Don’t Shun the ‘Un’

Why It’s Okay if Your Users ‘Unlike’, ‘Unfavorite’, or ‘Unfollow’

Like many sites with lists, Rent the Runway has the option to ‘Favorite’ an item. Favoriting is a useful way to weed through the 1000s of dresses available. Typically, when I’m looking for a dress, I will favorite all dresses that pass an initial bar, and then review that filtered list to make a final decision. Favoriting is good for Rent the Runway too — it helps them recommend dresses to me based on my tastes, so my experience keeps getting more personal and more interesting.

Following people on Twitter or rating things you’ve watched on Netflix is similar: the more data you give, the better the recommendations.

However, an important part of this is that it is easy to unfollow someone, change a Netflix rating, or unfavorite a dress if your tastes change. If you can’t reverse your preference, the recommendations will be stale, and it’s all for nought. Not to mention, it’s frustrating to take an action that isn’t easily reversible.

I was very surprised to find then, that on Rent the Runway, it’s difficult (perhaps impossible, still TBD) to unfavorite something.

Why did Rent the Runway make this decision? Here are few reasons:

1. Unfavoriting is thought to be a negative thing for Rent the Runway, so they’ve made it difficult to do.
2. Rent the Runway is tracking favoriting count as a sign of success for the feature, and thus are doing whatever they can to keep this number high.
3. Rent the Runway didn’t have the resources to build unfavoriting.

Let’s ignore 3, which seems unlikely anyway given that adding a 0 to a database isn’t much different than adding a 1.

The issue with both 1 and 2 is that only tracking and optimizing for favoriting is myopic. You don’t want people to simply favorite something, you want your users to perform some combination of favoriting and unfavoriting so that the resulting list accurately reflects your users’ tastes. The goal here is not to increase the number of favorites, the goal is to have the resulting list accurately reflect your users’ tastes. If the list doesn’t reflect tastes, favoriting has lost meaning. Your users might as well be randomly clicking images to turn them upside down or do something else devoid of meaning.

So assuming my reasons above are on track, the lesson here is twofold:

1. Don’t optimize for a first-level behavior metric. Think about the bigger-picture behavior you’re trying to drive.
2. Things like favoriting and following, aren’t just better if they are higher in absolute terms, they are better if they more accurately reflect a user’s tastes.

Make it easy for your users to unfollow, unfavorite and unlike. Don’t hide these actions in dropdowns or secondary screens. Easy reversal of these actions is necessary to get an accurate representation of your users tastes, and the only way recommendations based off this representation can be meaningful.