The Discovery Gap
Have you ever had trouble finding a movie to watch? Or lacked inspiration for a gift? Or been unsure what book to read next? You’re not alone:
Online retailers are notoriously poor at facilitating discovery.
But there’s a glimmer of hope for the future.
Lookup or lookout
Online retailers excel at helping customers find the products that they already know they want (call it “product lookup”). If you know the name of the movie, the product, or the book that you’re looking for, it will only take a matter of seconds to find it (and buy it) on your marketplace of choice.
But what if you don’t know precisely what you’re looking for ahead of time? Online retailers will typically fall back on directing you towards popular products — whether that’s the latest blockbusters or this week’s New York Times best sellers. But more often than not, those top-selling products won’t align with your personal taste.
Andreessen Horowitz partner Benedict Evans recently pointed out this discovery gap:
“Amazon is great at selling you what’s on the table in the front of the bookshop, and at selling one copy a year of a million or so obscure titles, but it’s not very good at showing you what’s on the shelves at the back of the bookshop. It’s not so good at selling the mid-list — things that you didn’t know existed, or didn’t know you wanted, before you saw them.”
So, how can online retailers better facilitate discovery?
Browsing is the most basic solution to discovery. Don’t know what book to read next? Then start by browsing the fiction category on Amazon, and go from there. In the early days of the web, browsing was confined to a single taxonomy; today, it’s typically multifaceted with the ability to filter by price, rating, creator, and other criteria. These are welcome additions, but, as with browsing in a physical store, online browsing is a time-consuming activity that doesn’t always lead to success.
A second approach to overcome the discovery challenge is to apply hands-on customer service. For example, Trunk Club and Stitch Fix — two online clothing retailers — provide a personal concierge who works with each client to understand their personal taste and then curate a monthly shipment of clothes. As with high-end department stores, this high-touch customer service approach is justified for big-ticket items, but it doesn’t work for low margin products.
An alternative to connecting customers with a personal concierge is to allow their social network to play a similar role. Facebook of course is the quintessential example, curating a news feed for you based on the actions of your network. But Facebook’s success in curating the content we read hasn’t been translated to curating the products we buy, as Benedict Evans pointed out in his aforementioned post. Apps like Foursquare may leverage your friends’ checkins at cafes and restaurants to help you decide where to eat, but in general, social signals are most useful when users are in comparison mode (e.g. should I book hotel X or hotel Y). I think of social signals as the browse approach with an added layer of social context — useful, but often not the primary driver of discovery.
A fourth method is to use algorithm-driven personalization, first to understand the customer’s taste, and then to tailor their digital experience around those taste characteristics. This mirrors the approach a personal concierge would take, but utilizes machine learning to accomplish the task instead of humans. Since it can occur instantaneously and inexpensively, algorithm-driven personalization can be deployed in scenarios where a human representative would be cost prohibitive.
For example: finding new music to listen to. In May, TechCrunch noted: “Apple Music, Tidal and, until recently, Spotify, failed at music discovery because they stuck to a blog-style format.” But Spotify succeeded in closing that discoverability gap by adding a “Discover Weekly Playlist” alongside other recommended albums. The playlist is updated every Monday with 30 songs that Spotify thinks you’ll like, and users have been flocking to it in droves.
We’ve also started using this approach at my startup Crema.co to understand what kind of coffee people like, and then provide personalized recommendations of other coffees with similar characteristics.
Not all algorithm-driven personalization is created equal, however. Amazon has long offered recommendations, for instance, but their people-who-bought-this-also-bought approach hasn’t adequately addressed serendipitous discovery, instead only surfacing narrowly related products. Other attempts are half-hearted; iTunes’ ‘Related’ selections are particularly poor, for example. In a future post, I’ll attempt to outline some of the principles I believe are necessary for achieving effective personalization.
Mind the gap
Most online retailers still rely too heavily on product lookup, leaving an unaddressed discovery gap. The starting point is to provide a multifaceted browsing experience, ideally that incorporates social signals from the user’s network. But increasingly, smart retailers are becoming more sophisticated in understanding their customers and delivering a personalized experienced that helps users encounter products they didn’t know they wanted.