Editor’s note: Jen Coriell is president of Impactrics, a Brentwood, Tenn., research firm.

Traditionally, the health insurance industry has priced its services on a cost-plus basis. The costs are estimated using a complex algorithm of past health care utilization and customer profile data. This process was developed and is used rigorously by the actuarial departments. However, a multibillion-dollar health insurer realized that an important element was missing from this cost-plus approach: market sensitivity to price changes. With the help of an econometric model and a revamped pricing process the insurer bridged that gap.

When an actuary examines a client’s cost and utilization of health care services they will typically find a pattern of significant gain from one year to the next. There are a lot of drivers of those costs, such as technology, an increasing number of specialist physicians, etc. - but that’s another article! They’ll then assume those same trends moving forward, adjusting for a change in the client’s profile (e.g., age, sex). This is the cost-plus approach used in a pinch by many companies. However, in the health insurance industry there is so much variability of costs across different profile factors that this actuarial approach has been the conservative mainstay of pricing strategies. Ironically, this approach can actually hurt the bottom line.

Unlike industries that sell a product and are immediately paid for it, in the insurance industry a guarantee is sold that future, unknown costs will be paid. When an insurer hikes up their prices by 15-20 percent, the “healthier” of their existing clients, knowing they’ll generate lower costs, are more apt to price-shop. The insurer is then left with the clients who tend to use more services…and generate even higher cost growth. It’s a self-induced upward spiral that whittles away at the bottom line. Why? Because even the not-so-healthy clients will tolerate only so much price increasing before pushing back on the insurer for lower margins.

Using the right drivers

Generally, successfully capturing market sensitivity through regression analysis is a matter of using the right drivers, measures, functional form and model evaluation methods. Because this was a major shift in approach for the insurer, most of the time and effort was actually spent in the first step: identifying the right drivers.

A cross-functional workgroup was established that first became familiar with the modeling process and requirements. (There were actually several workgroups as this was one component of a larger econometric model.) This was necessary not only for them to make valid decisions, but as importantly, so they could understand and have ownership of the new tool. This empowered them to communicate it to their functional areas with confidence. Once the group understood their charge, they quickly realized that customer growth doesn’t occur solely as a result of the insurers’ pricing actions. To model only their price as a driver would create an exaggerated estimate of price sensitivity. So, the workgroup developed a long list of drivers and then applied criteria in order to prioritize them. The result was a recommended handful of determinants of customer growth. These drivers were advertising and promotion expenditures, competitors’ pricing, substitute pricing, and the size of the market itself.

A second workgroup was formed to take the model to the next stage: estimation. Estimation began with the development of solid measures for demand and its drivers.

The workgroup determined that demand for the medium-to-large-size employer groups should be the first model since it was the segment with the largest profit potential. The group also determined that because the margins varied dramatically by product group, then demand and price sensitivity for each group would be individually determined. The measure of demand, by product group, was the number of subscribers. This measure was preferred to membership data as it nets out the influence of changing family size.

Measuring advertising and promotion expenditures was straightforward as the company maintained historical financial statements. Because the dependent variable was a non-monetary volume indicator, the expenditures had to be adjusted to net out the trend in overall consumer inflation. Not doing so would have forced the model to pick-up inflation as an influence, which the first workgroup had determined was not valid.

Competitor pricing data for an array of products was found within the state’s department of insurance records. Collected for a 10-year period, composite indices were developed which tracked the change in the company’s vs. competitor pricing for both similar and substitute product groups.

Finally, the size of the market was measured by employment in the geographic market. Weights were assigned by industry and employer size that reflected the propensity to insure.

Functional form and model evaluation

After the workgroup had identified the appropriate measures, the functional form of the model was determined. That is, are the interrelationships best described in logarithmic form? If so, to what base? Rarely are the series related in a simple linear fashion (e.g., a 10 percent price increase yields a 5 percent decline in the subscriber base). Other mathematical considerations were made to determine the appropriate functional form of the demand equations. (This functional form is different for each company as they are influenced in varying degrees by customer loyalty, brand value, etc., values captured intrinsically in price.)

A series of evaluative statistics were calculated and revealed a solid fit. Both in- and out-of-sample statistics were determined to ensure that the form specified was valid even when constrained to different time periods.

The result was a sound estimate of the impact from a change in price on the demand for an array of products.

Reign in the price increases

The pricing process used by this insurer began with the market segment team leaders being presented with the actuarial recommendations for price changes. Then, with the pressure of performance targets breathing down their necks, they’d attempt to reign in the price increases by scrambling to disprove the cost-plus calculations.

The estimates of market sensitivity were brought into the pricing process with forecasts using various assumptions. A third workgroup was assigned the task of developing forecast assumptions for the drivers. Combining the driver forecast assumptions with the statistically-determined relationships allowed for the market’s perspective to be represented in four ways:

1) Demand forecast resulting from the cost-plus pricing and the estimate of price sensitivity vs. the traditionally-used straight-line demand projection (Figure 1). The price sensitivity estimate allowed the market segment team to see that the cost-plus pricing had a stronger downward pressure on sales than previously thought.

2) Pricing and advertising/promotional expenditure scenarios needed in order to achieve sales targets (Figure 2). Pricing and promotion are classic marketing levers that can be used to reach targets. This chart used the company’s unique price and promotion sensitivities to determine different combinations of the two. A higher price trend requires more promotional dollars to compensate sales growth.

3) Gross margin forecast, assuming cost-plus pricing based on price sensitivity vs. straight line revenues (Figure 3). This forecast used the same assumptions as (1). In addition, the medical cost forecast generated by the econometric model was used. (Because of the complexities of the health care industry, the development of the medical cost model was a much longer process.) The traditional straight-line demand projection was misleading in its impact on gross margins. After adjusting for the company’s price sensitivity, the cost-plus price was proven to diminish margins.

4) Gross margin forecast, assuming various pricing and advertising/promotional scenarios to achieve sales targets (Figure 4). This forecast used the same scenarios as (2), combined with the econometric model’s medical cost forecast. One combination of price and promotion yields higher margins than the other. It was also evident that pricing yielded more bang for the buck than promotion!

ROI estimate

As with pricing, the model also yielded sensitivity estimates from advertising and promotions. A baseline demand forecast was first generated by holding pricing to the cost-plus trend over the forecast horizon and advertising/promotional expenditure increases at historical rates. Then, an increase of 15 percent was added to the advertising/promotions budget. This generated a ramped-up forecast of demand. The increase in demand was then multiplied by the price to determine revenue gain. The medical cost forecast was held at its baseline, creating an impact estimate. This estimate, divided by the 15 percent increase yielded a return on investment. The market segment team was able to use that ROI estimate for comparison to other investments that the company was considering during its corporate planning and capital budgeting season.