Companies should be adjusting their data and analytics strategies to better align with market realities as they unfold. Up until a few weeks ago, it was relatively clear that companies needed to become increasingly digital to thrive in an era of rampant industry disruption.
Then, the COVID-19 pandemic hit. Regardless of whether businesses have been shut down or they're operating above or below their normal capacity, every company's data and analytics strategy has been impacted because the underlying data has changed. Customer behavior has changed, supply chain behavior has changed, company operations have changed. If your data and analytics strategy isn't keeping up with what's happening, then you have important work to do, quickly.
Predictive analytics took a hit
Predictive analytics depends on historical information, which doesn't exist for the COVID-19 outbreak. Sure, 9/11 had both short-term and long-term impacts on consumer behavior. Just after the attack, people avoided public places. To this day, airport screening is not what it was before the attacks. The Spanish Flu shares more similarities with COVID-19 than 9/11 because it was a global pandemic but, there's little information about it.
"Data scientists like to talk about the concept of data drift, and typically that happens over time," said Brandon Purcell, principal analyst at Forrester. "That process just accelerated and now companies have to start collecting new data and creating new models based on the data from the point when folks started sheltering in place."
It's important to track how customer behavior is changing because it will continue to shift, perhaps radically, depending on several factors such as when the executive orders expire and whether those customers still have jobs or not. For example, since customers can't shop in some physical stores at the time of this writing, they're shopping online. However, when the shelter in place orders expire will they go back to shopping at brick and mortar stores like they did before the pandemic or will they choose to shop online more often?
Instead of relying on predictive models, Purcell said it's important to do descriptive analyses of customer journeys and the volume of customers going on different journeys. Pay close attention to whether a journey is functioning properly and if not, fix it quickly.
Understanding the voice of the customer is always important for brands, but in the present situation, businesses should strive to understand how their customers are feeling as the impact of the pandemic unfolds, which not all businesses are doing well. For example, some businesses are executing email campaigns with the same messaging they used before the pandemic hit. What those messages fail to acknowledge is that customers' states of mind have changed, and their buying power may have changed. If your company has a call center, text and speech analytics can help you better understand how customers feel so you can adjust messaging, strategy and individual customer interactions accordingly.
Purcell also advises companies to pay attention to metrics like customer lifetime value versus shorter-term metrics such as conversions and clicks. As is evident, customer behavior is and will remain more volatile at least in the short term compared to what it was just a couple of months ago.
"A lot of people have lost their jobs and filed for unemployment. Those people are cutting non-essential services, so all of a sudden, your churn rate is going to go way up. Any churn models you built before are looking for signals in a more normal period," said Purcell. "People are going to churn because they can't pay."
Big data has holes
Businesses have collected a lot of data on customers and their own internal operations, but the patterns of just a few weeks ago don't reflect what's happening now. Erick Brethenoux, VP analyst at Gartner said companies shouldn't overlook small data techniques.
"Because we have cloud and GPUs, people forgot there's a lot you can do with small data, so small data is coming to the forefront with a vengeance," said Brethenoux.
He also said knowledge graphs are making a comeback because they capture relationships in addition to facts.
"You don't need to run 5,000 iterations to find out that two things don't correlate to each other. You know with one or two that logically they correlate to each other for whatever reasons. Graph databases do that well," said Brethenoux. "Then, as you learn, you gather data, which helps you get better."
More generally, flexibility and adaptability are critical because as recent history has shown, supply, demand, and human behavior can change quickly and dramatically. Brethenoux said a system of interchangeable components can help.
He also said that people need to realize that machine learning isn't the right technique to solve every type of problem. It should be associated with other techniques such as rules-based systems, optimization techniques, and graph techniques so organizations can get to production faster in a more accurate manner.
"Going to production is the main issue except that you don't have three months now," said Brethenoux. "You need to stop the bleeding without knowing what's coming next."
Your own data isn't enough
The pandemic has given rise to an unprecedented level of uncertainty. Since organizations lack some of the data, they need to figure out what they should be doing next, they should consider outside datasets.
"This crisis is drawing attention to the lack of external data that organizations have available to them in a consumable way. [Third-party data] can be used for input into the forecasting models to help them forecast, not just using their own ERP POS data like they always have and the historical data, but also looking at the external and exogenous kind of data and signals that are absolutely necessary in the kind of space we're in right now," said Traci Gusher, principal of Innovation and Enterprise Solutions, Data & Analytics at KPMG.
For example, tracking pandemic data from other countries helps predict the impact of COVID-19 in the US with some level of probability. In addition, US data can be fed into models for inferencing purposes and to identify correlations.
"There's some really great datasets being published that can help you improve these models, everything from Johns Hopkins' data [to] the social mobility indexes being published by Google," said Gusher.
Like Gartner's Brethenoux, Gusher underscored the need for agility. Specifically, she said one of the things organizations have lacked is dynamic planning capabilities.
"New data is becoming available every day, new scenarios are changing every day, policies are changing every day. Every one of those changes impacts the way that organizations are able to deal with this crisis, so the ability to have that agility in planning processes, forecasting processes, modeling processes is key to balancing speed and accuracy."
Your data pipeline may be incomplete
Businesses are discovering that they're ill prepared to deal with present circumstances because their data pipeline is incomplete. They lack data or the data they have is unreliable.
Up until recently, it might have been fine to build a data pipeline one section at a time using different data engineers for data connections, data accumulation, master data management, data enrichment, and data packaged for consumption. However, given the current state of rapid change, the need for speed will cause organizations to automate what was previously done manually using intelligent technologies.
"People realize now that should you invest more in speed and nimbleness, you will lose less," said Igor Ikonnikov, research advisor at Info-Tech Research Group. "You have to have your tools in a fully automated manner figure out what are your integratable entities, how you can crystallize them out of your transactional data, how you can build your data muscle optimized for consumption data structures, how you can build a knowledge graph for deep inferencing, how you can build multidimensional if/then cube scenarios, and all done quickly. You don't have weeks or months anymore."
Circumstances have changed radically, and they're going to continue to change, often, over the coming weeks and months. In response, data teams should endeavor to become more agile so they can adapt their data and analytics strategy faster and easier.
For more on executive strategy read: