When most people think about AI, they think of the splashy stories and applications. They think about DeepMind beating a world champion at Go, Microsoft’s ill-fated Tay experiment, fleets of self-driving cars, Ken Jenning’s Jeopardy! loss: things of that nature. But AI is already woven into the fabric of giant businesses of every stripe, from big tech players like Google and Facebook to hospitals and doctors using it to improve patient outcomes and so much more. One place you might hear about it as much? In retail. But that’s changing every day. In five years, almost every major retailer and brand will be using AI integrations to drive core business decisions. We are moving to an AI-first world where every retail problem is an AI problem.
Now, before we forecast this AI frontier, it’s important to level-set on what it is we mean by AI. After all, artificial intelligence has become something of a buzzword with different people having wildly different definitions. For me, AIs need to do a few important things to be considered true artificial intelligence. They need to observe and learn and they need to decide and act. Not only that, it should all be done without the need for human intervention. In other words, true AI systems are autonomous decision makers.
And when we’re looking at how AI could change retail, we need to be thinking of these sorts of systems. We need to be asking: what problems can be tackled by autonomous decision-makers? A better question might be which problems can’t. In five years, you can bet almost every major retailer and brand will have AI integrations at key steps in the shopper journey, doing everything from merchandising, product selection, returns management, pricing–just to name a few.
Let’s talk about how this works in practice, starting with a fairly common retail problem: how do ecommerce sites know the best products and experience to show each shopper? Can we get to 1:1 personalization? With AI, we can.
Right now, many retailers batch users into cohorts or recommend items that “users like you” have purchased – which is a far cry from catering to preferences of the individual shopper. To optimize the experience, they might run A/B tests to find out which layout and design appeals best to those segments. A vast majority of those tests a) take a lot of time and b) only really allow for marketers to try a small number of ideas. These tactics haven’t improved retail conversion rates for quite some time. We’re still only seeing on an average 3% of shoppers who land on a site buy any products, and brand engagement levels have plateaued.
AI–more specifically, autonomous, decision-making AIs–are uniquely poised to solve this exact issue. They do this by interacting with individual shoppers, in the moment, to learn shopper intent, then act by surfacing products that meet that customer’s unspoken preferences. It’s true 1:1 personalization and it looks something like this:
Say a user and clicks a product. That starts training the AI about what that shopper is interested in. Now, say it’s a red dress. Since a smart AI can be trained and learn any catalog, it knows which products are more similar to that red dress, not by historical data and cohort analysis, but from the image itself. It can analyze the dress and understand everything about it: the scoop of the neckline, the hem, the shade of red, etc. The AI can then act by showing the user items that are similar in style to the dress in question. They could have identical necklines, they could be the same color, texture. in fact, they could be any mixture of like characteristics.
But that’s just the start. Say a different dress piques the shopper’s curiosity. Well, now the AI has two inputs to learn from. Did the shopper click on another red dress? A black one with a similar hemline? Artificial intelligence can analyze these actions in the moment and learn what attributes the user is attracted to. Each click trains it, effectively allowing it to intuit the user’s sense of style as they browse, finding products in the catalog that are available in inventory, and that which fits her individual preferences exactly.
How best to present the these recommendations on desktop and mobile? AI can help yet again. It can run massive multivariate experiments where every button, copy, layout and other design elements are being testing constantly, evolving through applications of genetic algorithms that find better converting sites from the ideas your marketing team already has but doesn’t have the bandwidth (or traffic) to try. Those winning tests will teach retailers about their customers, giving them a much better understanding of what they want to see and the messages they want to hear. Not only that, but these kind of tests run constantly, optimizing the experience as online traffic patterns and behavior patterns change with each season and campaign.
What I’ve talked about are just a couple of examples. But they’re instructive in that both involve AIs making autonomous decisions, learning and acting on their own. In the near future, you’ll see smart supply chains that learn how to spot gaps and issues earlier and learn how to adapt and fill those gaps without human intervention. You’ll see AI systems that learn what consumers are more interesting in and act by forecasting trends you hadn’t even noticed. You’ll even see AI systems that take a whole host of this information and actually design products within the framework of your brand.
Brands like Sunglass Hut, Skechers, Cosabella, North Face, Amazon and so many more have already integrated AI in key parts of their businesses and related digital disruption. And make no mistake: more and more companies will begin to do so. Because every retail problem really is an AI problem. And companies that solve pervasive problems with smarter technology have a better shot at succeeding in the long term.