What AI can teach you about your customers
Photo source: iStock.com
By Andy Narayanan, VP intelligent commerce, Sentient Technologies
When most businesses think of artificial intelligence, chances are they think about chatbots, logistics, or really any machine learning algorithm that improves their business processes. But what some companies are finding out is that AI can do a whole lot more than just streamline operations. In fact, it can teach them a lot about their customers.
Before we jump in, it's probably important that we level-set about what we mean when we talk about artificial intelligence. After all, it's become something of a buzzword over the past five years and a lot of people use it as a catch-all. My view is that true AI is capable of autonomously learning, autonomous decision-making and autonomous action. It can observe the world, orient itself against the situation, decide what action to take, and then actually take that action, all without human intervention. Some readers may know this as the famous OODA loop, short for Observe, Orient, Decide, and Act, and that's exactly the sort of systems I want to focus on here.
So what kind of AIs arewe talking about? Which AIs are going to both improve business outcomes and teach businesses about their customers? Let's start with a pervasive problem in retail as an example: the problem of truly personal shopping.
Right now, a lot of the sites we visit display products and recommendations that are quasi-personalized. They rely on cohort analysis to present products that "users like you" have purchased in the past. And while that's better than no personalization at all, it doesn't look at individual customers in the moment making individual choices about the products they like. Instead, it lumps us together in buckets.
Now, with artificial intelligence, businesses can get actually get to true, one-to-one personalization. It works by analyzing images. Here's how:
Say a customer hits a site and she clicks on a product she’s interested in, say, a red dress. The AI looks at the red dress, again, the image itself,and analyzes the image vectors to orient it amongst a retailer's catalog. In other words, it finds other images that look like that dress, but in different ways. It can now start showing recommendations based on that dress. It might show products that are also red, or have a similar length, neckline, fabric, or any combination of similar characteristics.
It gets more interesting with the next part. Say that customer clicks one of those new recommendations. Now, the AI starts learning what she's really looking for. It compares the original choice with the new one and finds the similarities between the pair. Each successive click through recommendations trains the AI to figure out what exactly she's looking for, what subtle characteristics the products share that really meet her intent and sense of style. And it doesn't take many clicks for that user to find the exact right dress among a vast catalog of like choices.
That in and of itself is a big deal. After all, it's true personalization, recommendations that don't lump shoppers into buckets but treats them as unique individuals. But a business can use this to really understand both that particular customer and their audience at large.
For example, by learning that customer's sense of style with AI, re-marketing to her becomes far more personal and exact. Email offers or living pages that highlight not just the products other people like her bought, but products that fit the journey she went on as well as products that match the look. Past that, businesses can extrapolate really interesting trend data from click-streams of all their customers. What brands and styles are they interacting with most? How can a retailer use that to forecast trends, manufacturer buys, coupons, and deals? The list of uses goes on and on.
Now, take another example of AI teaching businesses about their customers: AI-powered conversion rate optimization (or CRO). Most sites sites know about – and use – some CRO tool already and generally, that's an A/B testing solution. Those allow sites to learn about the offers, designs, images, and messaging that resonate with their customers and give them better conversions. The problem is velocity. A/B testing takes a lot of time and a lot of traffic and most tests don't actually move the needle.
But artificial intelligence actually solves those problems. Instead of testing A vs. B, AI-powered solutions can leverage evolutionary algorithms that allow sites to input not just a single idea but dozens or hundreds. Instead of figuring out which offer message resonates, imagine figuring out which message, headline, button color, site design, image size, image style, navigation bar configuration, and right rail offer combine to create the highest converting experience. It's a vastly different, vastly better way to test, one that speeds up the CRO process immensely, and lets sites try more ideas in the same amount of time as typical tests take. And when you know exactly what combination of those elements really resonates with your customers, you learn a lot about what they like.
Say, instead of hearing about free shipping, they actually would rather join a frequent buyer's club. Instead of seeing a picture of a man using a product, they'd rather see a family in the park. Instead of clicking on gray buttons, they like bright, bold ones. Each of these things can give companies real conversion gains while simultaneously helping them profile their audience and use those learnings to influence future marketing efforts.
With promise like that, you can rest assured that artificial intelligence really is going to be everywhere. The examples above are the exact sorts that you should expect to see taking off: AIs that help customers find what they want, help business get more for their dollar, and teach sites about what their customers want. And tell me: which of those things sounds like a bad idea?