Predicting unpredictable consumer demand: Forecasting in times of crisis

How can you predict what products consumers will want tomorrow? In standard times, retailers can rely on sophisticated algorithms that use historical sales to model a projection of demand. But when extreme events happen, sales patterns become erratic and unusual. Today, in the middle of the Covid-19 crisis, retailers are left with a few months’ worth of biased or inapplicable sales histories which they can’t use to predict future sales, and with a situation still in flux, with further limitation and lockdowns looming ahead.

As black swans – sudden, highly disruptive events – are hard to predict and to prepare for, retailers need to find strategies to create accurate projections of demand while accounting for periods of irregular sales patterns.

Combining artificial intelligence and human insights

Automated demand forecasting tools are extremely valuable, as long as you have a reliable sales history to work with. “When demand is unstable, you can’t rely on automations and historical data alone. To achieve a dependable projection of demand, you need to combine technological and human insights,” says Martin Kleindl, who heads the development of replenishment and supply chain solutions at software development firm LS Retail.

Here are three ways retailers can create reliable forecasts and optimize stock coverage during the Covid-19 disruption.

Predicting unpredictable consumer demand: Forecasting in times of crisis

1. Fix biased histories

For most retailers the past few months have showed extremely atypical sales patterns, which are unlikely to repeat even as the Covid-19 crisis progresses. Since machines can’t (yet) read the papers and realize that recent spikes in sales of pajamas bottoms and bralettes will probably decline as people go back to the office, retailers need to manually adjust biased histories.

“I always advise selecting a demand planning software that also includes tools to automatically detect extreme statistical outliers, and which uses this information to adjust the sales history. This level of automation will simplify and speed up work tremendously. At the same time, the ideal software solution must also give retailers the ability to do manual adjustments and fix abnormal histories so they are not considered during modeling. The most accurate predictions come from combining the machine’s analytics capabilities with the retailer’s knowledge of external factors,” says Kleindl.

2. Improve predictions

With both consumer behavior and regulations changing rapidly, neither recent sales patterns nor historical sales trends can be used as a basis for a trustworthy forecast. During times of disruption, the retailer’s sales expectations, based on their knowledge of future events – the delay in back-to-school plans; the cancellation of weddings or of Halloween celebrations – are tremendously valuable, and should be used to model and adjust predictions.

“Retailers need to be able to manually adjust for specific product groups, based on their expectations of future fluctuations. Then, they can use the software for the heavy lifting. The system can take care of shaping a curve based on trend expectations, and creating a forecast that factors in the correct amount of items and variants,” says Kleindl.

3. Redistribute items across the chain

As both regional and localized lockdowns are continuously eased and reintroduced, retailers should prepare for more forced store closures, which may occur with little to no advance notice. There’s a real risk of having valuable stock stuck in locked stores, which makes it both imperative and urgent to secure tools to redistribute items easily and flexibly across the entire retail chain.

“The ideal replenishment software solution gives you the freedom to propose where items should be placed based on rules, so you can use your knowledge of external factors, such as which stores will be open and doing business, to optimize distribution,” says Kleindl. “The best software will also make your job easier by automatically calculating the time and costs of the different redistribution plans, so you can easily devise the most cost-effective strategy.”


Related news