Have you ever been on a long drive, and decided to flip through some of the local radio stations, just to see what people in that area are listening to? Maybe you were in the mood to discover some new artists or hear a genre you hadn’t listened to for years. Or maybe you were tired of cycling through your same old Spotify and Pandora playlists, and just wanted to be surprised by a totally new sound!
What’s the common factor in all these desires? The element of discovery. In that moment, you’d suddenly had enough of music services that reinforced your existing tastes by playing songs that were similar to ones you’d listened to previously.
As smart as services like Spotify, Apple Music, and Pandora are, all they can do is predict what songs you’re likely to want to hear next, based on songs you’ve already heard. What they can’t predict is when you’re in the mood for an entirely new artist or genre — specifically because it’s totally different from everything you’ve been listening to lately.
Those are the moments when you tune in to the radio, hoping to be swept off your feet by a brand-new musical discovery. And that’s exactly the feeling you can deliver to your customers, when you step beyond reinforcement and deliver a diet of content that matches the customers evolving tastes and moods.
Reinforcement just tells customers what they already like — and that gets boring.
We all know how it feels to open up our favorite music app, and be greeted with playlists full of songs we’re tired of listening to. This isn’t the app’s fault, of course. Your music service’s machine-learning algorithms are hard at work day and night, creating intricate mathematical models of the relationships between your favorite artists and songs — and using those models to generate new playlists and song recommendations similar to ones you like.
This kind of reinforcement learning has also been standard practice in the retail space for a decade or more. From the moment you purchase a product on Amazon, ads for similar products follow you around Amazon — and maybe even around the web — informing you that “you might also like” a whole range of watches and jackets similar to the ones you’ve recently purchased.
Of course, in many cases, those recommendations are annoying, if not outright useless. The fact that you just bought a new smartwatch doesn’t mean you want to buy more smartwatches. Quite the opposite — it means you’ve now filled that need, and your interests and aspirations have moved on.
In other words, your content diet has changed. Now you want to discover new and different products that address your new dietary needs. And that’s exactly where machine learning can help.
Traditional machine learning falls short at moments when customers’ diets change.
Conventional recommendation algorithms are great when it comes to the “machine” side of things — they work tirelessly to build up profiles of all your customers, and keep showing customers a diet of products similar to the ones they’ve already bought (or playing them songs similar to the ones they’ve already listened to).
On the other hand, traditional algorithms aren’t so great at the “learning” part. Consider a case in which one of your customers tells you they’re not interested in a given product anymore. Like Spotify or Pandora, your algorithm is probably smart enough not to keep recommending that exact same product — but is it smart enough to figure out why your customer has lost interest?
More to the point, is your algorithm smart enough to figure out what that customer now does want instead? Conventional algorithms certainly aren’t. Like Spotify and Pandora (not to pick on them too much), your recommendation software probably cycles back to other products the customer has liked in the past — or maybe it casts around almost at random, offering up new suggestions from your most popular categories.
What no conventional algorithm can do, however, is make an intelligent, accurate prediction about where your customer’s taste is headed now — when (to go back to our original example) they’re flipping through the stations in search of a new discovery. To serve up a discovery that’ll click, you’re going to need a new class of predictive modeling.
Predictive recommendations deliver sustainable content diets by serving up moments of discovery.
The moments when customers’ tastes suddenly shift are precisely the moments when they need you most. When they turn to you in those moments, they’re hoping you understand them well enough to show them exactly the product they’re dreaming of — even if they themselves don’t yet know what that product is!
If your recommendations fall short in those moments, reinforcing customers’ past purchase patterns by showing them products similar to ones they’ve bought before, then suddenly their trust in you is broken. They’ll turn to Google — or tune into the radio — in search of another retailer who can help them discover their next favorite product or artist.
That means your machine-learning software needs to base its recommendations on more than just customers’ past actions. With an adaptive model of every customer’s ideal content diet, you’ll be ready to serve up those delightful discoveries at the exact moments your customers seek them out. Over time, those magical moments will build trust — and that trust will keep delivering value over a lifetime.
Interested in learning more about ow machine learning can make a difference in your marketing? Read “How Machine Learning Personalization Drives Meaningful Connections with Millennial Audiences.”
James Glover is President and CEO of Coherent Path, where he helps retailers transform their email program into a modern data-driven channel focused on revenue. Prior to Coherent Path, James was VP of Sales at Desktone where he was responsible for accelerating customer adoption and revenue growth from zero to over 100 customers to whom he sold through world-class partners like Dell and Time Warner Cable. Before that, James held executive positions at Memento (acquired by Fidelity Information Services) and Watchfire (acquired by IBM). He earned a Bachelor of Commerce degree from the University of Toronto.