What: It is tempting to simplify the number of variables taken into account to predict future sales, for the sake of simplicity, but it can lead to errors varying from 1 to 3.

Why it is important: Uncertainty modelling is surely an approach worth to be considered in these times of extremely high uncertainty from one day to another in terms of customer behaviour.

Even though companies are working hard on controlling their supply chain bottleneck, now that just in time production is a common thinking and this led to the disappearance of intermediary buffer stocks, the disruption in production prediction persists. According to the HBR, it is not due to a shortcoming in the approach (or, as they call it, the ‘software’) but more to its implementation.

Researchers categorize data analytics according to 3 types:

• Descriptive analytics, answering to ‘what happened’ and ‘what is happening’ (e.g. sporting good chain The Gamma Store),
• Predictive analytics, often using advanced statistical algorithms to predict the future values of the variables on which decision-makers depend (e.g. Amazon, Procter & Gamble, Unilever),
• Prescriptive analytics, informing decision-makes about the potential consequences of their decisions and prescribe actionable strategies, based on mathematical models (e.g. Airlines, UPS).

Usually companies use a cocktail of these 3 approaches, and are usually weak at the predictive part, according to the researchers due to the assumptions and choices around the generation of data analysed. In order to reduce the noise, and improve the quality of the predictions, researchers advise to use uncertainty modelling, which aims at identifying key parameters associated with data generation to reduce uncertainty around the predictive value of that data.

The example given is a customer which increases his orders by 500 units each time that an order is placed. Companies are going to reason on an average amount of products ordered each month, and will not be able to correctly predict the future amounts ordered, if they miss the specific information of +500 pieces ordered per month. Uncertainty modelling, through advanced algorithms, allow to increase the number of uncertain parameters taken into account, and help predicting with better accuracy, as shown with a tire manufacturer example in Turkey.

Using Uncertainty Modeling to Better Predict Demand