Estimating Sales Force Efficiency in the Pharma Industry

Estimating Sales Force Efficiency in the Pharma Industry
Ashish Ranjan Jha, Managing Principal, Analytical Wizards
Abstract:
The field force continues to be the most important channel of promotion for pharmaceutical companies. Every year manufacturers earmark a substantial portion of marketing budgets for sales force visits to physicians. Tighter regulatory environment, restricted access to physicians and diminishing returns on detailing mean that the marketers are under pressure to improve the efficiency of these sales force visits. Knowing the efficiency levels at which medical representatives are operating is a good first step towards giving that extra boost to portfolio sales. Training and hiring efforts are more focused and effective when the strongest and the weakest links in the sales force are known to marketers. Given the constant pressure on cost, it might be prudent to direct efforts towards improving the efficiency of the existing workforce rather than simply increasing the size of it. Stochastic Frontier Analysis is a potent technique not only to measure efficiency but also to gain insights on ways to improve it.

There is a great deal of variation in the efficiency levels of sales representatives. Those in the higher quintiles of efficiency have 1.5 to 4 times higher efficiency than those in the lower quintiles on average.


Keywords: sales force efficiency, field force efficiency, Stochastic Frontier Analysis

Business Context
A leading pharmaceutical company was looking for ways to improve its sales. Amongst other options, it decided to review performance of its sales force team. Efficiency at the sales representative level was chosen as the metric to be evaluated, as was any difference in the efficiency level of representatives in different tenure bands. In other words, it wanted to understand how the efficiency of a sales representative changed as it tenured within the organization. For this study, business team clubbed sales force representatives in three different tenure bands viz. 0 to 3 years, 4 to 9 years and greater than 10 years. They wanted to depute their training resources where it mattered the most.
Data Description
The project required cross-functional collaboration between client teams. The Sales department and the Human Resources Development (HRD) department worked closely to ascertain the objective of the study and then to collate data accordingly. Data included sales operations provided with monthly brand sales, market sales, number of samples distributed, and calls data on each sales representative. HRD was provided with the tenure of the representatives within the organization.  Data was limited to one country, one brand and was in a one-year time frame.
Challenge
Commonly used metrics for sales force efficiency include the total number of calls, number of calls per day, time spent per call, etc. But these are all surrogate measures that have nothing to do with the efficiency of sales representatives. Moreover, these measures do not take other factors into account e.g. market size, other forms of promotion, etc. Therefore, there is a need for a robust methodology which takes a comprehensive view of efficiency.

A benchmark is needed to measure efficiency. Standard ordinary least square regression techniques can be used to fit a curve between sales and input parameters. This regression curve can then be used as a benchmark. But a regression line passes through the mean and, therefore, it is a representation of average performance rather than the best performance. Moreover, regression measures each deviation from this fitted line as random noise. But, in the context of brand sales, we know that the deviation is caused by both random noise and the efficiency of the sales representative.
 
Another common methodology for measuring efficiency is Data Envelopment Analysis. Unlike least square regression, it measures efficiency with respect to the best performer. One problem with this approach is that it treats each deviation from the fitted line as a measure of inefficiency. But we know that deviations from the best frontier line have two components – one is obviously inefficiency, and the other is random noise in behavior which is inherent in any process.
Solution Design
Efficiency can be defined as the ratio of observed output and maximum possible output for a given level of input.  But, considering the stochastic element of sales, this ratio should be calculated after accounting for the random noise component.
Figure 1


Refer to Figure 1. Point A lies on a curve, which is called the sales frontier. The sales frontier is the line that passes through the maximum achievable sales at all levels of possible inputs. Therefore, it can be said that point A is efficient if random noise has already been accounted for. Point B, on the other hand, is inefficient because it is lower than the sales frontier. Since sales has a stochastic component we should account for the random noise too. Point C is the level of sales post adjustment for random noise. Now, BC is the random noise effect, whereas CA is the inefficiency effect. Therefore, if noise effect can be segregated from efficiency effect, then efficiency can be derived as OC/OA. It turns out that the Stochastic Frontier Analysis incorporates both random noise and efficiency in the model separately, and therefore, it is quite suitable for estimating efficiency of the sales force team.
Solution Development
As mentioned above, Stochastic Frontier Analysis incorporates both efficiency and the random noise in the model. It is a parametric method which means one can model any functional form between the output, which is sales in this case, and an array of inputs such as promotion activities and market size. Moreover, it treats the best performers as the benchmark and estimates efficiency of the rest with respect to this benchmark.

Stochastic Frontier Analysis subtracts a non-negative random variable, a measure of inefficiency in the process, from the standard production function and then employs a maximum likelihood estimation technique to estimate model parameters. The model can be represented by the equation below (see Introduction to Econometric Production Analysis with R, Arne Henningsen, 2015):
 


While v can take any value on number line, u is a non-negative number in this equation.
 
This u is a measurement of inefficiency. Once this inefficiency term has been estimated for each sales representative, then one knows the overall efficiency at which the sales force is operating and the zones of inefficiency in the overall pool.

Sales can be assumed to be related with inputs in a Cobb-Douglas functional form. In other words, log of sales can be modeled as a function of log of inputs viz. promotional activities, market size, competition activity, etc.

Since it is a parametric approach, one can assume any functional relationship and model for any business assumption, so long as the functional relationship remains linear in parameters. For example, there may be territories with a low level of brand share historically. One would expect efficiency of sales representatives assigned to these territories to be low. In this study, after a sales force restructuring exercise, the sales team had assigned new sales representatives to such territories and wanted to make sure that new representatives were not penalized with low efficiency for traditionally low levels of brand sales in their respective territories.
 
Two approaches were proposed to accommodate this in the model. First, it was proposed to add prior year’s sales to the model to act as a control for the historical level of sales. Alternatively, it was proposed to model panel data at sales representative and month level, instead of cross-sectional data, and evaluate efficiency for each month for each representative. The latter approach would allow the business to see if efficiency of a representative has been on an increasing or decreasing trend over time. With much deliberation, the company saw value in the second approach and therefore, panel data was modeled. 
 
The final modeling equation was
 


Model Result
A high level of variation was observed in the efficiency of sales representatives as derived from the model. Those in the highest quintile were approximately 2.5 times more efficient than those in the lowest quintile of efficiency. (Figure 2)
Figure 2

Almost a third of the territories with high market size were found to be served by those at below average efficiency level. This group was operating at an efficiency of only 0.43.  There was a good opportunity to increase sales just by pushing efficiency of the sales force operating in this market segment. Analysis revealed that even a minor 10% increase in efficiency of this section of representatives could boost overall sales by more than 1%. This was seen as a big opportunity by the business team. (Table 1)
Table 1


Though some decline was observed in the efficiency of representatives as they tenured within the organization, the difference was not found to be significant.
Conclusion
Given the prevailing market conditions, it is imperative for pharmaceutical companies to increase the return on investment they make on sales force visits.  Stochastic Frontier Analysis is an effective technique to understand not only the overall level of efficiency at which the workforce is operating, but also to detect the zones of inefficiency and the underlying reasons driving that inefficiency. It models sales with all relevant input factors using the functional form of choice. Thus, it leaves a lot of control in the hands of users to model the type of relationship between sales and input factors that are in sync with business hypotheses. Unlike regression and data envelopment analysis, Stochastic Frontier Analysis assumes that any deviation from the most efficient sales frontier is due to both random noise and efficiency. And, therein lies its strength.
About the Author
Ashish Ranjan Jha, Managing Principal, Analytical Wizards, has years of experience in marketing analytics space serving global clients in Pharmaceutical and CPG industries in areas such as marketing mix optimization, sales force efficiency improvement, understanding and predicting customer behavior, personalized marketing, etc. He is skilled at utilizing statistical modeling and machine learning techniques to solve diverse business problems.
References
1 Henningsen, Arne. Introduction to Econometric Production Analysis with R. 2015.