compatibility learnings

the experiment setup

a few months ago, a friend and I joked about hosting a matchmaking night. people spend hundreds of hours and thousands of dollars on dates to try to find their person, so what if we tried to help? one thing led to the next. we found a gorgeous mansion in the upper east side and a group of friends who were crazy enough to join as staff.

we wanted to matchmake ahead of time, so we launched a site with a google form, which we shared in a tweet that reached 42k+ people. soon, we had 100+ responses of NYC-based young adults searching for someone “kind,” “intelligent,” “ambitious,” with traits ranging from “can pick me up easily” to “best friend energy.”

after closing form submissions, the team convened for a matchmaking marathon and we quickly realized we were in over our heads. GPT was surprisingly powerful at generating matches but we didn’t trust it fully for such a high-stakes task. we decided perfect matches couldn’t be made, and even if they could, it would take hundreds of hours of manual labor.

regardless, I was set on attempting to match everyone, so the two weeks before the event, I spent hours perusing applications and sorting names, preferences, ages, heights, etc. on sheets. I got to know each and every applicant and formed a list of each of their match preferences. then, I roughly ran a manual version of the gale shapley algorithm, matching each of the gals with their top free preference and replacing guys if I found that there was a pair who would prefer each other over their current matches.

the results

the matchmaking night was a great time - the house was brimming with warm conversation and genuine laughter, and people were exchanging contacts and chatting until being ushered out so that we could clean. however, I was somewhat off on the matches. some couples formed unexpectedly, while matches that I was confident about didn’t end up engaging at all. I learned a few things:

1. 50% of attractiveness is in how people carry themselves, and doesn’t come through at all in pictures. how people walked into the room, or how they would touch arms gently to emphasize points, or how they simply appeared at ease. seeing people come to life from their google form application was always a shock for me.

2. surface values don’t matter, core ones do. in their applications, people described their love for food, or their career ambition, or their appreciation for humor, but this doesn’t get at the primitives. how do they view life and the world around them? I realized I agree with the study findings on political alignment in couples; political beliefs are the most direct manifestation of the core values that determine compatibility.

3. turn-offs are more important than turn-ons. as I was observing and reflecting, I realized there’s probably certain telltale signals in people’s first conversations that cause them to end the conversation early. hasn’t asked enough questions, too uptight and not enough banter, doesn’t carry the conversation. below the surface, I think we need partners who make us feel safe, and a turn-off could be a sign that they’re not equipped to do that. personally, one of my insecurities is not being funny enough, and I feel safe around people who make me feel funny, so I’ll get the ick if I receive a straight face after I’ve cracked a joke.

when I look at some of my friends in happy relationships, it makes total sense why they’re together. one of my friends and her partner share the same rational view of the world and a love for dissecting details about people and things. her mind is always racing about something but he grounds her with his calm, content energy. another friend and her partner both have hectic, ambitious startup careers but anchor each other by being silly and going to llama farms together. there’s so much nuance in the intersection of values required and the buckets each partner needs to fill for each other.

this was a fun experiment, but it was also a lot of pressure to be responsible for people’s love life. I do think matchmaking is possible with the correct datapoints, and maybe Hinge AI will get there at some point. in the meantime, I’ll be wingwoman-ing my friends at the bar as we continue to search for our soulmates.