Demand Forecasting – Science or Art?


Predicting demand used to require a mixture of statistics and good instinct. The gaps left by statistical models were filled by the experience of seasoned forecasters, with educated guessing essential to achieving the most accurate outcome. Today, the formula is different. AI has been a true game-changer, delivering results previously thought of as impossible. Can demand forecasting finally become an exact science?

It used to be believed – and to some extent still is – that one of AI’s biggest barriers was creativity, particularly artistic creativity. We now know that this is not true, and that AI can challenge humans in all areas. For example, the first AI art exhibition, which took place last summer in London, brought together the works of Ai-Da – a humanoid with an artificial brain. In just a few days, all of ‘his’ drawings, paintings and sculptures had sold out – for over a million Euros! Proof, if you needed any, that AI is permeating areas that we believed to be outside the realm of rationality and science – areas where only instinct and human talent could act or provide answers. AI shows that, contrary to firm beliefs, algorithms can find answers to questions that were considered untouchable by science. Demand forecasting is one of these areas.


Upsetting the apple cart

Although they are very powerful, classic statistical tools reach their limits as soon as confronted with a situation that would not be considered ‘normal’. For example, how do you anticipate demand for a product which is going to be the object of a big promotion next month, when the product has never previously been promoted? How do you know to what extent the launch of a new product, for which there are no existing statistics, will drive demand?

For any of these non-standard situations, in which statistics cannot be drawn upon, forecasters generally rely on their experience. Intuition kicks in. This is the point at which demand forecasting ceases to be a science and becomes an art.  It also is precisely at this point where AI is able to transform everything. By analysing these correlations and the probabilities of correlation with other items, other sales locations and other contextual data, AI is behaving in the same way as the instinct or intuition of the forecaster. This development flips everything on its head, and instead of an art, demand forecasting becomes a science.


Science is always more exact

In retail, like anywhere else, the value of a forecast can be measured by its reliability and its accuracy. While the forecast of demand for known products may have achieved high levels of accuracy, the forecast for non-standard products or situations always made the accuracy rates fall globally – creating a barrier that used to be considered difficult – not to say impossible – to break through.

AI can remove this barrier. Not only can it handle immense volumes of data, but it can identify correlations that make it possible to explain things that are outside the norm. It gives forecasters the means to take a major step forward in their work, providing them with the missing links needed to explain anomalies, effectively automating the act of intuition to provide unprecedented accuracy, even in situations where there is little or no data directly linked to an event.

With the use of AI, forecasters’ talents can be put toward improving these new heights of precision. Their intuition will allow them to put forward and test additional contextual data around events in order to enrich the work AI is producing, even further enhancing its performance. It’s not an exact science yet, but with AI, demand forecasting is getting closer and closer to becoming just that.  The results will be greater accuracy, faster execution  leading to near real-time re-forecasting and the inclusion of AI-forecasting as a primary driver of stock ordering, inventory management and supply chain execution.