Photo: Getty Images/ronstik
How murky has COVID-19 made retail data?
COVID-19 has made past purchasing data less relevant, impacting everything from loyalty program strategies to AI-driven product recommendations, personalized emails and merchandising decisions.
A few columns and blogs have explored the challenges marketers are having extrapolating pre-pandemic shopper intelligence to assess purchasing patterns as consumers have sheltered at home. Purchasing behavior amid the pandemic has also been too erratic to offer much future insights.
“In our post-stay-at-home reality, companies need to recognize that their existing predictive models, forecasts, and dashboards may all be unreliable, or even obsolete, and that their analytic tools need recalibrating,” Angel Evan, president at Particle Inc., a data analytics firm, wrote in a column in Harvard Business Review. “Although the objectives of a specific automated system or predictive model may not have changed, the input data and the users certainly have, which should cause companies to re-evaluate how outputs are interpreted and relied upon.”
Bryan Pearson, former CEO of LoyaltyOne and currently a strategic advisor, wrote in a Forbes column, “COVID-19 is causing consumers to change their once-predictable purchasing behaviors dramatically, and in doing so it’s undermining the traditional statistical modeling retailers use to help frame their inventories, merchandising and marketing.”
Will Douglas Heaven, senior editor for AI at MIT Technology Review, wrote, “What’s clear is that the pandemic has revealed how intertwined our lives are with AI, exposing a delicate codependence in which changes to our behavior change how AI works, and changes to how AI works change our behavior.”
Gary Saarenvirta, founder and CEO at Daisy Intelligence, an AI platform, wrote in a blog entry, “COVID-19 has broken customer shopping habits and historical trends. Retailers will not be able to rely on historical trends to understand the current landscape and adapt to ongoing change. Predictive modeling, by extension, will become less and less relevant. However, successful retailers will still need to invest in the ability to scenario plan with agility in order to run their businesses profitably.”
- Retailers Face a Data Deficit in the Wake of the Pandemic – Harvard Business Review
- Is Covid-19 Stymying Retail’s Intelligence – And Making AI More Critical Than Ever?- Forbes
- COVID-19: The End of Predictive Analytics in Retail – Daisy Intelligence
- Our weird behavior during the pandemic is messing with AI models – MIT Technology Review
- Has the pandemic changed shopping behaviors forever? – RetailWire
Discussion Questions
DISCUSSION QUESTIONS: What challenges have sudden behavior changes caused by COVID-19 put on leveraging shopper data? Can marketers correct accordingly? Do you see long-term implications?
Poll
BrainTrust
Carol Spieckerman
President, Spieckerman Retail
Dan Frechtling
CEO, Boltive
As the CEO of a retail analytics company, I can attest to the erratic patterns we are seeing in store traffic and conversion trends. Historical trends are out the window, and now the key is to remain flexible and monitor trends by day and even by hour. The key consequence of all this is that the interpretation of the data has become significantly more nuanced and complicated, so on top of everything else retailers are dealing with, they also now have to be even more careful with how they use and apply insights from their data.
The only thing predictable about consumer behavior during the pandemic is the unpredictable nature of it. However, harnessing the power of consumer shopping behavior, coupled with third-party market insights remains a valuable part of merchandising, assortment planning, pricing, promotion strategies, and execution.
The normal forecasting and buying models have to take into account the blips of unpredictability we are experiencing. What may have been perceived as an outlier has become the norm for the moment, with consumers leveraging digital commerce channels, spending less time in-store, and making very value-driven shopping decisions. It has become increasingly critical for retailers to have a holistic view of their consumers across all shopping channels.
Agility, flexibility and adaptability, powered by valuable consumer insights, will help win in today’s environment.
Depends on how long-term we’re talking. Three to five years: we’re looking at similar buying trends. Five+ years: It’s too early to tell. With Americans making decisions that will prolong the impact of the pandemic, current behavior changes could remain the norm for the foreseeable future.
COVID-19 has made all the predictive models obsolete – that much is obvious. But what’s less apparent is the long-term impact. It will be months or more before shopping behaviors stabilize, which means that all the data being collected now is of limited value. The status quo is out as an option, and there is no new normal yet. AI models themselves will likely need recalibration. The one upside is that COVID-19 resets everything to zero, so retailers that have stayed away from analytics and modeling have time to get into the game – if they act quickly.
The COVID-19 pandemic is an event much like a hurricane in respect to how it’s impacting retail; there have been severe commodity market cost changes, etc., but the difference is that this will last a lot longer than a hurricane or other natural disasters. Retailers (and the software vendors who depend on this data to perform analytics) will need take into account this period when planning next years budgets, sales and labor projections.
So does this mean we are on the cusp of a return to the merchant who knows what sells and relying less on AI? I thought the whole point of AI was unmatched prediction-making ability.
Any model that relies on data patterns over time is presently useless. The big question for marketing research is, what will shopping patterns be like when everything is back to normal (if that ever happens)? Predictions run the gamut from “no differences from pre-pandemic shopping” to “everything is totally different.” I favor the former, which suggests that while purchase forecasting models may not be so useful, snapshot-based models will return.
If nothing else, the data sets that result from COVID-19 will be invaluable for potential future calamities. COVID-19 may seem like an anomaly as we’re living it now, but it also serves as a cautionary case study for any future events.
True. This data should not be discarded or ignored. We cannot pretend this did not happen, because this is not a blip like hurricane that came and went in few days.
Data is only good when it is helping answer the questions key stakeholders are asking. The question is this – are the business leaders asking the data science teams questions that will lead the company towards the future? Or are they just asking the same questions as pre-COVID-19? If they are asking the same questions, the data is basically moot. However if they are asking questions about what the future of the company looks like, the data that can be culled from the past few months could actually be quite useful.
I wrote a short article in January of this year about “probabilities and possibilities.” It was before I had heard the first mention of COVID-19. A mere six months ago there was a world of pretty comfortable probabilities that would give the business some level of predictability. Last year and last week data gave some pretty good bedrock information. Then you could crank in current market trend dynamics and project (guess) at some possibilities that made sense for your business. Now in mid-2020, many of the probabilities of the business are gone. Even for basic product that might not experience much design change, what’s the demand going to be? How much of x, y and z did I already buy for the holiday season? I should probably reduce those quantities, but by how much? Spring 2021…? Total coin toss. And now it looks like I have to make different assumptions based on region. I can use one set of assumptions for the Northeast, and a whole different set of assumptions for the South. A resilient and responsive and flexible supply chain has never been more important.
The forecasting models from prior seasons are pretty much useless. So are year-over-year comparisons. There are demand forecasting techniques that – use recent months’ data, forecast, compare actuals, and refine. This requires more business intuition, assumptions and factoring in supply constraints all at once.
Even then, it is very likely to overshoot on some and have stock outs on others. To overcome this, businesses have to actively think of substitutes to offer and promote.
I tell my students that the three most important words in marketing are why, why and why?
Never come to a conclusion without asking why you came to that conclusion. Always look for the “why’s.” That is where the answers are. They are never on the surface.
Data should be looked at in exactly the same way. It is never reporting of a single point of time, it is always reporting of change, or no change. But it is not just the change or no change, it is the why?
The answer for the retailers and the marketers is not just what behaviors the new data indicates, but why it changed. The answer is not simply COVID-19 and what consumer were “forced” to do. The answer lies in why they actually did it. Only after understanding that will one be able to predict with some accuracy what the future holds.
For example, did the COVID-19 response surface brand new behaviors or surface latent behaviors that were perculating to the forefront? Or was something completely different happening that we didn’t consider at all?
it’s a double whammy — the predictability of shopper behavior is significantly dented, but so is the predictability of supply. Retailers are going to have to relearn how to fly by the seat of their pants, you know like in the “good old days” of retail. Not so good, is it …
The true COVID-19 effect will not be known until the customers who are waiting to return goods post-COVID-19 flood retailers. Long term, so many variables are in play. Schools opening? Vaccine? Work from home? The list is long. Marketers are going to need to do a deep dive into the data to make sense of new shopper behaviors. Socio-marketing seems in order.
The challenge is that pre-COVID-19 customer insights are of seriously limited value due to the disruptions of shopping habits, the formation of new ones and the new offerings and changes made by retailers and other brands. This all poses new challenges for organizations on top of what largely challenged them pre-COVID-19.
On a recent blog post of ours, we shared that “In our work, we see companies with more data than actionable insights. But we also see smart clients embracing the reality that things have changed and recognizing their pre-COVID insights are increasingly irrelevant. They’ve already reset their management dashboards to gather benchmarking data and analysis to see how far they’ve fallen and what customer segments have emerged due to the ‘new normal’.”
Count analytics and insights as just another one of the challenges that retailers face in addition to everything else!
COVID-19 has not been around long enough to truly have a formidable impact on retail data. COVID-19 is a five-month phenomenon, not even two quarters and certainly not a full retail year.
Reacting to COVID-19 after five months is premature and ignores all of our current, historical retail data. We cannot be so reactive in this dynamic, online retail environment, simply because there is a new force that changes our very short term perception of the omnichannel and online environment.
We can accommodate the current changing environment, recognizing that it is a transition and flexibility is critical.
Along with the challenges mentioned herein so far, we have to realize that we’re not out of the woods in this crisis as of today. This will easily impact retail in an anomalous manner for the balance of the year, throughout the holiday season. However that does not mean bad news for all retail. Food stores, big-ticket electronics, appliances, recreational vehicles, bicycles/leisure and other categories are seeing huge spikes this year. Over time, we’ll all see the painfully obvious data sets that need to be gut-checked.
When consumers clear shelves of products, the brand has no real understanding of actual demand in the supply chain. Did their products sell out because they were desired, or because they were the only option left?
The problem for consumer goods companies is that brand and product affinity goes out of the window during a panic buy or out-of-stock scenario. When there’s only one brand of toilet paper (or soda) on the shelf, consumers are forced to grab whatever is available.
Quotes pulled from this piece: “Pepsi Reveals the Future of Direct to Consumerism.“
Despite many who smartly and accurately speak about broken data models, I will take a step outside of the general consensus and suggest that data has become even more important now than ever before.
Consumer data still drives customer needs – and real time data is the most valuable. Companies like Amazon process over 29 million transactions daily (peak). Walmart is making $35 million per hour. The retail engine provides us with plentiful and reliable data about what’s next and predictive patterns to adjust and adapt.
The length of the pandemic has also been long – with notable points of inflection such as phase two openings of non-essential retail. There are enormous amounts of data that would be relevant and critical to predictive modeling. More important, the uncertainty of retailers means that we need to rely more on the most recent data trends.
Lastly, data models can be changed and tested. There’s no reason to believe an adjustable data model that has inherent assumptions are locked in place. Accuracy will be maintained as more data is added. AI is designed to adjust with changes in data trends. The best AI tools will provide relevant data and assumptions can be checked.
Long term implications are that with abundant data, which is still quite available, our data scientists will be able to leverage data for marketing and retailing purposes.
It has absolutely rendered the process more challenging but not impossible. There are many trends and buying patterns, etc. that are erratic through the pandemic and will continue to be, but aligning your retail store shopper data together with your e-commerce data and trying to understand attributes can help guide your planning. No one has a crystal ball and that is why it is critical that retailers and manufacturers use this time of uncertainty to truly collaborate, share data and work together in order to emerge stronger!
Historical retail data is far less useful today than it was before COVID-19. Consumer habits have been affected by externalities that seem likely to persist for the indefinite future. I believe that comparative competitive data becomes more important than ever, allowing retailers and brands to compare their performance against those like them. When assessing the performance of a retailer’s management team during the pandemic, market share rather than performance against plan is the only reasonable way to determine whether the right choices were made.
COVID-19 has thrown a monkey wrench in the predictive capability of pre-pandemic data. Customer shopping behaviors have been changed by an external force, so existing algorithms and machine learning are a bit less accurate. That impact is even more dramatic when you consider AI applications which rely even more on a wealth of data and some level of predictability.
The impact of the pandemic creates the need (and opportunity) to leverage purchase history data in new ways. The challenge is to dive into the dynamics of recent behavioral change to gain insights into how to compete and how to better serve your customers. The best way to be relevant right now is to help the customer adapt to the new normal. Looking at each customer’s behavioral change as a result of the pandemic can reveal insights into how they are responding and how the retailer can best meet their current needs.
The traditional models are essentially broken at this point. Relying on LY and LLY just won’t cut it going forward. Many of the currently closed and/or severely impacted locations will need to be treated like new stores. The reverse of that is true for those businesses like grocery who have had a big increase in sales. They will need to temper that surge next year when their historical movement comes into play in their models.
The big gains in eCommerce sales will also need to factored in. I believe we have increased the online sales adoption rate by 3-5 years in some segments of retail like which will have long lasting implications and hasten adoption of solutions like micro-fulfillment in semi-dark or dark stores.
Some data, like panic buying at the outset of the pandemic, is anomalous and can be ignored. Other data, such as the more dramatic behavioral changes among Millennials and Gen Z than among Boomers, can be segmented. But the most reliable data comes from identifying trends that have accelerated. Three years of transformation happened in three months, as Accenture’s CEO has pointed out. This includes online-to-store fulfillment, home delivery, and subscription services. Data in these areas is more robust and predictive on the path to a post-COVID economy.
Beyond behaviors, the challenge is “consumer emotions”. COVID-19 has created emotions that have created positive and some negative behaviors. Case in point: the paper industry had an extremely sufficient supply chain that balanced supply with demand. The demand variability was low allowing the industry to create a “pull” supply chain based upon a very predictable demand signal. COVID-19 created an emotional state of panic for toilet paper.
Can you simulate the impact of these emotional buying patterns? The answer is yes, using a wide variety of AI solutions and DATA (lots of it). However, even if you can predict them by running simulations using data from past events (pandemics) the real question is can you pivot your supply chains to respond to the drastic change in consumer emotion and behaviors? COVID-19 has exposed the inefficiencies that already existed (grocery retail) and has accelerated other industries to adapt to new and improved “paths to purchase” processes and technologies.
Retailers are struggling with “the new normal” and how COVID-19 has and will continue to impact demand. Machine learning algorithms require a mass amount of data to evaluate and learn patterns. The cliché “garbage in garbage out” is fitting because if the data (structure, or unstructured) is not the right data the output of any AI engine can and will create erroneous results. Hence the need for manual intervention and evaluations.
COVID-19 is an unfortunate event that has “made retail data murky” but COVI9-19 has created exponential demand for specific merchandise and retailers. Retailers now need to be concerned with calculating new levels of demand (LOD) and those that are not using statistical based forecasting tools vs. AI will be at a disadvantage.
Collecting data is all too easy. In a PostPan world, we need to use multiple technologies and collection devices to get to the truth.
Since there is no defined pandemic end-point at this juncture, it is hard to know if this is a blip in a longer-term cycle or whether there will be a significant period of time that more permanently alters shopper purchasing behavior in new ways in the next normal. For now, data provides a valuable window into current behaviors, helping the nimble adjust and pivot.
The unprecedented shift to online shopping due to store closures, the “black hole” of store-level sales data as a result of the shutdown and the shift in consumer spending from non-essential to essential product categories has disrupted retailers’ forecasting models and allocation/replenishment algorithms. While it’s very true that retailers are struggling with the missing sales data from this spring season (which I believe there are workarounds to this challenge in the short-term), the bigger challenge in my opinion — for both the short-term and the long-term — is that retailers must rethink how they engage with their customers.
Analyzing sales data is not going to be enough in the future. Executives must begin thinking more about what their customer’s beliefs are and how those beliefs affect their shopping behaviors. Companies that develop a nuanced understanding of their target customer’s beliefs, habits and shopping behaviors — and adjust their product offerings, customer experiences, and marketing communications accordingly — will be best positioned to thrive in the future.