Predictive analytics efforts can significantly impact commercial success by optimizing sales and marketing efforts. Commercial leaders experienced in the utilization of predictive analytics will have developed a good sense of where predictive analytic efforts can be deployed most effectively and with the highest probability of success. For organizations with little experience in this area, however, getting started can be daunting. Collecting the right data, building a reliable infrastructure, and assembling the required analytic talent is just the beginning.
Once all of this is in place, successful predictive analytics efforts depend on the proper selection of projects. One practical dimension for prioritizing projects is the degree of challenge and risk they present—a consideration of particular importance for organizations looking for early “wins” to cement managerial commitment to the discipline. A useful way of characterizing predictive analytics challenges is through what we call the Intervention-Environment Matrix (IEM). (See Figure 1.)
What Drives Challenge and Risk?
The IEM brings together two key dimensions driving challenge and risk in predictive analytics projects. The first is the familiarity of the intervention being considered. The second is the familiarity of the environment in which the intervention is to be implemented. Juxtaposing these two dimensions creates a simple framework for selecting projects and setting expectations for analytic success.
When both the intervention and environment are familiar, the challenge faced in predictive analytics is relatively low. For example, determining which physicians would be most responsive to increased sales call frequency in a stable market falls squarely in the “low challenge” quadrant (Box 1 of the IEM). By contrast, determining physician responsiveness to a novel reimbursement support program in a dynamic and rapidly evolving insurance coverage environment would fall in the “high challenge” quadrant (Box 4 of the IEM). Combinations of familiar environment and novel intervention, or the reverse, would fall in-between in terms of analytic challenge.
In quadrant 1, historical data, a clear definition of the intervention and a good regression model may be all that is needed to develop predictive algorithms that optimize a sales or marketing action. Moving to quadrants 2 and 3, or 4, however, the relevance of historical data diminishes. If the environment is changing or the intervention will take place reasonably far into the future, past performance becomes less of an indicator of future performance. The exact nature of the intervention may also be harder to grasp. As a result, the reassuring “objectivity” of mathematically determined predictive models gives way to an increasing amount of subjective judgment. Such judgment may be informed by market research, analogues, scenario planning exercises, or simulations, but it is still subject to the many cognitive biases that are well known to impair optimal decision making.
None of this is meant to suggest that predictive analytics projects that fall anywhere outside of quadrant 1 should be avoided. The IEM merely serves to ground expectations such as when the predictive analytics task departs from the familiar, the goal may be less to get the “right answer” than to narrow down the probabilities around potential outcomes and help management play the best hand possible.
