Based on Habel, Alavi & Heinitz (2023, Journal of Marketing Research)

Authors and Institutions

The study was conducted by Johannes Habel University of Houston , Sascha Alavi Chair of Computing in Engineering – Ruhr University Bochum and Dr. Nicolas Heinitz Ruhr University Bochum . Together, they explore why predictive analytics in sales often fails to deliver the expected results.

Research Question and Hypotheses

Predictive analytics promises to transform sales, yet adoption is inconsistent and performance improvements vary. The authors hypothesized that effectiveness does not lie in the tool alone but depends on two sets of factors: customer characteristics and salesperson attributes. They also examined whether managing expectations, by warning salespeople that predictions may include errors, could improve adoption. Finally, they proposed that the benefits of predictive analytics are dynamic and evolve as salespeople gain experience with the technology.

Methodology

The central part of the research was a 12-month field experiment in a B2B wholesaler selling construction supplies. The company implemented a churn prediction model based on XGBoost, generating around 13,000 customer-level predictions each month. Salespeople were randomly assigned to three groups: one receiving churn predictions only, one receiving predictions plus expectation management disclaimers, and a control group with no predictions.

To complement this, the authors surveyed 130 salespeople, measuring their trust in algorithms, selling abilities, and orientations such as adaptability and learning. They analyzed the results using causal forest models, which allowed them to estimate heterogeneous treatment effects across different customer and salesperson contexts. To validate findings, they also conducted controlled experiments with salespeople outside the field setting.

Analysis

The study introduces the concept of COPO or Characteristics of the Predicted Object. These are customer-side factors such as prior revenue, revenue stability, and the probability of churn. The analysis showed that COPO variables play a decisive role in determining whether salespeople can turn predictions into revenue gains. In fact, they explained 38% of the variation in outcomes, more than any salesperson attitude or perception.

Non-COPO factors (such as a salesperson’s trust in algorithms, experience, or selling style) still mattered, but their influence was more indirect. They shaped effectiveness primarily through interactions with COPO variables. In other words, the tool’s impact emerged from the combination of who the customer was and who the salesperson was.

Conclusions

The first key conclusion is that predictive analytics delivers the greatest value with large, stable customers who either have a very high or very low risk of churn. Salespeople are able to prioritize these accounts and either prevent losses or capitalize on remaining opportunities.

Second, while salesperson characteristics are less dominant than customer characteristics, they still affect outcomes. High-performing and learning-oriented salespeople were better at turning predictions into action.

Third, the benefits of predictive analytics increase with time. Salespeople with more experience and a willingness to learn were able to improve their use of the tool as they adapted, which shows that effectiveness is not immediate but requires a learning curve.

Fourth, predictive tools change behavior. Salespeople did not just use the churn scores passively; they adapted their strategies by dedicating more time and even granting deeper discounts to high-risk, high-value customers.

Finally, expectation management turned out to be a double-edged sword. Warning salespeople about prediction errors helped a small subset—those who distrusted algorithms but were highly motivated to learn. For most others, it backfired by undermining trust and discouraging usage.

Takeaways

The study offers important lessons for managers. Predictive analytics should not be rolled out universally but targeted to contexts where it is most likely to succeed: high-value, stable customers and salespeople who already perform well or show a learning mindset. Adoption requires patience, as benefits emerge over time with practice and adaptation. And while transparency about algorithm limitations is important, managers should avoid overemphasizing errors, as this can reduce trust in the system.

Predictive sales analytics is not a one-size-fits-all solution. Its success depends on matching the right tool with the right customers, deploying it to the right salespeople, and supporting its use through learning opportunities rather than simplistic expectation management. This makes predictive analytics less about the algorithm itself and more about the interplay between data, people, and context.

Silicon Valley Certification Hub