance pricing, we know that higher-paid individuals tend to have better claim termination rates and so do medical Standard Industrial Classification (SIC) codes. However, did you know that the highest-paid doctors tend to have worse termination rates? A traditional rate manual review would not be able to pick up that interaction, but predictive models can.
Pricing at the group level can also be improved by using de-identified prescription drug history of the group. This information can be fed through a predictive model that predicts expected claim incidence for either disability or life insurance products, depending on how the model is calibrated. This extra information can be incorporated into pricing to help reward those who are good risks or, just as importantly, to appropriately price cases that are poor risks.
When determining a renewal pricing strategy, a predictive model could be built to determine which groups are more likely to accept a rate increase. This information could then be compared with experience results to come up with an optimal strategy for focused use of resources in the field.
Predictive models have also been used in claim operations. When reviewing disability claims, a model can help determine which claimants should receive the highest levels of review. These assessments could be based on the likelihood of fraud or the type of disability that would most benefit from early intervention. This is another way to optimize staff time.
The sales team could use a predictive model to decide where the best cross-selling opportunities are, based on likelihood to buy. Alternatively, using the experience of other lines of business, a model could determine if a given group would be expected to have better-than-average results (potentially with rating adjustments to support the increased sales efforts). For example, for a company that sells both disability and health insurance, the health insurance results could be a predictor of the potential results of a disability policy.
Start your own “Project Insight” by using predictive models to allow your company to proactively identify profitable business strategies. They may involve strategic pricing decisions, optimally deploying limited resources, or even small improvements over prior strategies that can yield significant profits. A word of caution, though. It is important for your modeling team to combine modeling capability with subject matter expertise. If a data scientist doesn’t understand the business, there is a possibility, for example, that the models could find relationships in the data that are just anomalies and not expected to continue in the future. A knowledgeable insider would anticipate these issues and make adjustments to reflect them.