Store replenishment: How AI can address extraordinary events

Artificial intelligence goes beyond normal forecasting models when it comes to highly complex events that impact buyer psychology.

The panic that currently has many global consumers emptying food shelves has shattered all demand forecasts and led to severe issues with store replenishment. Of course, this is a situation that we’ve never faced before, but still reminds us – in the extreme – of how much demand depends on an infinite number of factors, some of which are particularly difficult to take into account.

An unprecedented out-of-stock situation

In the current Covid-19 crisis, completely ordinary products are out of stock in the face of unprecedented demand. Around the world, customers enter supermarkets and flock to the same shelves (e.g. pasta, rice, beans, hygiene and cleaning products). And if these shelves are not already empty, customers make excessive purchases. In pharmacies, Tylenol  evaporates, not to mention safety masks and hand sanitizer, likely exhausted since the first days of the crisis.

These unusual purchases are often called “panic buying.” But is it just simply irrationality or something more?

There are likely explanations. As I live in France, I’ll remind readers of the long lines at petrol stations at the end of 2019 resulting from strikes around the pension reform project. No shortage was expected, no blocking of fuel storage depots was announced. Still, consumers behaved as if there had been such announcement. Irrationality? In reality, it is the memory of past experiences (in particular the blocking of refineries in May 2016) and the projection of those historical events onto the immediate future that pushes consumers to anticipate a future of shortage, even if it means creating shortages themselves, in a sort of self-fulfilling prophecy.

The current explosion in demand for certain products is based on the same psychological dynamic.

And since we live in an era where various types of crises are more and more frequent, it is  not so surprising – or irrational – that consumers adopt these new kinds of behaviors. What is believed to be irrationality can be easily explained by an unusual form of foresight in the face of a crisis whose outcome clear.

New call-to-action

New forecast models

Forecasting demand is a complex science. Today, it is even more complex considering new factors which it must take into account, some of which are particularly difficult to grasp, like the Covid-19 pandemic. Certain forecast calculations must now consider psychological factors that trigger highly unusual behaviors among consumers. In addition, AI can use exogeneous data to identify a particular “event.” This can then be used to clean the data, but the model also incorporates the event so it can be used for the future.

Of course, for such an unpredictable event as the Covid-19 pandemic, forecasting fell short at the onset, as there was no warning and no historical data to even approximate an appropriate response. The only recourse was true stock visibility – knowing what you had, where it was and where you could get more was the best weapon to get through the initial “triage stage.” Now as we move forward, we have to record all of the data and contextual events (like government announcements that trigger surges in shopping) to feed the AI. It will then treat the event as an anomaly but will also be able to respond quickly if something like this does happen again (or a similar event like a natural disaster) and will know the types of products that sell out under such circumstances and what triggered the surges in demand.

As scientists we look to ways to infuse our offering with new and different techniques to help retailers. For example, for a retailer considering the start of a seasonal item like snow removal products or soup, AI can learn a signal from the item to detect when it would be best to start the seasonal forecast. It will also consider exogeneous data: weather, celebrations, etc. to increase the accuracy for this forecast.

AI keeps things in context, even when humans don’t

None of these calculations can be realistically performed by humans alone. The situation is changing too quickly, and crises can be very different in nature. But AI can learn. It learns which contextual data should be taken into account and which should be ignored, which correlations are valid, and which are not. It is not a crystal ball – retailers and manufacturers will still make mistakes – but it does consider factors and changes we could not have imagined. And, faced with the coronavirus and future breaches of normality, AI will be invaluable in anticipating what consumers will do next and what their next “panic purchases” will be.

Discover more critical COVID-19 insights and recommendations on our COVID Insights Hub

Learn more about what to look for in optimal demand-planning technology in the Gartner Market Guide for Retail Forecasting and Replenishment Solutions.