Predictive Analytics and the New World of Retail Healthcare

Major transformations are under way in health insurance. Employers, hard pressed to keep up with increasing costs, are moving toward consumer directed and defined contribution benefit designs. As a result, consumers–never the silent recipients of healthcare advice that many of us have assumed–are more actively involved in their healthcare decisions than ever before.

Major transformations are under way in health insurance. Employers, hard pressed to keep up with increasing costs, are moving toward consumer directed and defined contribution benefit designs. As a result, consumers–never the silent recipients of healthcare advice that many of us have assumed–are more actively involved in their healthcare decisions than ever before.

At the same time, health insurers are undergoing a major consolidation. The top 15 for-profit insurers have consolidated into nine national and mega-regional players over the past two years, and further consolidation is expected.

The recent M&A cycle has allowed national players to capitalize on scale. Factors driving their market capitalization and successful track records over recent years include:

• Size and pricing discipline;

• Superior negotiating power and contract discipline;

• Effectiveness of consolidation efforts;

• The positive spread between premium and medical cost growth;

• Their ability to control current and forecasted administrative expenses.

However, the commercial and group health insurance market in the United States is not growing. It is, in fact, shrinking. As this cycle of economic expansion and consolidation begins to come to a close, the leaders will all have scaled and they will have squeezed costs out of the existing process model. How will insurers then deliver increasing returns to their investors in a stagnant market where employers are increasingly aggressive from a cost perspective? The growing demands of an increasingly vocal and demanding consumer base pose an additional challenge. The industry is poised for a paradigm shift, replete with opportunities for those who morph their strategies to fit the current times.

But so far, the industry has not found the right combination of product and price, at least not on a large scale. The currently uninsured present an opportunity for top line growth. In part because the individual market in most states affords more flexibility than the small group market, it is reasonable to assume that the uninsured will most likely enter the insurance market via individual products. One way to deliver increasing returns is to grow the top line with profitable individual health insurance sold to the currently uninsured.

The other way to deliver returns is both a top line and a cost strategy. Historically, health insurers have been satisfied by driving margins at some aggregate level, such as a particular large group, a state or a product line. The smart players in the brutally competitive market of the future likely will demand positive margins at an increasingly granular level–that is, within each product line and from each market segment. Like any insurer, they are likely to accept risk, but are less likely to tolerate systematic losses, even when offset by profits from other business.

While this may sound futuristic, it is a model with a proven track record in other industries, and the leading health insurance players are already experimenting with it. This technology is the art and science associated with advanced data modeling and predictive analytics.

Defining Predictive Analytics
Insurers, of course, have always wanted to look at margins at a granular level. This has been historically difficult due to the high variance between health insurance claims and the limited tools that insurers have had available to them for analyzing experience. The industry has had to rely upon “slice and dice” techniques along a handful of parameters, such as age, gender, area, group and plan code. Slice and dice analyses require that each resulting “data cell” have credibility. The availability of data via data warehouses and the ability for business users to generate ad hoc “slice and dice” reports has advanced due to new software, and this has certainly helped matters.

However, the unavailability of a sufficient quantity of statistically valid data (credibility requires a minimum data set for at least 5,000 member lives, for the most part) has made it difficult for decision makers to assess the impact of a range of potentially significant factors. The ability of decision makers to analyze potential covariances between known factors also has remained limited. Health insurers usually treat age and geography as independent drivers of health costs, but geographic differences may be more relevant within certain age bands than within others.

Industry decision makers have been so inundated with historical claims and clinical data that efforts to pull in related data about primary customers have been limited. Data sets that represent significant potential for the health insurance industry include psycho-demographic and socio-demographic data about key segments within their target markets. Long used by the retail and financial services sectors, psycho- and socio-demographic group profiles such as “soccer moms” or “shotguns and pick-ups” are rich with details about the typical characteristics or buying patterns of particular market segments.

As a discipline, predictive analytics is, at its core, a set of techniques that does not rely upon traditional slice and dice techniques. It also does not require that explanatory factors be independent of each other, an assumption which is implicit in much slicing and dicing. In the hands of a skilled practitioner, predictive analytics works on large, complex sets of data to uncover hidden relationships and trends that could not be found via slicing and dicing.

Like any data analysis, for predictive analytics, data quality drives quality of analysis. Insurers that have a data warehouse which can deliver clean data relative to a large number of parameters are well positioned to use predictive analytic tools. But since predictive analytic software runs off large flat files and not relational databases, data–not data warehouses–are the core requirement for predictive analytics. The in-house data can be combined with quality psycho-demographic data readily available from outside sources.

Predictive Analytics in Use Today
Predictive analytic tools and techniques are being used to manage margins on an increasingly granular basis today. Risk predictors, with products such as Symmetry and DxCG leading the market, use predictive analytic techniques to forecast future claim levels based on the age, gender and detailed health insurance claims history of an insured party. The resulting predictions are much more accurate than the traditional age- and gender-based forecasts that were and are used for refined pricing of individual and small group insurance. Because they require detailed claims history, the Symmetry and DxCG risk predictors are used primarily at renewal. The Milliman Medical Underwriting Guidelines, however, rely upon similar methods for setting the debit points that are assigned to medical information obtained via medical questionnaires.

In addition, although companies are keeping the results proprietary, leading companies have “skunk works” of predictive analytic teams sifting through their data. The predictive analytics department at one Midwestern insurer has been steadily growing while the separately managed actuarial department, which relies upon more traditional analysis, has been shrinking. The director of that department says that the department’s growth is limited by the supply of qualified applicants.

Although predictive analytic techniques are already used in the health insurance industry and are here to stay, their use is in its infancy. It takes both a little imagination and a review of other industries to see their true potential. Here are only three of the potential uses of predictive analytics.

Individual Underwriting. Twenty years ago, obtaining a mortgage meant completing a lengthy, multiple page application and submitting supporting information. Mortgage underwriting was a manual “art,” not a science. The decision to accept or reject an application came back weeks after the application was submitted. In short, it was much like individual health underwriting is today. The predictive analytics pioneers in the mortgage lending business found that only a handful of questions from the application, when combined with information from other sources, produced superior underwriting results.

Today, a mortgage application can often be underwritten in minutes, at a price more appropriate to the risk, making mortgages more affordable to the best risks and opening up the market for the sub-prime risks that formerly would have been denied. The individual health insurance market could be posed for a similar revolution.

Large Group Segmentation and Product Design. Consumer services companies use technologies that may have applicability for large group insurers, where the insurer does not have individual underwriting discretion and has to set a single price for every employee within the group. Telephone companies that offer service packages are not content selling to the maximum number of people. Instead, they use predictive analytic techniques to segment their business by usage patterns and offer an appropriately designed and priced plan for each segment. Then, they develop marketing materials to steer people toward the correct plans for their needs, and subsequently evaluate whether they obtain both total sales numbers and the desired segmentation.

Of course, insurers have been designing products and prices for use in large group multiple-option environments with specific target segments in mind, but the employee segmentation and resulting product designs have been based on as much intuition as data. In June 2005, the

McKinsey Quarterly published a “leading edge” article entitled “Designing Better Employee Benefits” which proposes that insurers and employers take a systematic look at employee health plan selection. This is entirely consistent with current large group focus on increased consumerism and the defined contribution funding model.

Channel Management. Channel management has always been notoriously difficult for insurance companies. Insurers have struggled to attribute profitability (or losses) to individual agents and producers. As a result, agents often have had incentives to emphasize top line growth at the expense of the ultimate desirability of the business. In the future, the granularity of analysis made possible by predictive analytics will allow insurers to set not just top line growth goals but also goals for profit and performance within each market segment.

Predictive Analytics for Tomorrow’s Insurance Market
Predictive analytic techniques are providing insurance companies with new perspectives on their business and, in turn, will fundamentally alter how the business is organized. The full impact cannot be predicted, but it is easy to envision the following impacts:

Dramatic changes in rating methodologies. Health insurance rates historically have been based on a limited number of readily available parameters, such as age, sex, and area, available at the time of underwriting. Predictive analytic techniques have been used in auto insurance rating for years. Far more variables are used for calculating auto insurance rates than individual health insurance rates. Rating for individual and small group business has already begun to change.

Target marketing. When insurers can identify the characteristics of their most profitable insureds, they will naturally work to enroll more of these insureds. They will want to offer retention incentives specific to the profitable insureds they already have. The potential for target marketing will be enhanced if insurers can identify the psycho-demographic characteristics of a potential insured, prior to obtaining specific underwriting information. They then can market to that profile using databases from outside entities.

An explosion in benefit design. When desirable targets are identified, the next step is to offer appropriately priced products, which are attractive specifically to that target.

New products targeted at under-served market niches. Although the most advanced insurers may identify the most profitable targets first, other insurers will soon follow, chasing the same target and diminishing the profits. The result will be a search for ever more under-served market niches. At the appropriate price, a wide variety of business can produce positive margins. The sub-prime mortgage market and the impaired risk auto insurance markets exploded after the advent of predictive analytics. The result was less auto insurance in state mandated risk groups and the greatest access to home mortgages in history.

Toward a “retail model” of group insurance. Consumerism is putting the financial responsibility for healthcare decision making back into the hands of employees. Defined contribution approaches leave employees free to choose among a variety of plans. Insurance companies will use and are using data analysis enabled by predictive analytics to design products with appeal to specific subpopulations of employees. Group insurance design increasingly will need to respond to the preferences of individual customers.

Faster reaction time. Besides predicting the future, predictive analytic tools can be used to identify problems before they can be identified via traditional methods. But that knowledge is only useful if it is acted upon.

They see the potential power of predictive analytics; they also recognize the huge learning curve. At this point, the healthcare industry is just beginning to adopt such tools, as the players learn everything they can. Staying marginally ahead of the competition by using predictive analytics to manage profitability at the granular level is, therefore, becoming a key goal.

© 2006 Nelson Publishing, Inc