Using market segmentation in life insurance to benefit both insurer and customer TALEX DIEDE

Market segmentation is by no means a new topic, or even a particularly innovative one. Rather, segmentation has been considered a key marketing concept and focus of marketing research since the early 1960s.1 It is readily apparent that markets are heterogeneous, making it nearly impossible to develop a single product that will have mass appeal across all consumers. Even if such a product is created, or a product is able to be widely marketed, it is likely that different consumers will have different motives or uses for the product in question.

This diversity of consumers has motivated marketers of products to strive to identify segments, or profiles, to help them either build a more tailored product or market an existing product to more targeted subsets of consumers. This targeted marketing and product development is starting to be applied in the life insurance industry, where, to date, there are relatively few products that all tend to be widely sold using the same marketing techniques. By adapting other industries’ uses of segmentation, we can develop better targeted products and/or better identify the customer base for which a particular product will be most valuable.

Market segmentation has the potential to benefit both the insurer and the insured. Insurers can increase sales efficiency by targeting subsets of the population for marketing and distribution of particular products, and consumers can get products that are more tailored to fit their particular needs. By offering tailored products that satisfy the needs of the customer, insurers can look forward to increased customer satisfaction and retention. Better understanding the needs of the insurance consumer can be a win-win prospect, and segmentation can help get us there.

Segmentation at Work

Before delving into segmentation in the life insurance industry, we can take hints from uses of segmentation that have been pervasive across many industries. In the 1920s, General Motors overtook Ford in vehicle sales by producing “a car for every purse and purpose”—a famous example of a market segmentation strategy. More recently, the credit card industry has segmented customers based on their past credit behaviors to send targeted solicitation messages, products and benefits.

Moving closer to the insurance industry, there are examples of segmentation already at work. For example, auto insurance companies have evolved to serve niche markets—for example, USAA serves the military and their families, and Progressive serves high-risk drivers. By serving these specific markets, a company can increase sales efficiency and customer satisfaction as it markets to its selected segment, and it can provide offerings specific to those customers’ needs.

Better understanding of customers’ motives can help insurers design products that are more targeted and better suited to specific customer needs.

The life insurance industry has also already seen a version of segmentation. Clustering previously has been brought to the forefront of actuarial modeling, specifically being used for inforce compression. This idea is, at its core, the same as market segmentation. The clustering procedure takes a group—inforce policies in this case—and uses characteristics about the policies to create groups of like policies. A single record that characterizes the policies included represents each segment. This allows for projections to be run on a smaller number of records without loss of accuracy. The purpose of this clustering is to reduce run time for time-consuming projections, but we can take similar algorithmic approaches on the customer-facing side of the business.

Bringing Segmentation to the Life Insurance Industry

As the life insurance industry accumulates more data on its policies and policyholders, we, too, can better understand and target our customers. The existing experience data collected by the insurance company is used regularly to set assumptions related to policyholder behavior. And, more and more, the life industry’s growing mass of data is being used to set more precise assumptions for those policyholder behaviors by employing predictive modeling techniques.

Predictive modeling with policy-related characteristics is the first step in advancing assumption-setting to better differentiate policyholder behavior, but it is still relatively blind to the person behind the policy. This is where big data comes into play. Through the acquisition of additional data on policyholders available from third-party data vendors, we can move beyond predicting behaviors and begin to understand the motives behind them. Better understanding of the customer in this way can help insurers design products that are more targeted and better suited to specific customer needs. For example, customers with immediate liquidity needs are more likely to place value on the ability to get money out now than they are on a product feature that may provide them with more money down the road.

The first step in this process, as with any predictive analytics project, is to gather data. Insurers can start by gathering internal data such as policy values, policy experience, policyholder demographics and distribution data. The next step is to acquire as much additional data as possible to enrich the internal data, ideally information that will give a sense of a policyholder’s overall financial situation. This may include occupational, consumer marketing, mortgage, credit data2 and other types of information. Major types of segmentation variables include:

  • Geographic (population density, climate)
  • Demographic (age, family size, life stage, gender, income, education)
  • Psychographic (lifestyle, personality)
  • Behavioral (purchases, transactions, customer tenure)

This data can be used to create policyholder segments using any number of clustering algorithms.

The policyholder segments we create identify policyholders who are likely to behave in similar ways due to shared circumstances and motivations. To be useful, the segments must be:

  • Measurable. The size, purchasing power and characteristics of the segments can be measured.
  • Substantial. The segments are large and profitable enough to serve.
  • Accessible. The segments can be effectively reached and served.
  • Differentiable. The segments are conceptually distinguishable and behave differently.
  • Actionable. The segments can be attracting/served effectively.3

For example, a segment with low credit scores and/or high loan-to-value ratios on their mortgages implies liquidity needs that are likely to influence the decisions they make regarding their policies. Predictive models can then be fit to each policyholder segment to predict the behaviors of interest. We expect to see differences among the models for each segment, which will give us insight into how each segment makes decisions.

With an enhanced understanding of how each segment makes decisions, we can understand a lot more about these customers and their needs, and thus form better assumptions about their future behavior. Taking this a step further, these predictive model assumptions can be used in cash flow projections to determine the profitability of each segment. The calculated profitability can be used to determine segments to which the product should be marketed in order to improve bottom-line profitability.

Beyond enhancing the profitability impact that can be made, the enhanced behavior understandings can be used to develop new products that can better target the needs of a given segment. For example, if the analysis identifies a segment of policyholders who are exhibiting what would be considered inefficient behavior, this information can be used to target them with a new or different product that would better fit their perceived needs.

As data and technology continue to evolve, we should continue to look forward, beyond where we are now, to where we are likely to move down the road.

Most life insurance companies historically have faced a key problem in the inability to distinguish policyholders who are likely to behave quite differently from one another. This has led to overall inefficiencies and challenges in the marketing and development of new products. Market segmentation has been used in many other industries already to help alleviate those inefficiencies. Segmentation can be used simultaneously to improve company profitability as well as provide better value to customers based on their unique needs.

We see segmentation as the next step for product development and marketing in life insurance, but it’s worth noting that this is unlikely to be the ultimate state. As data and technology continue to evolve, we should continue to look forward, beyond where we are now, to where we are likely to move down the road.

The Future Beyond Segmentation

The ultimate level of segmentation leads to segments of one—individual customized products and offers. Products such as “Design your own Converse” sneakers currently incorporate “mass customization” of product features, but “segments of one” are seen more commonly today in marketing. In the era of big data, we have already started to see many companies make moves in data analytics that extend beyond market segmentation. They are no longer targeting markets solely based on groups of people, but rather they are targeting people as individuals. For example, Amazon uses individual customer purchase history to recommend products for future purchase. Looking at another recognized name in data analytics, Netflix performs similar analyses at the individual level to recommend movies or shows users may enjoy based on past viewing history and ratings.

Another, perhaps infamous, example comes from Target, which saw a backlash over coupons for baby and maternity items that were sent to a teenage girl. The girl’s father was furious, but later returned to apologize after discovering that his high school daughter was indeed pregnant. Target had analyzed other customers’ buying behaviors and was able to determine patterns of purchase behavior, at the individual level, that could identify customers who were likely pregnant. Target’s algorithm matched the expected purchase pattern to the teenage girl in question, and she was sent the coupons that were aimed at a pregnant audience.4

The ability to move beyond segmentation and into individual marketing and recommendations is heavily dependent on data size. Companies such as Amazon, Netflix and Target have frequent customer transactions and a wealth of data already collected; the life insurance industry is slower to generate as much data internally, but it can supplement its own data with external sources to move down the same path.

Using market segmentation, the life insurance industry can take advantage of available internal data and external data sources to better serve customers’ needs. As more data becomes available, both internally and externally, and as we as an industry continue to become even more sophisticated using that data, we will follow in the footsteps of other industries and move toward serving the segments of one.

Talex Diede is an actuarial analyst and data scientist at Milliman in the Seattle Life Practice. She has her M.S. from the University of Washington in computational finance and risk management. Her current work focuses on using data and predictive modeling techniques to drive business decisions in the life insurance industry.