H&M on AI, Sustainability, Power of Failure
Arti Zeighami, H&M’s head of artificial intelligence, weighs in on the company’s approach to AI.
At one point, H&M suffered the same swings and challenges as other legacy retailers seeing the Internet erode their business. But in the last few years, the Swedish company has been racing toward the future, trying to stem the downturn in sales and go beyond its old model of cranking though massive amounts of inexpensive clothes.
Artificial intelligence factors heavily into its efforts, and the fruits of that labor started emerging last year with earnings that beat expectations. The news immediately catapulted shares of Hennes & Mauritz AB up more than 13 percent, and the company rode the momentum into the first quarter of 2019, when it beat expectations once again with results that, it said, were proof that the “turnaround plan is working.”
Its stock has had some ups and down since then, which fixes even more attention on the company’s earnings report next week. No matter the results, however, one thing remains clear: H&M’s interest and investment in technology isn’t going anywhere.
And so it may be no surprise that the retailer would hit the tech conference circuit to talk about innovation, AI and the retail business. WWD caught up with Arti Zeighami, head of AI at H&M Group, on the heels of his latest session at Code Conference in Scottsdale, Ariz.
“We’ve been working on AI for three years,” he told WWD. “We started back end of ‘15, beginning of ‘16, working with this area, as a capability. And, of course, we’re a large company. We’re in 70-plus markets, [have] 180,000 people, 5,000 stores, and eight brands.”
Large, legacy organizations often struggle to latch onto new tools and platforms, which makes H&M’s quick-turn digital transformation all the more noteworthy.
“We started in 1947, so people see us, of course, as one of these huge, old industries,” he said. “But we’re also very young at heart, we are very innovative, analytical and entrepreneurial.“
That entrepreneurial spirit informs the company’s myriad projects and experiments — everything from changing up how it approaches logistics and backend analytics to increased sustainability efforts through intelligence, improvements to loyalty programs and even futuristic augmented reality-based holograms through owned brand Monki.
Most recently, the company inked a deal with German firm ZyseMe on an AI-powered platform to make customized clothing available online. Meanwhile, digital marketing firm SEMrush dubbed H&M the most frequented fashion e-commerce web site in the world.
“[People] see us as just a fashion company. But we are much more than that,” said Zeighami. “H&M democratized fashion many years ago, but everybody has access to fashion. And now we’re into a new era of digitalization — which is again about democratizing. But it’s different.”
Not that there isn’t a place for brickand-mortar in all this. In the new frontier, as Zeighami sees it, there’s still a role for brick-and-mortar stores. It’s different than simply focusing on sales.
“We have a touchpoint, different touchpoints, for the customer — it could be online, offline, it could be in the app...You know, all these touchpoints are a way for us to interact with customers,” he added. “And people still want to go to stores. But stores probably have a different purpose in the future.”
Certainly “click and connect” features, with online orders available for pick-up in stores, is part of that mix. But the AI chief also sees these physical outposts as valuable sources of data, particularly for chains with a slew of locations.
Many retailers have figured out that the Internet offers an endless-aisle scenario that extends beyond the specific merchandise any one store carries. But being able to track the buying, behavior and even stocking patterns across all those outposts yields valuable insights.
“Typically, when you have the business that we have had for many years, where we’ve flourished by expansion, where we’re adding new countries, new cities, new markets, and we’ve grown really fast… you make things simpler for yourself by creating all these different sets of models,” he explained.
All that data allows companies to look for averages or drill into important differences or similarities from one store to the next, one region to the next, one store size to the next.
“Say you have a store — type A, B, C, D, maybe. Or a shopping mall, type 1, 2, 3, 4. And you create a factory,” he said, “You can utilize the data to understand what is actually going on. You start averaging the data...Before, you maybe said, ‘Oh, we’ll never open a store less than this size. [But] today, you may say, ‘Hey, there is a need for it.’ Because the customer wants. You become much more customer-centric, customer-focused on what the customer actually wants, rather than making things simply for yourself.
“You cannot just go in and do personalization, or recommendations, and say, ‘Hey, you know, we do AI.’ It doesn’t work that way,” he added. “Everything’s interconnected, the whole value chain. So product is created for a customer, and the customer is all going back — that data is all interconnected.”
He cited critical areas — like fashion forecasting, quantification and allocation — and explained that new projects often start with small POCs, or proofs of concept, and pilot programs. They may extend to different markets, different concepts, different countries, and if the results are good, the team moves quickly on to testing. “It’s lots of Agile testing, and learning, and pivoting, and failing,” he said.
Turns out, that’s key. Think of it as retail’s version of tech’s “fail fast” mentality.
“You need to fail to learn how to do it right,” Zeighami said. “There is no ‘do it right from the beginning.’ You test, you fail, and you pivot. And at some point you say, ‘Hey, there’s no more pivoting. Let’s drop this.’” Or, if pilots show positive results, start industrializing and scaling them.
H&M also looks intently at the data for trend-spotting and logistical efficiencies, so it can ultimately reduce waste, both for its bottom line and for the environment.
Fast fashion takes a lot of heat for its negative environmental impacts, which isn’t lost on the company. So last year, H&M hired Christopher Wylie, fashion trend forecaster and whistleblower known for exposing Cambridge Analytica’s efforts to grab and exploit data across Facebook. Wylie signed up as the company’s director of research focused on using AI in sustainability efforts.
For Zeighami, AI has had a deep impact on how the company approaches its buying, supply chain and related areas.
“Sustainability’s impact is very important…you can quantify in the right manner, and make sure you’re much sharper in your quantification, and much closer to the actual truth in what you should buy, and not overbuy or buy too little,” he said.
In these and other ways, however, he doesn’t see AI as a zero-sum game that can fix everything.
“Already, from a very early stage, we realized that an algo [or algorithm] by itself is not going to help,” Zeighami added. “It’s the combination of the human and the machine. It’s the gut feeling, plus the data. That’s what we talk about with ‘amplified intelligence’ and artificial intelligence. Because we say that what we’re doing is helping our organization and amplifying their existing knowledge, helping them become sharper in decision-making.”
He recounted a particular example of an early pricing project. H&M’s algorithms were very good, better than the humans, in fact. But they weren’t perfect.
“It did not understand the difference between the colors. So in the Christmas sales, it couldn’t differentiate between a black sweater and a red sweater, whereas a human would understand,” he added. “Because nobody wants to buy a reindeer sweater [after Christmas]. The computer doesn’t know that, because it doesn’t have that cognitive thinking, or thinking about what’s right or wrong. It just goes binary about the data it looks at. And there’s always a bias in the data.”
The guard rails around data for computing things correctly, he said, are human beings.
“Those failures taught us to say, ‘OK, this is the combo, and how we can enhance, and how we can utilize the Amplified Intelligence’,” he explained. “The point is not to see those failures as a failure, but embrace them. And learn from them.”