Overflowing warehouses, empty shelves and problems in the supply chain. In recent years we have seen it all. Our economy has been slowed down a number of times by lockdowns, sometimes very quickly by stimulus and now we may be heading for a recession. This puts demand forecasting and sales forecasting in a new light. After all, the impact on the financial performance of food companies and others can be enormous. As a company, how can you better respond to this changing market? What role does data play to better predict your sales? And how can we use machine learning to properly predict demand, despite uncertainties?

Covid as an example of demand disruption

Everyone probably remembers, Feb. 27, 2020 the first corona contamination in the Netherlands. One corona report quickly followed another. No one could have imagined then how many challenges and uncertainties this pandemic would bring. Besides the fact that the pandemic had extreme consequences on a social level, companies also had no idea what to expect. A difficult time. Now we seem to have had the worst of it but are we ready for the future now that the pandemic is over?

According to a World Bank analysis, based on surveys of more than 120,000 companies in more than 60 countries, it was found that some companies’ sales skyrocketed, while others had to fight to keep from going bankrupt. Certain companies saw demand for their products increase 20% in the first few months and even 34% in the following months. On the other hand, the 25% most affected companies had sales drop by as much as 72% in the first few months and 50% in the months that followed. Of course, these differences are also due to the difference in how heavily countries and sectors were hit by COVID-19. Some countries were hit harder than others, and the impact on certain sectors was also greater than others. For example, the hospitality industry was virtually at a standstill, while supermarkets turned in top sales.

Are historical data still a reliable demand predictor?

Clearly, corona has had a big impact on historical sales data. How should you deal with this now? Whereas before you were pretty confident in predicting demand, now that may have changed completely. Suppose you expect, based on historical data from previous years, to sell 20% more in September than in August, then you can take that into account in the forecast. Based on this, you can optimize processes, targeted purchasing and capacity planning. But perhaps the pandemic caused consumers to exhibit different behavior, which is why sales were much higher in the past 2 years in the first place. Or the distribution of demand has changed a lot over the past few months.

The next question is whether this behavior has been permanently changed by the pandemic or whether the old demand pattern is returning. In any case, it is certain that you can no longer blindly rely on the old forecast of selling 20% more in September than in August. So what should your forecast look like in the coming years? Should you base the demand forecast on the years including the pandemic or should you remove the 2 pandemic years from your historical data? How do you deal with this as a company? And could it possibly also present opportunities?

The impact of information deficit on demand forecasting

The well-known Harvard Business Review pays attention to changing consumer behavior in an article. Businesses are facing an “information deficit,” or information shortage, due to recent changes. This occurs when insights derived from historical sales data are no longer useful due to an abrupt change in consumer buying behavior.

This sudden change means that consumer behavior is more difficult to interpret, predict and model. Demand forecasting is placed in a different light. That is exactly what happened when Covid-19 turned the whole world upside down. Companies can no longer assume that the data they have collected is going to yield good forecasts because of the (permanently?) changing consumer buying behavior.

Unique moment for catching up

Especially companies that in recent years have paid less attention to optimizing demand forecasting based on (historical) data can take advantage of this moment. Especially companies that are actually lagging behind data-mature competitors now have a unique opportunity to catch up. After all, data-mature competitors are also currently struggling with an information deficit.

The less data mature companies can now commit to a better data infrastructure and move toward data-driven decision-making. They now have an opportunity to catch up. Because everyone has the same questions right now; has consumer behavior changed permanently or is it going to change back? And what should I do with my sales data from the pandemic period? Do they give the right basis for insights? Or is this data actually causing noise?

Machine learning improves demand forecasting

Throwing away data is always a waste, because in the end it is information. You have to make sure that you can separate the standard patterns from the incidental effects. If you can include a feature in your data to indicate that something unusual happened, this data can actually be very valuable. In practical terms, you could think of an extra column in the data set that indicates whether there was a lockdown at that time. Then you can arrive at a useful forecast and even use this knowledge in possibly similar situations in the future.

For this, you can use machine learning. Machine learning is a form of artificial intelligence aimed at building systems that can learn from processed data or use data to perform better. The power of machine learning is recognizing underlying patterns.

Machine learning is also able to understand certain factors and the impact these factors have on the forecast. Thus, when you give historical data as input to your machine learning model, it will analyze the data, segment it and establish patterns that you can use to make data-driven decisions. Companies that smartly leverage these outcomes will always maintain a competitive advantage in the long run with their data-driven decisions.

Where to start?

Want to know more about how you should deal with the uncertainties in your demand forecasting? Then we can help you. Data Refinery Amsterdam has a lot of experience in this area and also has tooling to automate demand forecasting. Please feel free to contact Mathilde Aarnink for more information.