When Infinite Scroll Doesn’t Work

Dan Nguyen has an interesting writeup of Dan McKinley’s talk about why the implementation of infinite scroll at Etsy didn’t work. Essentially what happened was that Etsy had spent time implementing an infinite scroll interface (i.e. a list of results that automatically keeps loading more items as you scroll to the bottom of the list) before first testing their assumptions that more search results at a faster rate would boost user engagement. They found that the new interface just didn’t perform. Although the amount of purchases stayed roughly the same, user engagement has gone down – now people weren’t using the search so much.

I don’t think speed and volume are at fault here. I think the reason it didn’t work was because the kind of interaction the user has with the data is much more suited to a paginated interface than an infinite scroll. In the case of something like Etsy, the user is searching for something in particular within the list of results. In the case of something like Twitter, the user is scanning and consuming the flow of information. In the latter case, infinite scroll makes sense because the user is essentially reading or scanning down the list of tweets until they get bored. In the former case the user isn’t just linearly consuming the data, they’re mapping the information in their mind, trying to remember the items they like or think are worth researching further.

What this means is that people will want to go back to the list of search results to check out the items they’ve just seen, comparing them to what else they’ve discovered somewhere else down the list. Having a paginated interface lets the user keep a mental location of the item. They may not necessarily know the exact page number, but they will remember roughly what it was, and the paginated links will let them get there easier. Not only does the infinite scroll break this dynamic, it also makes it difficult to move up and down the list, especially when you return to the page at another time and find yourself back at the top, being forced to scroll down the list once again and wait for the results to load. In this way the infinite scroll interface is actually slower than the paginated one.

McKinley’s original assumptions – that users wanted more search results faster – were actually right, and their tests on each of these metrics individually have verified this (albeit with insignificant gains). A paginated interface that loaded more results faster would have worked a little better than what they had before. What they did wrong was use a new interface pattern in the wrong context, thereby upsetting other browsing mechanisms that people have relied on when using their site. An infinite scroll works well for something like Twitter where you’re linearly consuming an endlessly flowing stream of data without looking for anything in particular, but for search results pages where people are looking for a specific item and where the location of all the items the user has viewed matter, the paginated interface wins out.