One of the buzziest and most debated tech terms of the last 5 to 10 years is artificial intelligence. To start off with, what does the term artificial intelligence actually mean? If you just Google “define artificial intelligence,” you’ll discover that even computer scientists and engineers don’t entirely agree on what to call everything that fits under the AI umbrella. And when it comes to putting it to use, there’s even less clarity. For retail brands, many of whom may be sitting on a lot of consumer data, understanding what options are available now, and what could be relevant tomorrow is key to staying ahead.
George Wang is a designer whose work bridges areas of innovation and technology, product development, and software. And he’s going to talk to us about what artificial intelligence means for retail design.
Transcript
Melinda: Hi. I’m Melinda and you’re listening to Think Retail.
One of the buzziest and most debated tech terms of the last 5 to 10 years is artificial intelligence. To start off with, what does the term artificial intelligence actually mean? If you just Google “define artificial intelligence,” you’ll discover that even computer scientists and engineers don’t entirely agree on what to call everything that fits under the AI umbrella. And when it comes to putting it to use, there’s even less clarity. For retail brands, many of whom may be sitting on a lot of consumer data, understanding what options are available now, and what could be relevant tomorrow is key to staying ahead.
George Wang is a designer whose work bridges areas of innovation and technology, product development, and software. And he’s going to talk to us about what artificial intelligence means for retail design.
George, welcome and thank you for chatting with me today. Can you start us off by telling us a little bit about you and how you came to be interested in artificial intelligence?
George: Sounds good. Thanks for having me, Melinda.
To give you a little background, I’ve always been this person who’s fascinated with the intersections of ideas and people that don’t normally connect, right? And over my career I’ve become a bridge builder of sorts, which sort of explains my career – starting from graduating in Mechatronics Engineering at Waterloo to co-founding a startup, to getting into design and design thinking at Johnson & Johnson and later, BCG Digital Ventures in 2017, which for those that don’t know is Boston Consulting Group startup studio where we help clients launch new tech startups from the ground up.
It was during my time at Digital Ventures when I saw AI being loosely used as a buzzword by a lot of the clients and the business folks I worked with. And they understand the business challenges quite deeply, but their grasp on the capabilities and limitations in nuances of AI was a little bit limited. In the meanwhile, the engineers had the other piece of the puzzle normally, you know, what is under the hood of AI. But the domain knowledge was quite limited. And so when engineers and the business people are talking to each other, they’re oftentimes having two different conversations. And so being a bridge builder and as the only design strategist with an engineering background at Digital Ventures, I saw this opportunity to step in and create some frameworks to how to identify applications for AI in business.
One thing led to another, I left Digital Ventures and later joined a leading Canadian AI company called Dessa to lead U.S. expansion. Before acquired by Square earlier this year, the company’s focus was around building and deploying machine learning systems end-to-end for some the largest financial Fortune 100 companies in Canada and in the U.S. And the work that Dessa did was truly cutting-edge and, you know, we had only 12 engineers, but were able to create half a billion dollars’ worth of value in just over three years.
Melinda: Wow.
George: And there I worked along with machine learning engineers and saw what it was like to develop and deploy an AI system from end-to-end, from identifying this application to experimenting with the model, to deploying it in the real world in affecting actual business decisions. And that experience, it taught me a lot about the state of the art of AI and eventually brought me to where I am today, actually building bridges between AI and the design community so that we can have a continuous dialogue about the future opportunities where design can help AI reach its full potential.
Melinda: Excellent. Well, you’re the perfect person for us to talk to then. There are a lot of terms that get thrown around, A.I., machine learning, strong A.I., weak A.I., deep learning. How do you categorize names and talk about what is commonly referred to as artificial intelligence? It’s a big question, I know.
George: Right. It’s a big question. I think there’s a lot in that question and, you know, I doubt we have the time to cover it all, but, you know, you can quite easily Google this sort of stuff. But the thing that I want to talk to you about is AI is quite a buzzword these days.
Melinda: Right.
George: And it’s really hard to have a productive conversation around it because it kind of means everything and nothing at the same time. And I have two reasons for this. AI is as broad as health care, you know, as a word. And it’s like a collection of technologies that are united by common characteristics. Just like everything in healthcare, it’s about the human body. Right? That’s pretty much it. I mean, there’s a little more, but, you know, in principle, AI is this collection of things that are vastly different, but, you know, there’s a lot of math, there’s a lot of commonalities.
And the technologies themselves are very different. For example, between natural language processing and object recognition, you know, completely different field even though they’re all categorized under AI. So you say, “Well, we want AI.” What are you really trying to say? Like, when I have a bill, I don’t just get “healthcare.” I need to see a typical type of doctor for something particular, right? So for that reason it’s kind of a buzzword and one, you know, one reason I would discourage from using that word so much.
The other reason is that its definition is constantly shifting. The word AI was invented in the 1950s when they barely had any computers. I forget who it was that actually came up with the term, but, you know, there was a system called ELIZA and she was supposed to be this avatar that responded to certain queries. And entire thing was a rule-based system. So, you know, the program was, “Hey, if the person…if the user enters this in the input, then this is your response.” You normally wouldn’t consider that as AI these days, but, you know, that’s where the word came from and the definition of AI shifted with time. You know, one of the interesting quotes I’ve heard about it is AI in fact is it’s something, AI is something that a human can do, that a computer cannot do yet. And given the constant shifting definition of what AI is, with this very broad definition, it’s really hard to pin down when two people are having a conversation on AI, what are they really talking about.
And so that leads us to how can you more precisely refer to AI, what do you mean when you say the word AI. And, you know, I’ll start off with the reason why we’re even talking about AI these days so much as we do is because of the advent of machine learning and most recently, deep learning. And these are the words that I encourage you to use. And, you know, most of the time when you’re referring to AI, you’re actually referring to mobilities made possible by the technology of these…of machine learning and deep learning.
And so, what is machine learning? It is “the scientific study of algorithms and statistical models that computer systems use to perform a specific task without using explicit instructions.” The operative words is “without using specific instruction.” So, in the case of ELIZA back then, you know, that’s not machine learning, that’s just if the user says this, this is how you respond. That’s explicit instructions. What is not explicit? When your email is looking for spam, right? Like, if your email comes in, is it spam or is it not spam? There could be a million ways that email could be worded or phrased, what things could be in the email. You can never possibly come up with all the rules to let the computer discern whether it’s spam or not. Machine learning is able to let the user feed tons and tons of examples. Like “here is an example of a spam email, here is one that is not, and here is one that looks kinda like a spam email, but it’s actually not.” So, lots and lots of examples of both the positive and negative kind of versions of a spam email, and writing code, training code to let the computer essentially figure out on its own what is a spam or not spam email.
Deep learning is a method in some constant set, you know, a method of machine learning that’s responsible for much of the AI you’re seeing today. So, machine learning has existed for a while, and deep learning, it constantly resembles what you see in science fiction. Machine learning, you can use it to analyze tabular data, for example, transactional data and banking data, you know, everything that exists in a table. But what deep learning is able to do in addition to that is analyze unstructured data. What that means is the data that exists in an image or in a song, right? Like, it’s just ones and zeros. It’s not a tabular data with columns and rows, and it’s to identify, for example, the objects within those images. That’s propelling self-driving cars these days. So, it’s because of deep learning this technology that is why AI has made a comeback in a way since the 1950s.
Melinda: So, okay. You prefer machine learning and deep learning; you prefer those terms to AI. What is the most common use for machine learning or deep learning that the average consumer experiences in their daily lives?
George: You know, AI is already everywhere. I don’t think there is a most common use. I think we’re already using it. It’s already affecting us in a myriad of ways that we don’t even realize.
I actually don’t know if this is machine learning or deep learning because the most impactful applications of it are in the back end. It’s not so much the interface, right? So, the lights turning green at the intersection, that is based on some algorithm and perhaps it’s real-based, perhaps it’s driven by some statistical data. Maybe in the city of the future it is driven by something like deep learning, where you take into account a multitude of dimensions to determine whether the light should turn green or not. You know, another example is when you’re typing in Gmail, Google autocompletes the rest of your sentence and you just hit tab to kind of populate it. You know, that’s an example of deep learning, it usually has to do with language processing.
Melinda: So, in the world of retail there are a few uses for predictive analytics that are currently being deployed and one of them is for predicting product assortments based on a specific location. For example, H&M and Nike are both using this kind of a strategy to put different products in different places based on what they predict is going to sell better. Can you explain how that might work?
George: Yeah. Sure. I think in terms of retail, the brand we’re probably all familiar with is Amazon, and the secret to its success as you might already imagine is the highly, highly targeted recommendations that’s personalized to each individual customer, right? Based on your browsing patterns it’s able to figure out what you are most likely to buy, it recommends those products to you. And in the case of hyper local, it’s similar in the concept of accepting the greater emphasis on geography, and rather than optimizing your recommendations based on what you predict for each customer, you’re optimizing the inventory based on the collective preferences of each neighborhood. So, it’s really not as much of a difference in terms of concept, but the underlying technology that’s… You know, I don’t know what Amazon’s proprietary machine learning tech stack is. And that’s something now that kind of needs to be figured out for each retail company.
Melinda: So that wouldn’t be a really super complicated type of system that we’re talking about here. It’s relatively simple, what you’re saying?
George: It’s all relative, right? And it would take some investment, and I think there are lots of low hanging fruits that companies don’t realize that they have. When some hear AI or machine learning, or deep learning, “Oh, my God, these are words that are so new and some other people, they have the resources, they have the talent, they have the data to do it, but not us.” But in reality, it’s quite achievable, you know. I built an AI system in less than a day from not knowing how to build it in the morning and actually built one during the night that can classify an image based on what’s in it. So, I think this technology is getting to a point where it’s so accessible for individuals and for companies that, yeah, I mean, the value is there. And, you know, I just urge companies to give it a try, experiment, carve out a little bit of a section of customer base and just to run a few experiments. Limit your risk, but see for yourself whether, you know, with a little bit of investment, whether it’s going to pay the value you’re hoping for. And then if it works, scale it.
Melinda: Another challenge that a lot of retailers are facing especially now with people shopping more online is online apparel returns. How could retailers potentially use AI or a Power BI training course to address this issue?
George: Right. So, you know, AI has proven to be a great solution to this challenge already and before you think about how to implement machine learning or AI, I think there’s plenty you can do leading up to that without skipping right to the technology, and why are people returning. Could it because of sizing? Is it the style? Is it the texture? You can uncover some of these questions by just doing some research. And maybe it’s a rather straightforward fix. For instance, I’m tall and quite thin, and it’s hard for me to find a shirt that fits nicely. And, you know, maybe there’s to be a store-wide system to standardize the different sizes from the various manufacturers. But if you do decide to experiment with machine learning, having some preliminary insight into the problem, it will also allow you to be more intentional in where you apply the technology. And if it’s sizing, we may want to use machine learning to, for example, recommend the best size for the customer based on their past purchase and returns behavior. And if it’s style or texture you may want to develop a machine learning system that automatically photoshops a customer into a virtual fitting room where they can see themselves in everyday situations wearing the same piece of clothing.
So yeah, this technology does exist. And these applications are vastly, vastly different. The more strategic and intentional you can be about where the real problem is that you can apply AI, I think the better results you’re going to be able to see. You know, say, you do want to predict what items are going to be returned, what you do want to figure out is what combinations of the variables are correlated with the lowest and highest return rates? Just like the previous example we gave about spam and emails, you know, you’re giving this computer a thousand spam, a thousand not spam email, and letting it figure it out, the same thing here.
Melinda: Right. You just give it a lot of information and let it kind of determine the variables that are maybe contributing to the product.
George: Yeah. Right.
Melinda: Right. Okay.
George: And what do you do with that knowledge? You know, do you automate a certain action to take place in real time like disincentivizing a purchase on something that will be returned or perhaps offering a coupon for making the return nonrefundable before it is purchased? And again, it’s different for every company depending on the data they have. Take it step by step and don’t rush, you know, straight into implementing it.
Melinda: Right. Yeah. Okay. So, I want to talk about chatbots because chatbots or, you know, virtual assistants especially, again, everybody’s shopping online and even post-pandemic, a lot of these behaviors are going to stick. With AI-enabled chatbots and assistance – with complex language still being something that’s really being developed – is the cost of this type of technology worth the benefit do you think?
George: Right. As you can see I don’t give straight answers. I think there’s a lot of nuance, right? There’s a huge diversity of chatbots out there ranging on, you know, levels of technology sophistication to the use case and I think, yeah, there’s a level of rigor that needs to be applied to discerning certain technologies that give off the appearance of AI from the ones that actually are leveraging AI’s predictive capabilities.
Melinda: I mean, some of them are still just really like, ELIZA where they’re just answering you know A, B, C, or D response based on whatever you put in.
George: Yeah, yeah, yeah. Exactly. And in terms of benefit, it largely depends on where and how the chatbot is applied. If you haven’t charted the right customer problem to solve for the chatbot or you’ve implemented in a way that’s frustrating to use for the user, then it’s not going to generate ROI that you’re looking for. And, you know, speaking from the point of view of a designer, one of the biggest UX issues with chatbots is the customer not knowing what they should ask. So, I think an opportunity for machine learning is making predictions around the question that the customer is trying to answer at that exact moment in time, and when they’re on the page of your website, are they hesitating? What are they thinking about? Perhaps machine learning can be used to kind of draw out some of those conclusions based on their real-time behavior and having the chatbot initiate the dialogue in a much more personalized way.
Melinda: So the retail industry is often viewed as being really a very slow moving business environment where embracing innovation is more exceptional than normal. But given the current circumstances, many companies have been forced to accept new ideas really quickly. Do you think that the pandemic could precipitate an era of greater innovation in the retail sector? And if so, how could predictive analytics play a role?
George: It’s no question that COVID is driving massive changes in consumer behavior today, changes in discretionary income, in spare time, when you consider values and priorities. And what’s interesting about COVID is that it’s a stepwise change given the fact that machine learning models predict behavior based on past data. Companies that use a machine learning model could have learned by now that the model stopped working when the future is vastly different than the past. And some correlations will remain the same, like, we’re going to sell more sweaters in winter than in the summer, right? But there are new patterns that are potentially emerging. For example, is there a relationship between the COVID rates and the quota of the COVID purchasing? I don’t know. These sorts of invisible correlations are what machine learning is able to detect for us.
And so, to your question whether COVID would precipitate an era of greater innovation, yes and no. Yes, that consumer demands are changing, and the retailers that can keep up with these demands will succeed, and no because brick and mortar retailers are under a huge amount of pressure and many of them are clamping down budgets and focusing on a broader line rather than innovating. So, as we move into a future that’s changing more and more rapidly with more disruptive events, what is COVID for next year? You know, just based on past few years, we’ve come to see a lot of destruction, right? And it’s not going to stop here. And I think the most agile companies will adapt to the change and the ones that cannot will be outcompeted.
Melinda: Yeah, absolutely. I completely agree, and, I mean, I think many people think that once there’s a vaccine we’ll be back to normal, but, you know, climate change is another thing that is looming in our much near future than I think most people imagine as another disruptive force and there’s other things that of course, we can’t predict. So how could retail brands intentionally move to a culture where innovation is more intrinsic instead of one where, you know, you take Pepper the robot, and you put it in the store, and you call that innovation?
George: Right. Yes, Pepper the robot is kind of like this cliché, right? But, you know, in business what matters at the end of the day is how you’re reaching your goals. If Pepper the robot is actually giving customers positive experience and your investment results in real tangible business benefits, then you should definitely invest in Pepper the robot. In most cases, companies have put Pepper the robot in because it looks like innovation, but it doesn’t actually create real value for customers or for the business.
So what you should be doing is one, understand the capabilities of a particular technology for something like AI, with capabilities that are as diverse as it gets. I keep saying, I talk about it like ML and deep learning, I guess, is force of habit, but even what I’m referring to AI, I really mean this huge collection of technologies that are vastly different. So, you know, just get a level deeper into what the technology is capable of doing, and first and foremost really, understanding the challenges that your customers and your business face that, again, could be uniquely addressed by this technology by finding the intersections between the business need and the customer need.
I do want say that innovation isn’t real unless it’s implemented and companies either treat like an all or nothing game and shy away from investing in anything, whether they missed out on the opportunity or they’re so committed to change that they spend a ton of money and implementing something into panning out. So, what I would advise is determine part of the business as a POC experiment, and just, you know, see the results for yourself.
Melinda: It’s always a great pleasure to chat with George. He might not give you a black or white answer, but he does give you a lot to think about.
So, what’s hitting home for me? There are a couple things. One is the idea that AI is as broad as healthcare. I’ve never heard it described that way before and it does help me frame artificial intelligence a little bit more clearly. Another is the idea that AI can help you define your problem, and this is a big one because we do see a lot of retail organizations relying on historical information or traditional research practices to draw conclusions about a challenge, but it often turns out if we approach it differently, we find that the challenge is something completely different than what we originally assumed it was. And if machine learning and predictive analytics can help us define the challenge, this could help us break out of some of those biases that lead to inaccurate conclusions. The last thing that George was talking about was agility. It’s another buzzword that we hear all the time, but given how fast things are moving, it is really the only way I can imagine any business model staying alive. And the great news is that machine learning can help, and according to George, it’s more accessible than you think.
We would love to hear about how your brand is experimenting with machine learning. You can reach us at info@sld.com. Thanks for listening.
About
George Wang is the Head of Design at Wysefit as well as a product and design consultant. As a researcher and designer, he is passionate about solving meaningful and important problems.
Think Retail is a podcast where top designers, strategists, thought leaders and business people discuss what’s coming next. For more information, email info@sld.com.
