How to Improve Forecast Performance by Reducing Human Bias

This article is an Executive Summary of a NorthFind Management Whitepaper on the application of Behavioral Economics to Demand Planning. To read the complete paper, please email the author, Jonathon Karelse, at [email protected]

Promising advances in the areas of Machine Learning and Artificial Intelligence have disproportionately skewed conversations about performance improvement opportunities in Demand Planning and Forecasting. Despite the potential for gains in these areas, many companies face far more fundamental daily challenges. And whatever the degree of computerization and automation, every Demand Planning process by its nature contains multiple human touch-points.

Research into decision-making and Behavioral Economics has proven conclusively that while most humans have the capacity for rational, objective decision-making; we are all consistently influenced by unconscious biases and heuristics. Nobel-laureate Daniel Kahneman and his long-time collaborator Amos Tversky established more than 40 years ago that the effect of human biases is so prevalent that there are many parts of a company's portfolio that should only be forecasted statistically. If left unidentified and unmitigated, these biases will find their way into Demand Planning and forecasting processes and undermine the performance of these key operational functions.

To understand the prevalence of a set of major bias and heuristic types - nearly 20 - in both the general and Demand Planning populations, and in so doing to develop plans for mitigating them; NorthFind Management devised a study which engaged nearly 500 planners and non-planners globally, including a substantial component of Heineken’s global Demand Planning team. By posing a series of personality, practical and abstract questions, we were able to determine not only the prevalence of biases and heuristics in these populations; but also what impact they would have on Demand Planning performance; and in the process, we discovered some links to personality types.

At its conclusion, our study found that the prevalence of certain biases was much higher than anticipated – nearly 90% of Demand Planners, for instance, were impacted by a tendency to consistently over (or under) estimate in one direction – and that forecasting performance was often significantly impacted as a result. It additionally found that the Framing Effect - a bias by which the recipient of information has their judgment skewed by the way in which the information is presented - impacted more than two thirds of planners, and caused many to amplify their biases. Given that much of the information reaching Demand Planners in their daily work comes by way of stakeholders who are consciously framing data to benefit their position, this finding was particularly significant.

The NorthFind study also found clear connections between certain personality types and biases, though we caution against the overzealous application of this information in recruitment processes, since diversity of thought is also one of the antidotes to several heuristic types. There is no question, though, that a group of personality types that follow Jungian Typologies (popularized through the Myers Briggs methodology), have a much stronger proclivity to the types of cognitive biases and heuristics that impact Demand Planning than others.

Finally, by examining the prevalence of biases and heuristics in Demand Planners, we prescribe four strategies for mitigating their effects, all of which are Demand Planning best practices. First, test for individual biases. By making unconscious biases conscious, it is possible to take direct corrective action. Second, employ a standardized Demand Planning and forecasting process which includes FVA (Forecast Value Add) to consistently test and refine the inputs to the process; as well as diarizing and measuring the effect of human overrides. This can also support formalized decision trees which can ultimately allow automation for some SKUs with the appropriate demand profile. Third, consistent and standardized education and training on Demand Planning best practices, including elements of Behavioral Economics and specific biases and heuristics. And fourth, create a team with robust technical skills, but differing viewpoints and personalities. Diversity can create diverging views which encourage deeper thought, and mitigate against the Availability Heuristic and biases like Group Think.