New Normal: How AI Is Reshaping Post-Pandemic Retail with Warut Khern-am-nuai (Read Transcript)

Season 2 Episode 6 of The ‘New Normal’ hosted by Dave Kaufman: New Normal: How AI Is Reshaping Post-Pandemic Retail with Warut Khern-am-nuai
Dave Kaufman – host: Let’s go back, way back, to a period of great uncertainty. Let’s briefly return to March of 2020. More specifically, and my apologies in advance, let’s go back in time to the first half of March 2020 and take a trip to the grocery store. Did you just get a little anxious thinking about lining up, sanitizing your hands, walking with purpose through the aisles trying to find the products you need, and only coming across half empty shelves? Retailers aren’t immune to spikes in purchases, and they’re often able to predict based on seasonal purchases in the past. But March of 2020 was different, COVID was a seismic event, a shock to the system that left retail just as unprepared as everybody else. There were many items that were in short supply, but there was one that we all need that suddenly became the most sought after item at the grocery store, and that was toilet paper.
Dave Kaufman – host: Why did toilet paper fly off the shelves? At the time, many thought that it was about buying something that would give us a sense of comfort and control amidst the uncertainty. We didn’t know what tomorrow would bring, but we knew that there would always be a square to spare. The panic buying of toilet paper, of course, could not have been predicted. But knowing what we know now, a similar shortage should never happen again. And why is that? Because of thought leaders like this episode’s guest, who have taken a real world problem and offered up a solution that allows retailers to react well before the shelves ever go empty.
Dave Kaufman – host: Welcome to the second season of The New Normal, the podcast exploring management research brought to you by Delve, the official thought leadership publication of McGill University’s Desautels Faculty of Management. I’m your host, Dave Kaufman. On this episode of The New Normal, we will discuss how the first wave of the COVID-19 pandemic influenced panic buying, how a data driven framework can help retailers react to panic purchases in near real time, and how one small policy changed by retailers could have avoided the empty shelves in the first place. Joining me for this episode is Warut Khern-am-nuai, an assistant professor in information systems at the Desautels Faculty of Management at McGill University.
Dave Kaufman – host: His research areas are management of information security, platforming for online marketplaces, and predictive analytics. And his areas of expertise are online platforms, management of information security, and predictive analytics. In 2021, he and his colleagues received a major grant for the Natural Sciences and Engineering Research Council, for helping retail industry navigate the post pandemic world with artificial intelligence. To begin, I asked Dr. Khern-am-nuai to take us back to the early and uncertain days of March of 2020, and try to explain what happened to shopping and to purchasing trends for essential goods when the COVID pandemic first arrived.
Warut Khern-am-nuai: At the beginning of the pandemic, we saw the gradual increase in grocery demand and then suddenly, for whatever reason, and people considered that as a panic buying, people start panic buy stuff at a very short amount of time, and the demand of certain products, and especially essential products, skyrocketed in a very short time period. And if I remember correctly, this only (took) a week.
Dave Kaufman – host: It was. In one short week, a trip to the grocery store went from being something you didn’t think twice about to a reenactment of the hunger game games. It was the week of March 5th, when I first went into a grocery store and noticed that everything was quite different, lines were longer, people were more aggressive. A trip to the store would no longer be a happy one. We know how stressful this experience was for the customer, and according to Dr. Khern-am-nuai that in spite of record profits, he was surprised that the experience was far from ideal for the grocery store and their employees as well.
Warut Khern-am-nuai: It’s actually quite interesting in my opinion, because originally I thought the store would love it, because they saw a lot of increase in demand, a lot of increase in sales. So I thought they would love it, but actually they didn’t like that at all. And partly because first, people panic buy, but they didn’t panic use. Think about toilet paper, for example, they panic buy toilet paper a lot, but they didn’t increase the consumption of toilet paper.
Warut Khern-am-nuai: So essentially it is not the increase in sales, they just shift the sales in advance. So instead of buying certain amount of toilet paper over the year, they just do the same amount at one week. So in that sense, it creates a lot of trouble from the supply side. And in addition, it also put a lot of stress on the grocery store. When people panic buy and product were out of stock, several of consumers become stressed, and then they put that stress on the frontline employees and that creates a lot of issues, like psychological issues and stuff like that. So it turns out that the grocery stores and retailers didn’t really like this at all.
Dave Kaufman – host: So this had a negative effect on the employees, as Dr. Khern-am-nuai mentioned, who were, of course, put through such a difficult spot. But they were far from the only ones who had to adapt to a difficult and changing situation. I asked him to highlight others who were not easily able to adjust when panic buying and product hoarding became a reality.
Warut Khern-am-nuai: We are talking about multiple aspects here. For example, if we look from the socio-economic perspective, people with low income household cannot panic buy, they don’t have the resources to panic buy or hoard a lot of products. So in that sense, they couldn’t compete with the demand. And in addition, think about people who are disabled or people who have mobility issues, for those people they cannot come to the store as quickly as other people, and they cannot purchase the products as well, think about disabled people or elderly people. As a result at that time, there were a lot of issues with vulnerable people that they couldn’t access essential goods, toilet paper, sanitizer products, which arguably are more important to them than to healthy people like ourselves.
Dave Kaufman – host: There were and continued to be numerous issues where COVID laid bare the inequities in our society. Perhaps you were lucky enough to be able to fill up your car to the brim with groceries and toiletries. Some without access to a vehicle, or without the disposable income needed to make a massive shopping order, weren’t so lucky. The way we all hoarded goods in the first wave of the pandemic may have been the first sign that we are not, in fact, all in this together. It was every man and woman for themselves.
Dave Kaufman – host: So we’re going back to March of 2020, when it became obvious that the pandemic would change the way that we live our lives, at least in the immediate. At that moment in time, chaos rained, remember disinfecting your groceries, being told to put your clothes in the wash the second you got home, remember the uncertainty? Canadians were told that the only trips outside of the house that they should make were for essential services, like going to the doctor, the pharmacy, or the grocery store. And anyone who had to go to a store when COVID first appeared saw that there were problems with stocking, with supply, with people’s irrational purchasing desires for rolls of toilet paper. In that moment, Dr. Khern-am-nuai and his colleagues saw a real world issue that was our society, and wanted to see how they could improve upon the situation. Fortunately, they had already established a relationship that would allow them to quickly get to work.
Warut Khern-am-nuai: We partnered with this retail store before, and we had ongoing project together. And then they suffer from this issue and they came to talk to us and then we brainstormed and eventually we got this idea.
Dave Kaufman – host: I asked if he could explain a bit of the process to me, like when and how they started collecting data?
Warut Khern-am-nuai: Well, fortunately this project is in partner with one of the largest retail store in Canada. So technically speaking, we don’t have to collect any data because the retail partner collected the data for us and they share all of the data with us. After we got the data and it’s a huge data set, we spent some time exploring the data and then use that data to develop the machine learning AI model to detect the panic buying.
Dave Kaufman – host: When I asked if he could share the name of the retailer, he said that if he told me he would have to kill me. So since I love podcasting and want to live to record many more, I decided that I wouldn’t push the matter further. Instead, I asked him what type of data him and his colleagues were looking at?
Warut Khern-am-nuai: The primary source of data is the consumer sales transaction. So we got all of the sales transaction that happened at all of the stores in Canada.
Dave Kaufman – host: Okay. So you and your colleagues were able to pour over millions of data points, and millions more sold rolls of toilet paper than your average store sells in the average month of March. I would imagine that it’s one thing to predict the purchasing of candies and pumpkins around Halloween or Christmas lights in December. But it’s another when trying to predict what people might be clamouring for during a rare event. So how does one use technology to detect or possibly even predict what consumers could be asking for somewhere down the road?
Warut Khern-am-nuai: That’s an excellent question, and in fact, we have to clarify this a little bit. So typically for this type of data, we would try to predict. In other words, we would try to predict that the panic buying behaviour will occur in the future. And in this project, we didn’t even try that because we knew in advance that predicting will probably not work. Because as you correctly mentioned, panic buying is a rare event. For machine learning and AI to do the prediction we need historical data that is rich enough to do the prediction, and because it is a rare event, it rarely happened before in the past.
Warut Khern-am-nuai: So if we were to develop the predictive model to predict panic buying behaviour, it’s not going to work as good as we intended. There are several techniques or tools that allow us to do just that, but they’re not going to be very useful based on our opinion at that time. So instead of doing prediction, we decided to do the detection model instead. So instead of trying to predict that the panic buying behaviour will happen, we instead detect the panic buyer behaviour pattern in real time. When panic buying behaviour occurs at any store in Canada, we can detect that as soon as five minutes after it happens.
Dave Kaufman – host: Dr. Khern-am-nuai is crystal clear. This is not guesswork. The ends aren’t achieved by shaking a Magic 8-Ball. He and his colleagues have developed a data driven framework that takes the guesswork out of product distribution, even during a pandemic. Specifically, they examine the effect of COVID-19 on three key categories of frequently purchased goods, toilet paper, canned soup, and household cleaners. According to the data cited in the study, sales of these product categories doubled, and in some cases tripled, starting at the beginning of the pandemic in mid-March 2020. The research is the first to leverage data, statistical models and machine learning methods to not only detect potential panic buying behaviour in real time, but also help retailers identify what is happening and allow them to strategically react. And it’s this reaction mechanism that I wanted to know more about, specifically what levers would be used to cope with a demand once a panic buying anomaly is detected.
Warut Khern-am-nuai: The first one is the rationing policy, or in other words, the store wants to limit the number of products or product categories that each household or each customer can purchase at the checkout. By limiting that then the store can make sure that more consumers or more household can get essential goods at the stressful time.
Dave Kaufman – host: This explains the hastily written notes you’ll see printed out and taped to a shelf, limit two packages of toilet paper per customer, limit six cans of Campbell’s tomato soup, et cetera. In spite of consumers being used to always being able to buy as much as they want whenever they want, Dr. Khern-am-nuai says that the retailer is sure that this concession is both understood and acceptable.
Warut Khern-am-nuai: This is interesting because there is no formal research on this topic, but based on the testimony from the retail partner, they seem to believe that the trade off between the customer dissatisfaction on being limited to the number of products that they can purchase and the satisfaction from being able to get the product is a good trade off.
Dave Kaufman – host: Dr. Khern-am-nuai tells me that social media apps and Google trends also play a part in their data collection and analysis.
Warut Khern-am-nuai: So in the detection of the panic buying behaviour, we essentially look at the spike in sales. Because as I mentioned to you, we have the sales transaction data, so we can look at the patterns in the sales. So suppose we look at the sales of toilet paper, for example, we can see that the sales of toilet papers, X unit per hour, and then we see that there is a huge spike in toilet papers at this point in time, so that is probably panic buying. That’s the naive approach that we can take.
Warut Khern-am-nuai: But once you hear this, there could be other things that can happen at that time that could lead to a spike in demand. And the obvious example is, for example, price promotion. If there is a price promotion at that time, then the sales can increase a lot and that can have a spike in demand as well, but that’s obviously not panic buying. So the idea at that time was to collect the data from social media because when panic buying behaviour happens, it’s typically spread through the social media. If you remember that panic buying behaviour on toilet paper, at that time, people spoke about people hoarding toilet papers or toilet papers are out stock everywhere on social media a lot. And we use that information the help the model to take the panic buying behaviour and rule out other alternative explanations.
Dave Kaufman – host: And since we’re talking about looking at trends derived from social media, I feel I’d be remiss if I didn’t ask about misinformation. I wonder if there’s a concern that misinformation on social media could be harnessed in a negative fashion and perhaps mislead stores into purchasing things when they don’t need to. Could panic buying information derived from social media be harnessed to deceive?
Warut Khern-am-nuai: Yeah, that was a concern at the time as well. But fortunately, at least, for us at this time, we did take a look at the data that we collect from the social media and we didn’t see an evidence of any misinformation at that time. But you are right that if misinformation, in fact, happens in social media, this approach might be problematic, and then we need to apply some filter to make sure that we don’t take misinformation into account.
Dave Kaufman – host: And at this point would something like Google trends be a safer pick than looking Twitter or Instagram or Facebook?
Warut Khern-am-nuai: Maybe, but unfortunately, if misinformation happens, systematically, Google trend will suffer from the same problem.
Dave Kaufman – host: Are there differences when predicting panic purchases in brick and mortar stores versus online?
Warut Khern-am-nuai: Technically speaking, no. Actually for the offline store, we have the less information than in the online store. In the online store we can track everything, we can track past behaviour, current behaviour. But in the offline store, if the customer didn’t use that membership card, then it was almost impossible to track the identity of each person. So the approach to detect the panic buying is essentially the same thing. But when we do apply, let’s say, rationing policy, we can make that policy more effective in an online environment versus in the offline environment.
Dave Kaufman – host: Is the massive data from the retailer better for predicting panic buying than the data that comes from social media?
Warut Khern-am-nuai: They serve different purposes. So the data from the store are the sales transaction, and then we can use that sales transaction to detect the real time demand. And as I mentioned that real time demand is one side of the picture because you cannot differentiate between, let’s say, panic buying versus spike in demand because of price promotion. And we use that social media information to help ruling out the alternative explanation. So I would say they supplement each other.
Dave Kaufman – host: And now to the results of the study, when applied in the real world, what did the modelling show? Did the findings help reduce panic buying and allow customers to be able to buy the goods that they needed?
Warut Khern-am-nuai: For one thing we can show using computer simulation that if we had this tool before the first wave of panic buying behaviour, and if the store apply this tool, then we can improve the access to essential goods by approximately 56%.
Dave Kaufman – host: The study findings show that the toilet paper panic buying crisis in March of 2020 could have been avoided, if only retailers had imposed a limit of one package per customer, starting in mid-March. Had retailers simply ration toilet paper from March 12th to March 19th when most of the panic buying period in north America occurred, 56% more customers could have been served. I asked Dr. Khern-am-nuai if he could confidently state, if his modelling had been established prior to March of 2020, would the shopping experience for customers have been different in the lead up to COVID-19?
Warut Khern-am-nuai: I couldn’t say this for sure, but we didn’t have the panic buying after the first wave. And we know that the store used this tool inside the store, so maybe that’s also the reason.
Dave Kaufman – host: Have we not had any panic buying since? I recall running from store to store in the summer of 2020 desperately trying to find a Nintendo switch.
Warut Khern-am-nuai: We might have panic buying, but we didn’t feel the same effect. We didn’t see the toilet paper out stock everywhere or hand sanitizer out stock anywhere after the first wave.
Dave Kaufman – host: No, what I saw, I saw it more with higher end goods.
Warut Khern-am-nuai: That could be a completely different issue. That could be a supply issue, because they might not anticipate a higher demand for high end goods, so they didn’t ramp up the supply.
Dave Kaufman – host: Right. When looking at the Nintendo situation, there were many other factors such as the worldwide semiconductor shortage or the disproportionate number of people who like to buy video game consoles and sell them for astronomical prices. But Dr. Khern-am-nuai and his colleagues’ data is reliable for grocery shopping, and that’s what we’re looking at for this study.
Warut Khern-am-nuai: It can also, scalable to other segments, but only for mass buying segment.
Dave Kaufman – host: Finally, after pouring through mountains of data, I was curious as to what surprised Dr. Khern-am-nuai the most about panic purchases made during the first wave.
Warut Khern-am-nuai: So to me, from the consumer side, I was very surprised by the product category. It is not too surprising to see products such as face mask or hand sanitizer being out of stock. But some products are super surprising, for example, yeast, and especially high end yeast that is used to make high end bread. People panic buying yeast for whatever reason, that surprised me a lot. And from the supply side, I was very surprised by the fact that the companies and the retailers care about these panic buying problems a lot, and they were willing to spend a lot of money to try to fix this problem.
Dave Kaufman – host: We might not be able to predict exactly what consumers will panic buy in future critical events. Maybe instead of bread making, we’ll see pickling or jam making takeover in the future. But thanks to thought leaders like Dr. Khern-am-nuai and his colleagues, retailers will now be able to solve panic buying problems rapidly and ensure that customers have access to the products that they need. And while there are never any guarantees about the future that we’re facing, we can remain confident that our best and brightest are learning from the past and are endeavouring to make our world make a little bit more sense. Stay tuned as we navigate this new normal together. The New Normal is brought to you by Delve, the official thought leadership publication of McGill University’s Desautels Faculty of Management, I’m your host, Dave Kaufman. Producers of today’s episode, Dave Kaufman, Robyn Fadden, and David Rawalia. The technical producer of the new normal is David Rawalia.