Follow Datanami:
May 4, 2020

Detecting Consumer Signals in the 90% Economy


As COVID-19 lockdowns are lifted across the United States, consumers will venture out of their homes and begin to spend money again. But the new buying patterns in the 90% economy are likely to look dramatically different. Will machine learning-based forecasting methods still work?

Before COVID-19, companies in the retail and consumer goods sectors were adopting machine learning at a healthy rate. That’s because AI gives them powerful tools to use data to detect what customers want and predict what they’ll buy, increasingly at the individual level. And with a better demand signal, consumer-facing businesses can better match supply to it, which helps reduce costs.

Then COVID-19 arrived, and it changed everything. We’ve seen the results play out in real time, as non-essential stores are shuttered, certain items fly off grocery-store shelves, and consumers flock to e-commerce sites like, which hired 80,000 additional workers to handle the surge.

During May, shelter-at-home orders will expire and businesses will re-open. As that happens, American companies will be looking to data and AI to help them guide decision-making. But nobody is sure how customer sentiment will play out in a public environment where social distancing and facial coverings are still required, fear of a second COVID-19 wave are extant, and one in five people are out of work.

It will be a challenge for businesses to decipher the economic signals created by the COVID-19 pandemic, says Sankar (SN) Narayanan, chief practice officer at Fractal Analytics, one of the biggest AI companies you’ve never heard of.

“Almost all of the forecasting work that everybody has done until now does not really stand because of this important new data point that has emerged,” Narayanan tells Datanami in an interview. “In such a scenario, when you don’t have the luxury of using past data to forecast, what can you do? What sort of algorithmic approaches can we adopt to forecast better?”

Will shoppers return when COVID-19 lockdowns are lifted?(RockerStocker/Shutterstock)

Fractal Analytics is one of many AI software and services companies looking to help its clients cope with the new normal. The company has clients in retail, consumer packaged goods, insurance, healthcare, life sciences, technology/telecom and financial services, and is engaging in COVID-19-related work with many of them.

“What we’re seeing is that different industries are experiencing different effects because of the pandemic,” Narayanan says. “ Some industries, such as retail and consumer goods, are going through a significant change or shift in consumer preference or behavior, which is leading to considerable volatility in how consumers [shop] and therefore there’s a lot of work going on there.”

China is a couple of months ahead of the U.S. on the recovery, and according to The Economist, some things appear to have returned to normal. “Factories are busy and the streets are no longer empty,” the magazine wrote recently. “It is better than a severe lockdown, but it is far from normal.”

This is what “the 90% economy” looks like, The Economist says. “Discretionary consumer spending, on such things as restaurants, has fallen by 40% and hotel stays are a third of normal. People are weighed down by financial hardship and the fear of a second wave of COVID-19.”

Back in the United States, consumers are set to make significant changes in their buying behaviors as the COVID-19 lockdown eases, according to IBM, which released the results of its Institute for Business Value survey last week.

The IBV survey found that many consumers will forgo ridesharing and public transportation, and will avoid large crowds, with 75% saying they will not attend a large event or trade show this year. Consumers are more likely to use contactless payment options if offered, and are less likely to buy a new car due to personal financial concerns. They’re also more likely to shop locally, although most say they will still eat in restaurants and bars.

Will COVID-19 permanently alter consumer behavior? (GoodStudio/Shutterstock)

As business executives look to chart a course through the COVID-19 pandemic, this type of data will be critical. Bankruptcies will climb in the months to come as some companies either fail to align their business strategies to the 90% economy or simply succumb to the new reality, like clothing retailer J Crew, which today announced it was bankrupt.

“The [IBV] study provides further evidence that COVID-19 is permanently altering U.S. consumer behavior,” Jesus Mantas, a senior managing partner with IBM Services, stated in a press release. “There are long term implications of the new consumer behaviors for industries like retail, transportation, and travel, among others. These organizations need to quickly adapt their business models to serve the new consumer behaviors in order to survive and thrive.”

The big question is what role AI will play in helping companies to adapt to the new normal of a 90% economy. According to Ryohei Fujimaki, CEO and founder of dotData, the machine learning algorithms will begin giving good data-driven answers again, but it will take time.

“Some of our customers are asking us, how reliable is the model they built six months ago?” Fujimaki tells Datanami. “In this situation, no one can really capture the actual trend, no matter if it’s AI, machine learning, or human intuition. It’s not reasonable to say that a single method can capture the real trend. So people should expect AI and machine learning models to become less accurate at this stage.”

Normally, the more real-world data that a company has to train an AI model, the more accurate the models’ predictions become. But in the case of COVID-19, that’s not the case. Fujimaki recommends that companies pay close attention to the performance of their AI model. If the model become less accurate, then they should retrain it using more current data.

Model maintenance is more important than ever, and customers should increase their use of AB testing and champion-challenger tournaments to help them detect new consumer behavior signals as they emerge, Fujimaki says.

“We recommend having multiple machine learning models based on different types of features with similar accuracy,” he says. “We don’t know which features are really stable and which features are unstable, so we need to prepare a few different models and then run these models in parallel, so they can see which model performs more stably.”

AI will automatically detect changes in consumer sentiment in the long run, and the models will adjust accordingly. But in the short-term, the AI-generated answers may not be as accurate as they could be, so companies should take precautions to ensure they’re using the data appropriately.

Related Items:

How AI Is Helping the Supply Chain Cope With COVID-19v

How the Lack of Good Data Is Hampering the COVID-19 Response

New MIT Analytics Tools Predict COVID-19 Patient Outcomes and More