Fashion Forecasting: Arti Zeighami on Implementing AI at H&M Group
22:12 Share

Today, we're airing an episode produced by our friends at the Modern CTO Podcast, who were kind enough to have me on recently as a guest. We talked about the rise of generative AI, what it means to be successful with technology, and some considerations for leaders to think about as they shepherd technology implementation efforts. Find the Modern CTO Podcast on Apple Podcast, Spotify, or wherever you get your podcast.

When you're putting a new tire on your car, you don't want to tighten one bolt all the way and then tighten the rest. You want to tighten them all a little bit and continuously tighten. What does that have to do with artificial intelligence and fashion? Find out today when we talk with Artie Zaghami from H&M. Welcome to Me, Myself, and AI, a podcast on artificial intelligence and business. Each week, we introduce you to someone innovating with AI.

I'm Sam Ransbotham, Professor of Information Systems at Boston College, and I'm also the guest editor for the AI and Business Strategy Big Idea Program at MIT Sloan Management Review.

And I'm Shervin Kodabande, senior partner with BCG, and I co-lead BCG's AI practice in North America. And together, BCG and MIT SMR have been researching AI for four years, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and deploy and scale AI capabilities and really transform the way organizations operate.

Today, we're talking with Artie Zaghami, who leads artificial intelligence at H&M. Artie's joining us from Stockholm. Welcome, Artie. Thank you. Thank you very much for having me here. Really, we'd like to hear a little bit about your background. Why don't we start there? How did you get interested in artificial intelligence? And for our podcast today, I'm actually wearing a nice shirt because we're talking with a fashion person. Even though this is audio, I can assure everyone that I look fabulous. How did you get interested in artificial intelligence, though?

I think my interest in the area started many years ago as a young teenager when, like a lot of people, I was studying math and physics and loved all those stuff. And I read these books by Isaac Asimov, the science fiction author. And I was reading about this theory that he had about psychohistory, they called it, which was about how you can start learning

somehow predict the future by looking at the past and apply mathematical models on top of that. And I was so intrigued by that when I was 15, 16 or 17. I was like, this is something amazing. This guy made it up by himself. And, you know, what if this can be for true? What if I could work with something like this in the future?

And, you know, it took me a long time to get there because there between school came and then I started working as a consultant and I was doing programming. I was an architect. I was in business developer. I was a strategist. I did, you know, different startups and all that. And then I ended up in a fashion retailer.

Then I got this opportunity to start looking at advanced analytics as an AI, as a capability. I don't have a formal data science background. I do have an engineering background from engineering school. I even started business school parallel to that. And then life brought me here. Somehow it's like the universe brought me to artificial intelligence. And somebody made fun of it. Yeah, your name is Artie. That's Artie Artificial. It's probably that's why. So, yeah. Oh, yeah. That's pretty funny. Yeah.

Yeah, I always joke about that. Joking aside, I would say like what you described as a diverse and colorful background and experiences, you know, architect, engineer, strategist. How does it help you now? That diversity versus if you'd been all focused, do you have a point of view on that? Yeah, absolutely. I think it has helped me tremendously. I think one of the major things

you know, parts of working with introducing a new capability such as AI into a, you know, old industry like a retail, which have done certain things in a certain way for so many years, is about shift of mindset. It's about transforming people's way of thinking. It's, you know, it's very little AI, it's very little tech. I usually, you know, refer to that 10% AI, 20% tech,

but it's 70% people and processes. So you try to shift people into thinking differently, to ask different questions, to look at the world differently. And working as a consultant, I think that was one of the biggest help I ever got. Because as a consultant, you always make sure that it's not about you shining, it's about your clients shining. And it's about making them the heroes of the day. And it's always been about that. And I kind of sort of even internally with my colleagues and my peers always said, listen, let's make sure that

that we are almost like internal consultants because it's about helping our colleagues to achieve their goals. Ultimately, if you're talking about transforming people's mindset, it's about the rhetorics. It's how you make them understand what you're trying to do and how you make them understand what you're trying to help them with. So, you know, it goes back to the Greeks, rhetorics, ethos, paltos, logos. That's a long way from the ancient Greeks to...

From Greece to Sweden here, I guess. Is there something specific that AI has been able to help with at H&M? Absolutely. You know, we as a company has always been analytical. You know, if you go back to how the company was brought up, even, you know, the company is a family that started as the person family. The third generation is now chairman of the company. Erling, who started this back in 1947, he was very analytical on his way already back those days. You know, there's stories about,

something they called following the bags. You know, when he was trying to enter a new city, he sent out people with pens and papers and were walking in the street looking at the bags of the people of where they were going and if they were crossing an intersection on this side or that side and didn't understand, okay, what sort of intersection they should build a store. We started to look at this artificial intelligence back in 2016 and tried to understand what it entails for us as a retailer.

And we entered a very small area and we did a proof of concept of that area within the personalization to understand, you know, how we can enhance the communication, the personalization, the offering to our customer in a way utilizing AI analytics.

And we saw that based upon the amount of data that we have, you know, the vast information that we have about product, about sales, about customer, we can be really precise on that. That by itself was not pivotal. It was more of understanding how you change the mindset. As I said, you know, you want people to come on Monday morning and ask different questions. And to do that, you need to get them more analytical. And in order to...

penetrate that into the entire organization, we took an approach that was a little bit different. Because a lot of people asked me back in those days, how did you pick your AI use cases? And I said, I don't have AI use case. I have business challenges that my colleagues have. I look at the portfolio of our project and see we have a lot of problems, we have a lot of challenges. And then those challenges have entailed into projects that are going to change and fix those things. And here I can come in and

help in a very small part of it. And I think that was also a very important part, how you infuse something new in an organization, because a lot of people, again, take one part, a very small part, and then they do deep dive on that small part. And they create that, you know, a sexy app or a cool customer-facing stuff. And that's fine. But then you make one part of your organization to become very, very good at that. And then the rest are still, you know, lagging behind. Yeah.

I believe you need to elevate everybody a little bit. So instead of, it's almost like putting a tire on a car, right? You don't, you know, screw one bolt really hard and then do the next one. You just do every little bit and then tighten everything up. And I think that has been a really good approach for us to do that to everybody. And I'm enhancing stuff in the beginning of the value chain with

fashion forecasting, with quantification, how you quantify how many pieces you buy, to how you allocate the garments throughout the whole value chain, to how to put set prices on them, and maybe also working with personalization and all those fancy customer-facing stuff as well. And for us,

AI has not meant artificial intelligence. We have always talked about amplified intelligence because we're amplifying an existing knowledge and competence of our colleagues. So it doesn't have to be the AI that does the decision. It could be combination. And we see that when we do the combination of AI and machine and human, the gut feeling and the data, the art and science, that's when we get the most out of it. I see a lot of things that we do today is that mixture.

I want to pick up on what you referred to as amplified intelligence. I think it's very elegant. And it sort of also underlies the theme you started this conversation with around organization, people,

you know, 70% is the people and organization. And it also ties very well with the research we just did, which is all again, talks about the role of human and sometimes the misunderstood or understated role of human. Comment more on that and particularly different ways that human and AI can interact.

you know, across these different business problems. Right, right. When we did our first pilot, a test of utilizing AI and advanced analytics in end-of-season sales, and that was my first attempt to try to use it on an actual use case and actual business challenge, we saw that very early that the AI could actually enhance what was much better than the human on putting the prices in end-of-season sales results.

And an important part of that journey was to make sure that the teams that were actually applying it, not my team, not the AI team, but actually the people that were working in the merchandise and online on very small selected markets, I let them actually calculate what the outcome was. So they both were responsible for the test.

for setting up the tests, putting all the constraints they want to me and making sure that my algo was not getting anything else than the merchandise we're getting on a daily basis. And then they're trying to understand how better it is by actually calculating themselves. It's perfect. It helps us to enhance our job.

Let's do a next test for the next season, mid-season sales. So a couple of months later, we added the test and we made it a little bit larger. We added another country, another market, two warehouses, same amount of products, and we tested that. I want to add a little more complexity to this, but one thing that we want to do is also add

A third bucket. So we're still going to have a few products to look at, but we want to divide it not in two buckets, but three buckets, where one bucket is the algo putting the price on, one bucket is the merchandising, and one bucket is the algo putting the price on and the merchandising coming in and tweak those prices. Because we saw there are some things that the algo isn't good at.

And we found that very interesting and said, let's absolutely let's do this. And then we did the mid-season test. And the outcome of that was even more interesting because the algo was again, you know, a few percentage better than human. But the algo in combination with human was twice as good as the algo itself.

And actually it was then we started talking about amplified intelligence because we realized the machine by itself won't help us. It's a combination of the human and the machine, the gut feeling and the data. You may have to change your name from RD to Amphi.

Yeah. Oh, my mom won't be happy. Your mom might not be happy there. You described a process where you went further in depth into a pricing process. But earlier you were talking about a process of saying, well, we need to do lots of things in different areas. Yes. How do you balance that process?

tightening all the bolts versus tightening this one bolt harder? Because it sounds like in that example, you were doing some more tightening of one bolt. How do you balance those out? Well, you didn't hear the whole story. Oh, there's more. Okay. Already from the first tightening of the first bolt, the first test that I got, which was good result, but it was not enough,

Arty and I was happy about those results. And I took that result and went to another part of the business and said, listen, we did this with these guys. They were really happy about the result. We saw that we were X amount of percent better on net sales by using Enalgo, which we created in four and a half weeks.

And it has a huge impact on the business. Do you want us to help you to look at this area? Because I know you have a problem here. And then we brought in the data scientists and we put experiments around that. And then when we started that discussion and that conversation for that specific project, it's,

it was connected to another part of the business. And they say, hey, if you're going to do change there, maybe we should do a change here as well. Arti, do you want to help us in this case? Yes, please, let me help. And then we started doing that. And then that follows. And by the end of the year, suddenly we had those eight, seven, whatever use cases that we had. And then we saw that we're actually applying this throughout the whole value chain.

And then, you know, meanwhile, you start finding each of the bolts a little bit more and more. It's like a pit crew. Yeah, exactly. And it's the whole idea of being agile, right? You start small.

You dream big, you start small, and then you scale fast. So you start small with something here, and then you start the next wave, and the next wave, and the next wave, and all these waves have a cycle of starting small, testing a little bit larger, failing, pivoting, and testing more, failing, pivoting, learning, and then goes on and goes on. And in organizations such as ours, we are huge, like 5,000 stores, 70-plus countries, 180,000 people. So there's a huge amount of things when you start to industrialize that.

And AI doesn't mean anything if you don't industrialize it. And Arti, as you were talking about industrializing AI and getting real scale out of it in the context of amplified intelligence, you know, machine is easy to get them to play with human because they don't have choice. How do you get the human to play nice with the machine? And what's the kind of pushback that...

That you would expect or you get. And how do you deal with that? So it's different from. Different levels of people. In the organization. One thing I found out is that. Maybe it's also about the culture. In the organization as well. So we come from a company with a culture of.

always been entrepreneurial, always seeking for, you know, improving ourselves. And that has been part of the organization. It's in our DNA somehow to always try to be better. And that's why also we talked about, you know, the combination that we're just taking a small part of the work and we're actually internal consultants.

So when they own that, it also makes them feel pride about what they're doing. We really went for this DevOps model where we put people into teams where you have a use case lead and then you have business experts and you had machine learning engineers, data engineers, software engineers, UX designers, all working together on a daily basis in a

very agile way by sitting together in the same office, having morning stand-ups. So they were part of the whole development team. And it was my job to make sure that my other colleagues also understand the impact and then utilize this to realize their value in their part of the company, their part of organization, making our company continue to thrive and become even better.

I think there's an important lesson in there for everyone as well, including for CEOs and heads of businesses. We are underscoring the importance of focus on value versus a single sort of minded tech centric focus on building capabilities and building more and more capabilities and the need for the business and the users and the people to come in from the beginning to design that.

Artie, thank you so much for making time. Yes, thanks for taking the time today. Thank you, guys. Sam, I thought that was such a stimulating conversation with Artie. Let's quickly recap. Sounds good.

Sure, and it's interesting, you know, when we talk with Porsche, we never even talked about tires and hubcaps and tightening bolts. But when we turned to fashion, he talked about the importance of doing that, and his analogy was very auto-related. That's right. And when we talked with Porsche's Matthias Ulbricht, we talked about coffee. Exactly. Everyone is, you know, joking aside, I think the common theme we're hearing is the importance of

archetypal problems and translating or transferring these learnings across problems, which interestingly enough is a topic in AI itself, like transfer learning. I mean, it does not exactly mean what we're talking about here. I also thought it was quite interesting how from the beginning Arti said, look, it's about changing the mindset of the people and it's about the organization and it's about

You know, he talked about it as amplified intelligence, you know, bringing human and AI together rather than all one or the other. I think if we pulled out the words, he said learning more than anything else. That's right. That's right. The other important point he made, which I think might be lost on many, is that you can't just start with technology and capabilities. And

reminds me of Field of Dreams. If you build it, they will come. He said the opposite. Exactly the opposite. You can't build it and wait for them to come. You actually have to build it together. You need to get them first and you got to build it together. I think it's an important lesson here. Yeah, he emphasized value first and then structure and that was important. There's an element too of

weakest link thinking that came through that he talked about lots of different places that they were using artificial intelligence. And he didn't actually use this phrase, but part of the idea was that it doesn't do any good to strengthen one area extensively, but not another one. And so that kind of speaks to his, the tension with

value versus structure. He wanted kind of at the same time progressing. He wanted different parts of the business to be progressing at the same time as well, not too deep in one area, not just completely shallow everywhere either. And so almost everything seemed to be about a balance. You don't start by saying, I've got to move everybody all into one center of excellence, then I'm going to go build this baseball field and then everybody will come and play, you know?

Right. The other part of that was that he said that he didn't have AI use cases. And that sort of structure first would lend you towards a thinking of AI use cases first. He said not AI use cases. He said business challenges that we sometimes and often solve with AI. That's right. That's right. Well, that's all the time we have for today. Join us next time with our last episode for this season. We'll be talking with Kay Firth Butterfield from the World Economic Forum. See you next time, Mr. Irvin. You too.

Thanks for listening to Me, Myself, and AI. If you're enjoying the show, take a minute to write us a review. If you send us a screenshot, we'll send you a collection of MIT SMR's best articles on artificial intelligence, free for a limited time. Send your review screenshot to smrfeedback at mit.edu.