Supply Chain Planning: Why we find it hard to trust Probabilistic Technology

As humans, we grapple with uncertainty, and our preference for stories over statistics often influences our decision-making processes.

The fundamental issue lies in our cognitive biases, particularly our struggle with probability. Our brains are wired to focus on immediate threats and historical occurrences, making it challenging to engage in thought experiments about future events. This innate bias leads to misconceptions and miscalculations of probabilities, shaping our distrust in technology that deals with uncertain outcomes.

Deterministic technology, which follows fixed rules and algorithms, is relatively easy for us to trust. Computers executing instructions deterministically provide consistent outcomes, adhering to known rules and making us feel in control. However, these systems struggle with uncertainty and adapting to novel situations, risking catastrophic failure when faced with unexpected events.

On the other hand, probabilistic technology introduces statistical methods and probabilities to decision-making processes. Machine learning models and natural language processing exemplify these systems, offering adaptability to changing conditions and handling complex, real-world data. Probabilistic models assess risks and prioritize actions, but their outcomes can be unsettling for users who prefer certainty.

The crux of the matter lies in our familiarity bias. Deterministic technology aligns seamlessly with our intuitive understanding of cause and effect, fostering trust in what we can predict and control. Probabilistic technology, by contrast, introduces an element of risk that provokes unease. The potential for errors due to imperfect data or assumptions, coupled with the challenge of understanding why a probabilistic system made a specific decision, contributes to the skepticism surrounding these technologies.

Which 50% of our orders can we trust the system to place automatically? The (unsatisfactory) answer is ‘it depends’.

The article “” uggests that demand forecasting could be a game-changing technology thanks to AI. With access to increasingly vast oceans of data, Machine Learning promises to transform our ability to predict the future. Uncertainty won’t be eliminated, but it will significantly reduced and better understood.

We can start by looking at the accuracy of machine predictions, compared to the accuracy of human predictions. This is sometimes called “forecast value add” – positive when human intervention improves forecast accuracy, negative when technology comes up with a better answer. In practice, we see more negative forecast value add, than positive. Humans are usually over confident in their ability to predict the future compared to machines.

It is only part of the answer, but if we start by looking at items with better statistical forecasts than human consensus forecasts we have a good list of candidates for automation.

As we develop this theme, in future articles we’ll explore the importance of intrinsic predictability vs forecast accuracy, and discuss the relevance of six sigma techniques to the automation of supply chain planning.