Before the outbreak of COVID-19, demand planning hardly ever matched actual sales. But since the virus reached Europe, the level of sales in different product categories became even more difficult to predict.

How could the garden centers have imagined that consumers would invest their money in a roof, pavement or new garden furniture? How could bakeries have predicted that pastry sales would plummet by up to 80% because there would be no more birthday parties or weddings? This volatile demand led to, among other things, empty toilet paper shelves in supermarkets and packed shelves of clothing for spring that were not sold.

Consumer behavior has changed. Of course, channels shifted with the exploding growth of e-commerce, but even when people went to a store, they made different choices. Local retail organizations struggled for a time, for example, as consumers preferred to do all their shopping at once at the supermarket.

Supply chains not recovered yet

On top of rapidly changing consumer behavior, organizations had delivery problems. Containers were at locations around the world where they were not needed. Factories and ports in Asia, in particular, went into lockdown with just one infected employee. And if they didn't shut down because of coronavirus contamination, they did get a "power cut" because energy production couldn't keep up with demand for energy when it turned out that the global recession wasn't that bad.

There were other factors involved, some of which still play a role: political instability (between China and Taiwan with the United States in the background), the Suez Canal being blocked, and a shortage of raw materials and labor.

Nevertheless, this fear of a deep recession forced many companies at the beginning of the pandemic to adjust their demand planning. As a result, suppliers downsized and put building plans on hold for additional factories. When it became clear that consumers were investing their money massively in long-lasting consumer goods, companies all turned to their suppliers at the same time. This was the main reason for the global shortage of computer chips, to name just one example.

Beware of bullwhip effect, improve your demand planning

This shortage of computer chips is a good example of the so-called bullwhip effect that happens when demand planning is inaccurate. If demand is higher than predicted, delivery problems occur. If demand is lower than predicted, warehouses are filled to capacity, and that is very expensive. Both situations are undesirable. But the risk of these scenarios is looming in today's VUCA world. The four letters in the acronym come from the English terms: volatile (rapidly changing), uncertain, complex and ambiguous.

Integrate open data into your ML models

All the more reason to improve your demand forecasting. If not for increasing your customer service levels, then at least for reducing your supply chain costs. The question is, how on earth can you make predictions in such an uncertain world? In any case, the old method no longer works because your historical data no longer has any predictive value for the future. You will have to include more data in your predictions.

The good news is that a lot of data is also available as open data, such as the R-factor, the unemployment figures or consumer confidence. You can even import this data into your predictive models fully automatically.

And the other good news is that today's AI and machine learning (ML) models are so advanced that they can look for the correlations between the data and the effect on the sales figures by themselves. The model can then automatically make recommendations. This data will help you improve the quality of your demand forecast, but it doesn't have any predictive value.

Another good example is a proof of concept (PoC) that SAS did at a well-known CPG organization. The demand planners were used to manually upsizing or downsizing the demand forecast provided by their system because they knew that certain developments had not been taken into account in the forecasts. By enriching the prediction model with more variables, we showed that we could improve the accuracy of the demand prediction.

Based on this PoC, we further developed our software, which now includes an Assisted Demand Planning module. On one hand, the demand planner can identify how likely it is that a new modification will indeed make more accurate forecasts. On the other hand, the software recommends which product categories are best to spend time and energy on.

Optimize your supply chain with analytics

The great benefit of better demand planning is that you can reduce or even avoid the bullwhip effect further down your supply chain. But data and analytics can also help you optimize your logistics network. You can use analytics to:

• Calculate where to build production capacity.
• Decide where to locate DCs.
• Determine how often to supply customers.
• And other similar supply chain issues.

Combine domain knowledge and data science

These all seem like nice promises. But many companies are far from getting the desired return from their investments in analytics. And yes, unfortunately, we see this happen all too often. In many cases, this is because analytics is still handled by the IT department. People are trained as data scientists, and they get to work without specific domain knowledge. They create nice dashboards. But in practice, these hardly seem to match the real need. On the other hand, the business managers do not succeed in getting the business needs across to the data scientists. The result is a mismatch and the feeling that the investments in software and expertise are not yielding sufficient returns.

If you want to reap the benefits of deploying data and analytics, make sure it becomes a joint project between the business and IT. Because without the proper input of domain knowledge, all those beautiful algorithms will fail.