Aktana CEO David Ehrlich and I had the opportunity to attend the Enhancing Sales Force Productivity Conference recently at the University of Missouri. It’s an annual conference that brings together sales and marketing researchers from around the world to share their findings on using AI (artificial intelligence) to improve sales force management and performance. As the only commercial entity in attendance, we listened to several talks by academic experts and shared our own industry perspective in David’s keynote address.
After two jam-packed days of networking, panel discussions, and research presentations, I came away with three main takeaways:
1. It’s hard to use machine learning in marketing
One recurring topic at this conference was the challenge of solving marketing problems with machine learning (ML). Dubbed the “black box prediction model,” the predictive performance of machine learning models comes at the expense of interpretability. For example, random forests, which Aktana uses to personalize the sequence of emails sent to a given healthcare provider, build a series of decision trees based on random re-samplings of the data and combine the results across these trees. While random forests have built-in variable importance measures, they do not have the same interpretability as the coefficient in a regression model — hence the “black box prediction model.”
As I shared with fellow conference attendees, Aktana overcomes the interpretability challenge by using the ML model as the basis for estimating counterfactual outcomes under hypothetical interventions. By generating parameters from the causal inference literature, we then rank the potential interventions by their impact on an outcome of interest.
Finally, we must determine how to use ML models to drive human behavior. We may be able to predict future behavior well by using machine learning, but ultimately we want to make suggestions that actually modify behavior. This remains a challenge in the field and is an area we aim to collaborate on with professors and students from University of Missouri.
2. Humans can and should co-exist with the machine
One of the most illuminating sessions during the conference was a panel titled Sales Force Management at the Cusp of the 4th Industrial Revolution, discussing the use of AI in sales force management. The panelists optimistically (or perhaps pessimistically) posited that the sales rep will eventually be replaced by AI. Machine learning, after all, replaces tasks that are near-impossible for a human to do efficiently, such as using real-time and historical interaction data between a sales rep and healthcare professional (HCP) to predict the best next action. However, as the panel discussion evolved, it became clear that there are intrinsically human abilities such as emotional intelligence, capturing nuance, or understanding new unknowns that simply cannot be captured in a ML model.
At Aktana, we use ML models to not supplant but rather enhance sales reps’ relationships with HCPs. We also face the unique challenge of modeling not just the sales rep-HCP relationship, but how the rep interacts with our software. Ultimately, the impact that our suggestions have on HCPs is mediated through the sales rep, and we have more control over how the reps interact with our product than how the reps interact with the HCP. Our ML models aim to strengthen both of these relationships by providing output that “feels right” to reps and in turn maximizes effectiveness for their interactions with HCPs.
3. Real-world application is vital to the advancement of data science
There were many interesting theories regarding sales rep and customer behavior presented at the conference that require validation through data and analytics. One researcher, for example, is modeling salesperson resilience and how that predicts success. Another is working to identify optimal times for upselling. These are fascinating areas of work that, substantiated with real-world data and stress-testing, could significantly advance sales force management.
At Aktana, we’re fortunate to work closely with large global life sciences companies and utilize rich data to solve real-world challenges and business questions. Looking ahead, we are excited to collaborate with these academics and use Aktana’s data to bring some of these exciting, relevant theories to life.