by Patrick Burns, Principled Strategies

Maximizing ROI Using Optimization Techniques

Optimization techniques can be employed to maximize the ROI associated with a brand’s promotional mix. In this case study, we’ll look at how optimization techniques were effectively applied by measuring physician response to a myriad of promotional and non-promotional activities, and accurately forecasting physician-level prescribing behavior. This work began by calculating a promotion sensitivity measures for nearly 300,000 prescribing physicians. Then, by reallocating tactics from physicians who were less responsive to those who were more responsive, it became possible to extract maximum ROI for each dollar invested in the promotional mix. In this particular instance, optimization unearthed an opportunity to improve this brand’s revenue by $50M.

Optimizing return on marketing investment (ROMI) for a pharmaceutical blockbuster brand facing patent expiry.

Class: Oral contraceptive
Revenue: >$1B
Field force size: 4,000
Physicians: 300,000
Achievable ROI: $50M

Situational background

For purposes of anonymity, I’ll use the generic term PharmaCo to refer to this company and will replace any proprietary data. PharmaCo, a top-25 pharmaceutical manufacturer not long ago turned to our team to assess the effectiveness of its promotional mix investment for its flagship brand, a popular oral contraceptive with dominant market share. With the patent on the oral contraceptive set to expire in eighteen months, PharmaCo wondered when to begin withdrawing promotions and accept the inevitable massive decline in market share that was sure to accompany the loss of patent protection.

Conventional wisdom suggested that an immediate, though measured, decrease in the overall level of promotional investment was called for, to include a slow decrease in the number of sales professionals representing the brand in the market. Marketing, Market Research, and Sales leadership, though, did not feel sufficiently armed with strategic and tactical information that would guide them in the process of minimizing the expected adverse financial impact. In other words, with loss of patent protection looming on the horizon they needed a solution that would minimize the expected loss in share for their flagship brand.

PharmaCo wondered in particular about the impact of detailing (a presentation by a sales professional to a physician of the benefit of PharmaCo’s branded drug) and sampling on prescribing behavior. Product samples are dropped off at physicians’ offices frequently. The sales representatives hope that dropping off these samples will encourage physicians to start new patients on this particular product, following the sample with a prescription. Does dropping off samples result in incremental sales?

Other groups tried to analyze this question and struggled. A number of reasons were provided to explain this. First, when a sales representative drops off a sample, that visit to the physician office generally begins with a detail where they educate and inform the physician on the product benefits. So the incremental effect of the sample may be difficult to distinguish from the detail itself. Second, it is possible that sales representatives may not accurately record when they drop off samples and when they conduct details. Third, statistical models may not have included all relevant data that can impact the promotional responsiveness for the brand.

The Solution

Our team worked closely with PharmaCo to gather all of the available electronic data available on the oral contraceptive market. We prepared the data, loaded it into the statistical modeling environment, and conducted an analysis of the impact of sampling and detailing on sales for the oral contraceptive. Initially, we anticipated either providing insights on the level of responsiveness or discovering that the available data was too incomplete to tackle the question. We worked closely with members of the Marketing, Market Research, and Sales teams to develop models that could be simply understood and easily explained to management but that also were developed within a robust and powerful analytical framework to be highly predictive.

Data collection

In the first phase of analysis, we worked with PharmaCo staff to identify variables that were likely to have the most predictive power. The data sources that were investigated included:

• Prescription sales for own brand and competitor brands
• Demographics for both physicians and sales professionals
• Census, in order to better understand regional differences in the patient population
• Disease incidence and prevalence
• Promotional information for key tactics, including details and samples
• Sales force activity

PharmaCo collected the data internally and delivered it to our team. Our team then conducted interviews with key PharmaCo personnel in order to ascertain relevant business domain information and to provide us with historic context for the oral contraceptive market.

Data analysis

Data analysis was aided by the use of off-the-shelf statistical tools that were configured to provide automated data pre-screening, a process specifically designed to aid the modeler in determining the appropriate data to use when building the predictive models. The power in this approach is its ability to detect and extract complex relationships (e.g., nonlinearities and interactions) that may be beneficial in the modeling process. We designed analytical processes to identify unique combinations of variables that were best candidates for modeling inputs and ranked them according to predictive power. We then made use of flexible nonlinear (non-parametric) transformations of the original data to allow the most predictive nonlinearities and interactions to rise to the surface.

Next we employed a multi-purpose, anomalous data detection tool to help locate potentially miscoded or highly unlikely constellations of input attributes. The tool is designed to pinpoint the types of data that can cause serious and otherwise unsuspected problems during model estimation, such as multivariate outliers and leverage points.

Model estimation and optimization

The primary focus of this phase was to develop forecasts of monthly prescriptions written by physicians, conditional on sampling activity, detailing activity, medical journal advertising, direct to consumer advertising, physician characteristics, sales professional characteristics, and patient population characteristics.

The modeling and estimation processes were designed to address three of the leading problems typically encountered by modelers—speed, accuracy, and interpretability. Our techniques successfully and rapidly identified highly predictive data variables and transformations of interest, developed optimal modeling coefficients around these variables, and ultimately embodied the most predictive final model structure in a readily interpretable form. Our team succeeded in identifying the models with greatest forecasting ability for the key drivers of promotion response for the oral contraceptive.

Once the appropriate models were developed, we examined the consequences of proposed changes in the overall level of promotional investment. Literally hundreds of scenarios were considered, all requiring various combinations of changes in the investment in samples and details for this brand.

The Results

PharmaCo was faced with an intractable problem that it was unable to solve: determining the optimal detailing sampling level for its flagship oral contraceptive brand. Using a principled approach to data analytics and predictive modeling, the problem was analyzed and successfully solved.

The key finding of the analysis was that PharmaCo was undersampling and overdetailing its physicians, and that an optimized detailing and sampling strategy would result in achievable ROI of nearly $50M.

Realizing any portion of achievable ROI, though, is never as simple as flipping a switch. It requires careful planning across all sales and marketing disciplines within the firm, and in this case resulted in recommendations in three key areas: sales force transition strategies, sales force compliance strategies, and performance measurement strategies. With the proper measurement processes in place, PharmaCo is now well-equipped and empowered to drive program performance at a pace comfortable to both marketing and sales management.

Patrick Burns is President of Principled Strategies, a firm that provides their clients with the intelligence and insight they need to make smarter strategic, tactical, and clinical decisions.