Towards Better Online Dating
It’s clear that the existing online dating sites are all broken.
I thought about working in this startup space for a few months last year, so I wanted to document my thoughts on the subject.
I ended up deciding against working in this space for now, but I still find it highly interesting and know that for the right kind of entrepreneur it could be a great place to dig in.
As you’ll see, the right kind of entrepreneur for this space is one who is willing to iterate and try lots of things, who understands human interactions and economics on a deep level, and who has the patience of a saint.
A few introductory thoughts to get us started:
- Meeting friends-of-friends is the best way to find cool people, because liking is transitive AND because the low social distances force people to be on their best behavior.
- Online dating is on the OTHER end of the spectrum. Social distances are incredibly high. 99% of the time your dating profile doesn’t reveal your real identity, or real friends, so there is no disincentive to behaving antisocially.
- This ‘behaving antisocially’ manifests itself in multiple ways, namely spamming, hypergamy, and false advertising. Let’s talk about each of those.
- Spamming is the act of sending messages to tons of people. The more messages you send the higher your chances of getting dates, but that also lowers the efficiency of the site. It’s a greedy individual behavior that hurts systematic efficiency.
- Hypergamy is the act of pursuing people who are higher social value than oneself. Dorky Dillon wouldn’t have any shot at getting with Model Mary, but he won’t hesitate to lob a message over in high social distance situations where there’s no personal downside, risk of public rejection, etc. Again, this is an individual behavior that hurts the broader system.
- Finally, false advertising is embodied in the “MySpace Pose,” the overly-witty profile, pictures of me on my BMX bike (ha!), etc.
- It’s important to understand that you can’t blame the users for any of these things. The users are just responding to the incentives created by the modern online dating site.
My favorite approach to the online dating problem is to lower the social distances. But before we get into that, let’s talk about the other major challenge: network effects. Also known as the chicken and egg problem.
I’ll leave the network effect problem as an exercise to the reader, and refer you to this Chris Dixon blog post, specifically the part about irregular network topologies, and this slide deck RE getting lots of users.
Aside from the network effects encouraging incumbent laziness (a la eBay), there are three other reasons why the status quo in this space is quite poor:
- The big players are making lots of money off of their self-destructive business model. We’ll talk more later about why their business model kills long term value, but suffice it to say that they’re not looking for ways to grow the market that involve cannibalizing their revenues.
- Second, there are an astronomical number of new approaches a new entrepreneur could try. The search space is quite large so most people throw their hands up in the air and copy large chunks of broken DNA from existing approaches.
- Finally, when new entrants do try new approaches, most of the time they aren’t well contemplated.
But there is still a ton of opportunity. In the words of my respected friend Zao Yang (FarmVille creator), “Online dating is like the mobile world before the iPhone.”
That quote hits the nail on the head, both in terms of the magnitude of the opportunity and in terms of how hard it is to do really great work in this space.
And to redefine the problem a little bit, I suspect there’s more value and fun in helping people meet new friends generically, and only incidentally maybe a significant other. I only talk about this problem in terms of ‘online dating’ because that’s the existing anchor in peoples’ minds.
So let’s say you’re crazy enough to take a shot at this unicorn. How should you think about winning this market?
I think of dating sites as the sum of three components: the business model, the back end, and the front end.
The status quo sites like Match and eHarmony have all three components wrong, I believe.
Let’s talk about the business model first.
The most profitable sites charge users for the ability to send messages to other people. While this scheme generates $350M in annual revenue for Match, it kills long term retention, word of mouth, and the overall user experience.
The average user lifetime on Match is abysmally low. I can’t remember the exact number I heard from industry insiders, but it was about six months, and maybe shorter.
Imagine if Facebook had a user lifetime of six months!!!!!!!!
(I realize that once there’s a successful match the couple leaves the site, which is another reason such a site would be better positioned as a way to meet new people in general. But I’d wager that most people leave because they didn't find a significant other via Match.)
Anyway, here’s how it unfolds. I sign up on the site, and buy a subscription. I find six or seven girls I like, and send them highly personalized messages. About 15% of people on Match are premium members, and therefore only one of those girls can even reply to my message. I get one reply, at most.
So then I start spamming, because I have to, and I'm incentivized to.
…It gets worse.
It’s also in Match’s best interest for me to send spam to a bunch of non-paying members, because Match builds revenue when new users convert because they’ve received a message they can't reply to! It’s for that reason that Match doesn’t tell you who’s a premium member and who’s not.
This broken business model works for big sites that have the network effect lock in, and niche sites that serve particularly affluent markets, like Jdate.
Yet I predict those sites will all die when someone really figures out the online dating puzzle in the next decade. And it won’t just be an innovation in business model; it will come from changes in all three components of how dating sites work.
OkCupid wrote about this business model problem extensively on their blog, and I learned most of this from talking to those guys. (Update: since I originally wrote this OkCupid got bought by Match and took down the blog post I wanted to link to. It’s called Why You Should Never Pay For Online Dating, and Google Cache has a copy here.)
The backend to me is everything that calculates matches, who can see who, etc.
The backend is the only of the three components that I could nail if I were working on online dating. It’s right up a data engineer’s alley.
It’s no surprise that 18 year-old Match takes a simplistic approach to everything. As far as I can tell, every user can see every other user, there’s no opaque throttling down of a spammers’ message delivery, and so on.
These sites also don’t build real signal into people’s dating profiles. A girl can learn more in five seconds of "real life" interaction than from reading a guy's OkCupid profile.
A bad backend?
…No…a horrible backend.
Here’s the principled way to approach this problem: close the loop and get real data about what kind of matches work and which don’t. eHarmony claims to do this but they seem to employ PhD’s in “love science” rather than statisticians. Call me cynical but I’ll take the statisticians, thanks.
(This situation reminds me of speech recognition. Decades ago at the dawn of speech recognition it was the linguists that people looked to for guidance. Guess what? All of their techniques failed. This led to Frederick Jelinek's famous quote, "Every time I fire a linguist, the performance of the speech recognizer goes up".)
So you run a speed dating event with random people. Get them to submit a copy of their Facebook profiles. Run computer vision algorithms on their Facebook photos to see how much skin they show in photos, how often they’re smiling, how many people they’re in pictures with, and so on. Generally extract as much signal as possible from sources of ground truth.
Then watch which people are interested in each other after the speed dating event, and use off the shelf algorithms to build a classifier that will, given two Facebook profiles, predict the probability that any two people would enjoy four minutes of real life interaction.
Then only allow people to access others they are likely to match against.
You should also watch this talk by Joel Spolsky about the sociology that goes into Stack Overflow, if only as an example of the flavor of thinking you need to do.
OkCupid seems to be at the forefront in this space. They get people in India to rate the attractiveness of new members and use that to limit your access to people significantly more attractive than you are. That’s not a principled approach but a step in the right direction compared to other incumbents.
You could really spend years optimizing the crap out of this. And it would be worthwhile.
This one is the most nuanced. Most of it has to do with behavioral psychology and how people interact with technology.
The front end is everything that makes up the product that isn’t the backend or the business model. It’s the positioning, the user flows, and so on.
First, I’m not convinced that positioning oneself as a dating site is optimal. I like to think of the problem as Helping People Meet Cool People. Not only does this make a difference in positioning/stigma, it feels more wholesome, more valuable, and more practical.
Secondly, everyone knows that dating profiles suck to create and suck to consume, as mentioned above. Not only do they lack real signal, they also make for a horrible onboarding experience. You want me to fill eight huge text areas with witty banter, and check or uncheck 250 radio buttons?
Overall I’m not bullish on the trinity of people profiles, messaging, and people searches. I know that’s how all sites are built today, but if I were in this space I’d keep an open mind on redefining the primitives.
For example, what about a site that just schedules weekly group dinners with new people? Every Wednesday night you’re set up for dinner with five other people we think you’ll enjoy. Not only would this get you past the high social distance problem, it moves the online posturing-and-judging step into the real world where it probably belongs.
That particular idea might not work, but I stand by its disruptive nature.
(Since originally writing this I’ve learned about Grubwith.us. Their product has paradox of choice all over the place, though. Even more recently Grouper has done the Grubwithus model but with a more focused value proposition and without the paradox of choice.)
I would also be excited about anything that approaches the problem circuitously. Imagine if you had a mobile product which helped friends hang out with their existing friends. And then you added the ability to meet new people you might like. And so on.
At the end of the day, whoever figures this out is more likely to stumble into the solution rather than foreseeing the right ecosystem in a monolithic, Genesis 1:1 fashion.
Overall, the world will be a better place once this space is revamped. Meeting people is a huge part of life. Today it's just too O(n). Motive, and opportunity.
(Update, 1 year later: One year later, many new ideas have been tried, with limited success so far. One of the prodomenant themes is connecting users to people closer to you on the social graph, for example with Facebook integration. This kind of targeting is akin to what I discussed in the 'Backend' section, and is a step forward. It's largely enabled by broader social acceptance of online dating displacing the need for pseudonym-based online dating. But the larger, harder questions -- how to overcome adverse selection bias, and the bootstrapping problem -- remain insoluble by all of the approaches I've heard about lately. For example, at the present day, if your user onboarding process begins with "go to xyz.com," I think you've already generated a lot of selection bias; you'll be missing out on a whole segment of the population that would otherwise add a lot of value to your ecosystem. I also haven't seen any significant advances in solving the bootstrapping problem.
One of the more interesting approaches I saw was at the Startup School office hours session in 2011 where one of the companies was proposing building better tools for matchmakers. If you ONLY target the matchmakers as users of your website you might be able to avoid the selection bias and bootstrapping problems. I think it remains to be seen whether that approach has other problems, for example whether matchmakers are incentivized enough to risk their energy and social capital to try to make things happen.)
(Update 2: I think there might be a business in building backend matching software and data sets, and selling it to every dating site. There are so many folks building products that are intended to match people, and they all lack any measure of sophistication in how they do it. Their innovations are usually in the front end; they could certainly use help with their matching backends. You could achieve network effects by accumulating training data from a variety of sites, by requiring that they send you feedback data regarding how any given match turned out.)
(Update 3: Tinder has recently gotten broad adoption, but the more I've gotten perspective on this market the more I think it's just not worth going after unless you have a breakthrough business model. Match.com, for being the leader in this space, doesn't generate a ton of revenue. OkCupid got bought for relatively little compared to the work put into it and how much traction they got, etc. It just turns out that some ideas have great business models, like web search, and others have terrible business models, like online dating. Most ideas are in-between these two in the elegance and productivity of the business model; it's clear that online dating is at or near the bottom.)