The COVID-19 pandemic has disrupted everything from consumer behavior to supply chains, and the economic fallout is causing further changes. The data analytics field faces a complicated problem: how to use past data, and predict future behavior, in the face of uncertainty. Few organizations are facing business as usual or as expected.
“It’s hard to get good data about the future, so we have to use data from the past,” said Thomas Davenport, a professor at Babson University and fellow at the MIT Initiative on the Digital Economy. “And if the past is no longer a guide to the future, we’re going to have a tough time doing any sort of predictive analytics.”
In a webinar hosted by MIT Sloan Management Review, Davenport and Jeffrey Camm, a professor and associate dean of business analytics at Wake Forest University, talked about what they’ve learned from discussions with data and analytics leaders about how the pandemic is changing their industry. For starters, they said, companies are taking time to pause and assess what data is still relevant.
The professors outlined several other trends, including how the pandemic is hastening adoption of external data and turning data scientists into amateur epidemiologists, why companies shouldn’t throw out outlying data, and why data jobs might be more important than ever.
Firms are reverting to descriptive analytics and pausing machine learning
Organizations have gravitated toward predictive analytics in the last several years, as they use data to anticipate future trends and needs. But forecasting demand is difficult even in normal times, and the pandemic’s unpredictability has been challenging. Since the pandemic started, Davenport said, executives have sidelined predictive analytics programs and pivoted back toward simple descriptive analytics — good data about the present and recent past that is quickly available.
“The simplest predictive model is what happened yesterday,” Camm said. “That’s what we’re going to use to predict what’s going to happen today.”
Companies are also pausing machine learning programs that use existing data, taking the time to figure out what information is still relevant. In this respect, companies could use a quality control-type system, similar to what’s often used in manufacturing, for machine learning programs, Camm said. As errors start to accumulate, automated machine learning would stop.
Facing more external forces, companies are embracing external data
The last year has accelerated a trend toward using external data, Davenport and Camm said. With outside factors causing significant disruption and internal data about past activities no longer a good predictor of the future, companies are turning outside to figure out what's going on, particularly about consumer behavior and demand.
Davenport said this should be a lasting change. “One of the big problems that we've had with data and analytics in a lot of organizations is they've been too internally focused,” he said. “Switching to figuring out what's going on in the outside world, with people who aren't our customers — what people think about us, what's happening with our suppliers and their suppliers, and so on, are all good things to do.”
Companies need agility to easily incorporate external data, Camm added. Too much red tape or lengthy processes can delay using data from other sources, he said, though companies should also think carefully about the data being brought into their models to avoid bias, inaccuracy, or other problems.
Data professionals are behaving like epidemiologists
Analytics professionals have been asked to predict the impact of COVID-19 on the business, and “to do that, you have to predict what's going to happen with COVID-19, a typical activity of an epidemiologist,” Davenport said. This is complicated in the U.S. because centralized, complete COVID-19 data has been hard to find and information has been reported and organized in different ways, he said.
Publicly-available information, like a Johns Hopkins website that tracks COVID-19, has been especially important for people following the pandemic, the researchers said.
Some companies are looking for other economic indicators, like movement through ports and consumer confidence levels. Car companies have been looking at smog levels in certain cities as an indicator of how much driving is taking place, with more smog meaning more driving, and a hint that activity is returning to normal and people might be buying cars again, Davenport said. People raising livestock are looking for information about meatpacking plant closures, which means looking at county-level data where plants are located.
Companies are reaching for, or creating, disaster models
While the last year has likely produced a lot of outlying or unusual information — skyrocketing or disappearing demand, or a sudden increase in the number of people who cannot pay their mortgages — companies shouldn’t totally disregard data from the pandemic period. There will likely be other pandemics in the future, as well as disasters of other kinds, Davenport said. There have also been similar disruptive events in the past, like the 2008 economic crisis.
“I don't think you should throw that data away,” he said. “But you should put the data and the resulting models on the shelf and say, ‘Okay, when something bad like this happens, let's bring them out again.’” Walmart, which is known for its supply chain preparation in the cases of disasters, is already doing things like this.
Companies that have models for different scenarios, like mergers and acquisitions or events like hurricanes, should consider reaching for them. (And companies that don’t should consider creating them.) Some consumer goods companies have been reaching for hurricane planning models that have plans with variables like how broad reaching the hurricane is going to be and which distribution centers should receive extra goods. “That same kind of thing might be transferable to the pandemic,” Camm said.
Organizations aren’t, and shouldn’t be, giving up on data
After a drop in the early months of the pandemic, data from the company Burning Glass has showed that data analytics job postings have bounced back, Camm said. “As soon as the end of June, new job postings in analytics were sort of back up to where they were before the pandemic drop occurred,” he said. “And it seems to have stabilized … I was shocked at how quickly new job ads bounced back very fast in early summer.”
Davenport said data professionals in some industries, like anything related to travel or entertainment, were hit hard by the pandemic, but hiring in industries that are prospering has made up for the downturns.
Camm and Davenport said a company’s likelihood of strengthening its data analytics program despite the recession will likely depend on whether a company has already seen a return on investment in their analytics programs.
Analytics is still a good bet, the researchers said, and when things get tough, demand for analytics can increase. “When the company goes into cost cutting mode, they want to do that in a smart way,” Camm said. “I would say in these disruptive times analytics and data science probably are stable, or maybe their demand even increases a little bit.”