Hey! It’s me, economics bat.
For a little background, I wrote about half the backend code for a dating site with over a million users. That is, if a dating site were a restaurant, I wrote the code that did the cooking, cleaning and busing. I’m responsible for a really wide range of features covering basically every aspect of the software. I’m also responsible for most of daily maintenance, and some other stuff I don’t want to tell you about.
I want to see if I can give you some ideas about how dating sites operate. I’m going to keep this stuff nontechnical and focused on the human-relevant aspects of this industry. Specifically, I’m writing for:
- people who want to caricature the startup world in writing and art
- cyberpunks and futurists asking what dating sites do to their users to extract more money and produce more satisfaction
- people who are wondering why they never personally get any attention on dating sites
- people who need to be talked down from the idea of making a dating site
There are no hot tips in here, but you might be able to use your newfound knowledge to derive some!
Promotional and personal disclosure
Some quick disclosure: a personal friend, Hayley, from The Character Consultancy, encouraged me to write this. As mentioned above, I’m assuming, you’re an artist, writer, or even a person who wants to make an app — if that’s so, you should check out her affiliate page.
As for me, I’m being compensated in coupons to the tune of less than a penny a word. I’m going to use her service as an illustrative example later in this essay.
Also, personal sidenote. This is my first post on Medium, and maybe the first internet post where I’ve been candid about my identity. I’d love to talk, but if you figure out who I am, don’t post it! I use this online guise because it lets me write and act differently than I do in my daily life.
An awesome bat photo you can look at before the article starts
The unsuccessful dating site
Most dating sites attract no humans, and fail immediately.
I’ve hung out with a lot of people who were trying to start a new dating site — they write me pretty often, and they like to write our marketing team too! — and usually they propose their idea around a niche. The people who message me tend to assume they have found a niche that has never been tried: often gay dating, dating for real people, dating with verified photos, dating for geeks and roleplayers, Christian dating.
Every new app maker should, at a bare minimum, Google their idea and see if it exists, but many don’t even do that.
While a niche might be necessary, it probably isn’t sufficient. New dating site developers often have the unrealistic expectation that, because they’ve got their niche, people who see nobody on their first login will still come online again and again. They’ll also make errors like counting people who haven’t logged in for years as users, or abusing their users by reengaging them with email notifications when nothing new has actually happened.
The bottom line is that, niche or no niche, it’s impossible for a user to have a good experience on a dating site if they log into it and nobody is there. This is probably the least effective thing about any new dating site. Developers who lack the cynicism to recognize this are likely to be the only users on their platforms.
For a cynical developer, this is a problem that needs solving. And there are options!
- Some dating ventures will copy the user pool of an existing site onto their new dating site. Of course, for users, none of the profiles on the dating site will correspond to anyone who has ever used the site.
- Others rely on bots to generate human interactions. These sites often charge the humans to reply to messages, under the theory that by the time you realize the person you are replying to is fake, you will already have spent money.
These strategies, emphatically, don’t lead to any actual dates.
While there’s money to be made for developers who are willing to do unethical things like the above, there’s money to be made from too-optimistic developers and dreamers. In a gold rush, the moneymakers sell spades.
It’s very cheap to start a dating site, but not free. Let’s say you decide to make a dating site for bat lovers. There are vendors who, for meager three-digit monthly fees, will develop a dating site for bat lovers — by editing a template. They’ll even supply it with a userbase which, unbeknownst to you, consists of computer-generated fictional characters, each given a name like LucinaBat or GottaBatIt1330.
In real life I get occasional messages from people who got suckered in by this latter kind of thing, who are now grossly in the hole financially, and who could really use a little promotional help from the operators of a real dating app. Many of them have Play Store pages, meaning they’ve bought some kind of deluxe package that includes an Android app.
Unfortunately, this is basically roleplaying — or worse, vanity publishing. How depressing!
How dating sites market
So you, a real human who is single, are probably not going to use a dating site unless you have faith that it will work. If you never have any faith, you won’t register; if you lose your faith, you’ll probably leave.
To earn or retain you, a dating site needs to defeat your negative assumptions. Basically, those negative assumptions are that:
- the profiles on the site are imaginary people
- none of the people on the site are active
In most cases, as you know from the previous section, these assumptions are absolutely true.
There are some additional gender-specific negative assumptions. My experiences, put really short: men are worried no women will talk to them; women are worried that the people who do talk to them will be creepy. There are also some gender-specific patterns of behavior: men are likely to sign up for dating sites experimentally, but need to be retained; women are less likely to sign up in the first place.
What this means is that a lot of dating sites try to retain men by marketing heavily to women. They market to women by interviewing successful couples, demonstrating that:
- there are real people on this site
- it is active
- the men on this site are high-quality
This makes the most sense even though most of the revenue on a dating app is expected to be generated by men.
There are some exceptions to this which I don’t really understand. Some expensive dating services for rich people (such as It’s Just Lunch and Kelleher Group) have unabashedly male-centric marketing. I don’t really understand why, and I find it really discomforting. I assume it must work for them, but their marketing tends to remind me uncomfortably of the illegitimate, bot-oriented dating sites. I suspect (but don’t know) that these services might, like the bot-oriented services, be designed around extracting a large amount of money from the user before they even begin to date.
Ad campaigns and targeting
So I wanted to write a few quick words on the technical details of internet ad campaigns. This is a major aspect of how we do business, but it’s one of the things I’m least personally engaged in.
Humans who are desperate to date are pretty hotly competed over, so we and our competitors do something that’s pretty common in online advertising: we market to a population that’s adjacent to dating-interested humans.
Take the interest “Seattle dating.” It might be hard to rank for. Let’s say, though, that lacrosse players are disproportionately likely to be single, willing to date, and rich enough to spend money on a subscription. Specifically, let’s make these be the numbers:
- Seattle dating: 10% conversion rate per view, $1.00 per view
- Seattle lacrosse: 1% conversion rate per view, $0.05 per view
It turns out that, while “Seattle lacrosse” searchers are much less likely to convert, their cost-to-convert is actually much lower.
So, remember Hayley from the intro?
You can scroll up if you want to see me gush, but here’s the short summary. She runs a service (The Character Consultancy) for people who like designing characters, designing worlds, and writing stories, but when it comes to who to advertise to, she targets furries. That is, the online subculture of people (like me!) who like depicting themselves and friends as colorful animal characters.
It turns out that while your average furry is less likely to be a worldbuilder than someone who, say, types “worldbuilding” into Google, your average furry is a lot cheaper to advertise to: furry-specific ad networks such as FurAffinity charge an unusually low cost-per-view compared to general-purpose ad networks like Twitter’s.
As for my employers, I couldn’t possibly reveal who we target. It would shock and amaze you! That’s a question for the Homework section, at the bottom of the article.
Building a matching algorithm
The matching algorithm is the AI part of a dating site. It decides which other users’ profiles you should see first in the app.
There are a few approaches to this — at the least exciting, you often have search-oriented techniques like the ones Google uses, but which happen to be done over profiles. At the most exciting you have algorithms like Gale-Shapley — what Hinge uses — which they colorfully described as “an algorithm that goes on all your bad dates for you.”
You can imagine the Gale-Shapley model of reality as a tiny fictional setting for simulated interactions between imaginary people, each constructed from character profiles written by the apps’ human users.
Unfortunately, I have to get a little cryptic here. My employers and I have spent a lot of time iterating on our algorithm and we’ve seen consistent improvement, but we couldn’t have achieved that improvement without a lot of market research and testing. We think our matching algorithm is worth at least as much as the middling five-figure sum we spent getting it right, meaning I can’t go into detail about it for the internet!
(Fun facts, though: it’s a couple thousand lines of code, and we’ve done enough rigorous testing to decide that it’s probably fairly irreducible from that. Depending on what line of work you’re in, that could seem tiny or it could seem giant.)
So I’ve had a lot of fun conversations where I’ve asked someone else to design me a matching algorithm, and rather than revealing secrets I’d like to give you some approaches that, in my opinion, don’t succeed on their own.
Here are three ways not to design a matching algorithm:
- show each user the users they would find the most attractive
- show each user the users most likely to find them attractive
- show each user the most attractive users who have no one better to date
Here are some tips:
- your matching algorithm is an extension of your marketing: it needs to defeat the assumption that your app will not work
- letting a user eliminate half the userbase from consideration is like cutting your active userbase in half
- the order in which a user sees candidates apparently matters
- men tend to have very different usage patterns than women
- a lot of users are practically indistinguishable from other users
- simplicity isn’t always a virtue
A picture of a happy couple which I found on Google
Well, I think they look pretty happy.
Last thing, I promise! I thought I’d give you all some questions to stew over in the comments. I’m not going to reveal the answer, but I might have some fun hinting.
- If you wanted to find “single women in need of someone to date,” what interest areas would you target? (Hint: being sexist isn’t going to get you far here.)
- How would you design a matching algorithm?