Featured Story

What Executives Need to Know About Predictive Analytics in 2020

Analytics is an issue of leadership. It needs champions who have overarching business strategies in mind — minimizing the prospect of multiple uncoordinated efforts and unshared insights across the enterprise.

Editor’s Note: The following article was first published by Carrier Management on January 8, 2020.

By this time, most leaders in the insurance industry have heard of using predictive analytics to study previous patterns that can help businesses understand what could happen in the future. However, the amount of noise being made about predictive analytics can be deafening. Ultimately, this can lead a company to jump into a flurry of activity that may or not be useful, or inactivity due to confusion about where to begin.

Predictive analytics is an effective tool that can lead to significant benefits when applied correctly, and we have identified six key themes that executives should focus on in 2020 to help ensure these benefits are achieved.

1. Analytics is mostly a leadership issue.

We often think of analytics as the purview of statisticians, actuaries or data scientists because that’s often who performs the work. We assert that it’s a leadership issue because executives define what questions to answer, what problems to solve and what opportunities to pursue. These questions need to relate to the company’s overall business strategy — thus beginning with the end in mind.

Additionally, executives should champion an overall analytics strategy versus the pursuit of multiple uncoordinated efforts. Consider how many business units or functional teams are relying on similar data to chase the same questions (What drives the cost of claims? What drives customer satisfaction?) or leveraging different technologies (data science platforms, chatbots) and not coordinating efforts or sharing insights across the enterprise.

2. Predictive analytics has little value if it’s not generating favorable business outcomes.

Predicting which policies are likely to lapse is very interesting and often generates a lot of discussion, but if that prediction is not being leveraged to refine existing processes, behaviors or decisions, it has not delivered much business value. It’s critical for executives to understand the intended application and success criteria of analytics initiatives and set expectations to measure return on investment after implementation.

Insurance companies often approach the insurtech space backwards. Insurers act as surveyors, identifying what solutions are out there and ultimately deciding to incorporate technologies that demonstrate (or promise) results that are impressive.


The allure of insurtech offerings, especially, can lead insurers to chase solutions that may not generate favorable business outcomes. Insurance companies often approach the insurtech space backwards. Insurers act as surveyors, identifying what solutions are out there and ultimately deciding to incorporate technologies that demonstrate (or promise) results that are impressive.

Unfortunately, an impressive application that is not related to a company’s mission will generally not result in favorable business outcomes. It would be better for companies to approach the insurtech landscape as miners, identifying the business needs of the company and searching for solutions that address the business need.

3. Ensure data strategies align with analytics applications.

Have you ever asked a question that your data could not answer, either because you did not collect the right information, you were not confident in the quality of the information or the information was not easily accessible? Not only will a well-defined data strategy help alleviate these issues but exposing the desired business outcomes to those involved in executing the data strategy will help underscore the importance.

Executives also play an important role in translating the company’s strategy to help data teams prioritize efforts. This means striking the right balance between continuous improvement efforts in core data assets and innovative efforts to create new data assets based on novel combinations of disparate data. For example, if you want to study how shortcomings in your digital self-service capabilities drive contacts to your service center, you have to build a data platform that can stitch together the customer’s experience across digital and service center transactions. If resources are limited, should more resources be deployed to undertake this or to pursue defects in core data assets (e.g., policy and claims data)? Which aligns more closely with strategic objectives?

4. Push beyond the hype to attain a working knowledge of machine learning and artificial intelligence.

We’re inundated with sound bites around machine learning and AI, and it’s easy for executives to proliferate the hype without a solid understanding of what these terms mean. This is exacerbated by the fact that the business world does not have consistent definitions. Executives may choose to become conversant through books or classes aimed at business leaders, but it’s equally important to push your own teams for business-relevant definitions, an understanding of advantages and shortcomings, as well as knowledge in what methods can and can’t do, likely applications, etc. In particular, leaders will be well-served to better understand the balance between complexity and understandability (e.g., the most complex model may not be well-suited to changing agent behaviors). It’s also prudent for leaders to ask about potential unintended consequences of a model.

5. Keep the “last mile” in mind.

What technology will be used to deploy the predictive model in operation?

Many legacy systems were set up to handle if/then/else rules or simple algorithms but not complex models. Modern systems can often call models via application program interfaces. What are the downstream and upstream implications of introducing a model into a broader process? What is the change management plan to ensure people who are part of the broader process have “bought in” and are working with the model and not fighting against it?

It is also important to remember that the “last mile” of the journey leads to the starting line of the next journey. This is for two reasons. First, no model is perfect, so part of the ongoing maintenance of the predictive model application will be to identify situations where the results are unusual or unexpected. Second, no model is implemented in a scientific laboratory. Once the results are implemented, customers, competitors and even company employees may adjust behaviors. This will result in an underlying shift to the process being modeled.

Consistent monitoring of results and incorporation of updated data in the modeling process will constitute the continuous future journey.

6. Promote and inspire multidimensional teams.

No matter how much science advances, it will not displace the need for business context. In fact, one might argue that as science advances to the point of training machines to learn on their own, the business context becomes even more important. Executives establish the business climate that enables diverse competencies (e.g., quants, domain experts, technologists) to row in the same direction.

If business executives keep these six things in mind in 2020, they can help improve the likelihood of predictive modeling applications driving improved business outcomes and achieving the desired strategy.

Roosevelt Mosley, FCAS, MAAA, CSPA, is a principal and consulting actuary with Pinnacle Actuarial Resources, Inc. and has 25 years of property-casualty actuarial experience. His skill set includes predictive analytics applications for all insurance functions, ratemaking and product development, competitive analysis and litigation support.

Claudine Modlin, FCAS, MAAA, is an insurance executive and actuary with more than 25 years’ experience in the industry. She is currently the head of personal lines product for the Western zone at Farmers Insurance Group.