Tinder: Increase Hookups Using Metrics-Driven Product Design

This was originally posted on Quora, Inside the Product, on 2/9/2013.

A friend recently introduced me to Tinder, a new(ish) mobile dating app by IAC’s Hatch Labs. The concept is simple and reminiscent of HotOrNot. Login with Facebook and instantly begin flipping through profiles of nearby women (or men). Swipe to the left to reject, right to like. Once two people mutually “like” each other, a private chat is started.

Tinder is fueled by variable rewards. Every profile you view “could be the one”. For every profile you like, there’s an opportunity to get connected. The more profiles you “filter”, the more invested you become as your probability of being connected with someone increases.

Ultimately Tinder’s goal and one metric that matters is the number of people connected. This is driven by the following high-level metrics:

(# of profiles viewed) x (# of profiles liked) x (% of mutual likes)

Everything Tinder does should be designed to improve these three metrics. Here are some initial ideas to improve each of these these metrics:

Increasing: Number of Profiles Viewed

• Reduce unnecessary taps to make browsing even faster. Anecdotally (and without shame), I always view additional photos before committing to a “like”. To do so, I have to (1) tap on the profile, (2) stretch my finger to the top right of the screen to tap on the picture icon, and finally (3) slide through additional photos. Alternatively, Tinder could present thumbnails of additional photos beneath the face of the profile to provide an instant, tappable preview.
• Give hope and build anticipation by providing a running counter of the number of users that liked your profile. Unless the number is zero, this gives the user a sense of hope that the next profile could be “the one” and also communicates a sense of progression to keep users more invested in the service (see Nir Eyal’s Desire Engine model).
• Re-engage users by sending a push/email notification when their profile has been liked (e.g. “Someone .5 miles away has the hots for you!”). OKCupid instruments a similar tactic, notifying users when an “exceptionally good match” visits your profile.

Increasing: Number of Profiles Liked

• Learn from user’s behavior and surface profiles that are similar to those previously liked. Does the user only “like” white women? Do profiles with mutual friends and/or interests receive more likes? Is there a relationship (no pun intended) between the date each user joined the service?

Increasing: Percentage of Mutual Likes

• Manufacture serendipity by surfacing profiles that have liked the user. As far as I can tell, Tinder presents profiles in random order which means it may take several dozen or hundred profile views before there’s an opportunity for a mutual match.
• Avoid wasteful impressions by suppressing profiles of people that already rejected the other user. Unfortunately (for some), this may lead to a much smaller sample of profiles to browse through so outliers must receive careful consideration.

There’s More to the Story

Of course I’m not recommending Tinder focus solely on those three metrics. There are several others that support these goals such as:

Number of profiles viewed - network response time, number of crashes, number of people using the app

Number of profiles liked - percentage of profiles with a “complete” profile, average user attractiveness

Percentage of mutual likes - number of profiles views where one person likes the other

You might visualize the relationship between these metrics as a bonsai tree with several carefully cultured branches, all connecting to the trunk - number of connected people (the one metric that matters).

What is your one metric and what metrics do you look at when designing your product? Let me know in the comments or on Twitter (@rrhoover).