AI Done Right: Assortment Planning in the Age of COVID

Simulations can predict without historical examples
Photograph: Shutterstock

Analyzing historical sales data with artificial intelligence technologies is one method by which grocery leaders develop forecasts and try to predict future conditions. Obviously, the value of those predictions depends on how accurate and meaningful that historical data is, which presents a real quandary in the current and upcoming post-COVID era.

One could say, “Well, let's forecast for post-COVID from data gathered from before the lockdowns began.” That is one way to simply ignore that this “Black Swan” event pandemic ever happened. Probably not such a great idea in terms of staying in business, but certainly, that's one way to make all those complex predictive models work. It also assumes that the post-COVID environment will be very similar to the pre-COVID world, and I'm not sure anyone would agree that is a valid assumption.

Predictive analytics solutions guide decisions based on correlations uncovered in the data, while running simulations based on causation can evaluate billions of options in a way that humans can’t. In statistics, the well-known phrase “correlation does not imply causation” means observing association or correlation between variables doesn’t explain why something happened.

The retail AI community has hit a historical inflection point because of COVID. Analytics that provide scenario-based predictive models are of questionable value. The ability to use technology to simulate means grocery retailers no longer need a big longitudinal history. Today’s AI predicts without using historical examples.

Traditional Assortment Planning

Assortment planning is the term used to describe the process of figuring out what and how much inventory should be carried in a particular merchandise category, over the course of a specific period: daily, weekly, monthly, quarterly. The overall objective is pleasing shoppers and driving revenue.

One-hundred-percent-optimized assortment planning would mean a grocery retailer would be able to provide every shopper with all the products they want, in any and all combinations, when they want them, at a reasonable price point. A retailer with 50,000 SKUs must evaluate a mind-numbing 103,600 basket possibilities every month. There are 1,080 atoms in the universe. That means there is more product combos then there are grains of sand on planet Earth. Obviously, this is an impossible goal, but one that grocers increasingly will need to come close to. When assortment planning isn’t optimized, inventory costs increase, and shelf and warehouse space is wasted. Maybe grocery retailers can’t get an A+ in assortment planning, but they need to get at least a high B in order to stay viable.

AI-Powered Assortment Planning

Deploying assortment planning strategically to raise up the average value of each transaction is one thing. For example, stocking more suntan lotion in July and putting up an endcap with cranberry sauce, gravy, and foil baking trays the week prior to Thanksgiving. Or simply moving inventory around to accommodate new product offerings from suppliers and taking advantage of trade deals.

Quickly reacting to a completely unpredictable Black Swan event like COVID and answering the skyrocketing demand for toilet paper (due to 90% of office workers staying home for months) and baking ingredients and sanitization supplies get sold out is a whole other ballgame.

For example, the mix in consumer baskets in grocery has changed to reflect more cooking at home than pre-COVID. Categories like baking supplies, spices, and meal kits have seen significant increases—which are expected to remain elevated post-COVID.  Cleaning supplies, paper products and sanitizers are also elevated. Using AI-powered simulations enables grocer to quickly and accurately calibrate and adjust store space and assortment needs to reflect these changes. Post-COVID assortments will be different and many of the changes are likely permanent.


One of the more significant longer-term changes that is going to stay with the food industry is that retailers and brand manufacturers alike are going to really take a hard look at product assortment; a fresh eye towards product assortment, whether short-seasonal products or multiyear basic replenishment items.

Price competition, we believe, will be even more intense after COVID subsides than it was prior to the pandemic. The reality is, there’s still a war for customers going on, but it's been in large part hidden by the fact that sales have increased. Whatever percent of the $1 trillion in annual domestic food spend that grocery has taken from restaurants and foodservice ultimately remains within grocery after COVID, every grocery retailer will always need to vigorously defend against underpricing by competitors and Amazon.

When grocery retailers arm their merchandising team with AI tools to augment product assortment decisions, they’ll be able to find incremental gains they couldn’t before.

As the slowed economy drives customers to be more selective and price-conscious, retailers who rely heavily on historical date to drive assortment decisions, are headed for trouble. Where “let’s just do what we did last year” approach was ineffective pre-COVID, now its fatal.

Gary Saarenvirta is an aerospace engineer by trade and the founder and CEO of Daisy Intelligence.


More from our partners