Over this past weekend, I came across a Wired called “Write the Perfect Email to Anyone with this Creepy Site” on using machine learning and natural language processing to analyze a person’s social media profiles and offer tips on how to communicate with them effectively. Most people immediately thought about sales, but one thread of the conversation drifted towards dating sites..
It’s not on my resume but over a decade ago, I worked part time for a dating website. It wasn’t one of the names you’d recognize.. in fact, they were quietly acquired and have been out of business for nearly a decade.
At first glance, the business model for a dating website is simple:
People sign up. They upgrade. They find the love of their life.
Well, it’s not quite that simple. Some people will use it constantly, contact people, and go out on a few dates. Others will sign up for the site, use it a few times, forget they have a membership, and continue to get billed for it, sometimes for years. No kidding. Others will only return to the site when they get good or close matches via email. The final group will find their dream person and live happily ever after.
Therefore, it’s in their best interest to notify you of good matches. It gets you to come back to the site and probably spend more time looking around. As a result, you’re less likely to cancel your account.
But what happens if they give you a great match?
Let’s think about their model again:
If someone joins the site and is unsuccessful, the site loses one customer.
If someone joins the site and is successful, the site loses two customers.
That’s right. Making you successful can be bad for them. Now obviously, they need success stories to promote and recruit new customers but too many success stories and.. they lose.
As a result, there’s a sweet spot between giving good matches and great matches. This can be done honestly if the matching algorithm isn’t very good. It could be something as simple as not including all of your factors and requirements. Or alternatively, it could be done dishonestly by explicitly failing one of your requirements. Imagine getting a match that checks all your boxes but happens to smoke. Or someone that is nearly “perfect” but lives just a little too far away. It’s subtle but they’re effective at making the experience a little less effective.
Even when you are paying for the product, you may still be the product..
(For the record, I bailed a couple weeks after figuring this out. In our case, the matching engine was awful so we never had to “taint” our results. It worked on gender, location, and age but was a crapshoot beyond that.)