Stop “feeding” your users
Feeds are a popular design pattern, but it’s time we came up with something better.
What do Facebook, Instagram, Youtube, Twitter, and LinkedIn, all have in common?
They use crazy powerful machine learning to help users sort through a ton of content to find things they actually want to see.
Oh yeah. And they all share a common design pattern: feeds.
Feeds are the most common way to design a recommender system (that’s a fancy term for software that uses machine learning to predict what users are most interested in seeing.) Feeds either link to more content, or allow the user to consume content without leaving the interface. Feed length varies from product to product, but they are almost always designed to facilitate scrolling, sometimes to infinity!
Feeds were designed to solve a noble and important problem: information overload. Ninety percent of the information in the world today was created in the last two years alone. Every day, 2.5 quintillion bytes of new content get added to the noise. How is anyone supposed to find what they need without drowning in all that!?
Enter machine learning. It promises to sort through all that information so you don’t have to. For Google or other search engine products, that means finding the content that most likely relates to your search query. For Instagram and Facebook, it means finding posts from people you care most about. For Spotify, it’s music. For Netflix, it’s movies. For Goodreads, it’s books. For Steam, it’s video games.
Machine learning is almost ubiquitous in digital products these days. Anytime a user has to sort through a lot of information, machine learning is usually there, working quietly in the background. And anytime machine learning is working quietly in the background, it almost always shows up as a feed in the front-end design.
Anytime machine learning is working quietly in the background, it almost always shows up as a feed in the front-end design.
At Shopify, I work on one of our machine learning products called Shopify Home. And, you guessed it, it’s a feed:
But the longer I’ve spent working on Shopify Home, the more I’ve realized that feeds create user experience problems too. Machine learning is evolving and becoming more powerful everyday, but the way we design machine learning interfaces is stuck in 2006 — the year Facebook first launched News Feed.
Why feeds are bad design patterns
Technology should speed up or eliminate mundane, tedious, or repetitive tasks so we have more time for things that really matter. At Shopify, this means helping our users manage their business more efficiently, so they can focus on the kind of work that makes entrepreneurship feel great — like inventing new products or building meaningful relationships with their customers.
But I’m increasingly skeptical that feeds actually save anyone time. Have you ever logged in to Facebook to look up the address of the party you’re going to tonight, only to lose 20 minutes scrolling through the News Feed before you realize what you’re doing?
Even though the address of the party is arguably the most important thing for Facebook to show you, it never shows up at the top of your News Feed! As Tristan Harris (a former product philosopher at Google) explains, this happens because, “Facebook wants to convert every reason you have for using Facebook into their reason, which is to maximize the time you spend consuming things.”
This means that feeds often pull users into content consumption and away from the task they’re trying to accomplish. That’s fine if they’re on the bus to work or if they’re just chilling on the couch, but if your product exists to help users get shit done, then feeds can actually get in the way of helping them to complete their tasks.
Feeds also create a heavy cognitive load for users. When we did user testing on Shopify’s Home feed for example, many participants remarked that there was “a lot to take in” and that they “weren’t sure what they should look at.”
Users feel this way because feeds usually show a long, uninterrupted string of objects that look very similar. Like a “Where’s Waldo” picture, feeds are information dense and visually undifferentiated — which creates a higher cognitive load. And if the cognitive load is high, then users are more likely to feel overwhelmed and anxious. Often this means they’ll leave or go somewhere else, potentially missing a nugget of relevant information that’s buried in all the noise.
A final problem with feeds is that there’s very little predictability or information permanence. Because content is constantly moving around, users are forced to consume information they are most interested in right away lest it disappears. This creates a “now or never” situation that compromises the user’s autonomy. So instead of letting users consume at their own pace, feeds force them to engage immediately by making it difficult to retrieve content at later time.
A better way of designing feeds
Machine learning technology certainly isn’t going anywhere, but that doesn’t mean we need to keep recycling the same old design pattern. It is possible to create an interface powered by machine learning that doesn’t look or behave like an infinitely scrolling feed.
Thankfully, there are some brilliant people out there who are starting to think about ways to tackle the user experience problems created by feeds.
One of my favourite features about Google Now is the [More stories] button at the bottom of the feed. It’s a small touch, but it creates a “bounded experience”: a little bit of friction that prevents the user from getting lulled into the time-wasting stupor of an infinitely scrolling feed.
Similarly, Netflix’s “broken” or “segmented” feed tries to address the problem of heavy cognitive load. In this design, movies are grouped together under different categories (Documentaries, Comedy, TV shows, etc.), which helps users narrow in on the content they are most interested in without needing look at everything at once. There’s still a lot of content competing for your attention at once, but it’s slightly more manageable than consuming everything at once.
But is there a way to ditch feeds all together?
The closest step I’ve seen in this direction is on the host calendar page in AirBnb.
On the right hand side of the booking calendar, AirBnb surfaces insights to help hosts better manage their listings. These insights are powered by machine learning, but they’re designed as contextual tips rather than a long scrolling feed. Users can tab through the tips one at a time, easing the cognitive load and putting an end to endless scrolling.
The AirBnB host calendar is a great example of a machine learning design that’s assistive without overwhelming, and empowering without being time consuming.
The content you need appears right when you need it, and you don’t have to scroll to find it.
Design better feeds at Shopify
Are you interested in designing machine learning products that have a great user experience? Then I would love to work with you at Shopify! Check out Shopify’s spiffy careers page, tweet at me, or email me at gillian.massel@shopify.com.
Special thanks to Penny Allen and Gabriel Muniz Antonio for their user research that inspired this post.
