Thoughts on an Optimal Dating App
It is clear to most people who have used dating apps that there is no incentive to actually match a user to their optimal partner. Whether the revenue structure is based off of ads or off of subscription, a user who is in a committed relationship is a user who no longer uses the app (committed being the operative word here). This is specifically relating to committed relationships, as an app whose user base is only interested in flings or one offs can successfully match users and still expect the user to continue to use the app. But for those looking for love, the success rate is probably much worse than typical methods, as even though a plurality of relationships have met online as of today, the absolute rate of people those had to swipe through is staggering.
I've had this understanding for a while now, but it is coupled with a hypothesis that many of my friends don't share. That is, a dating app that was actually incentivized to match you with an optimal partner could have an extraordinary success rate based off of known algorithmic matching methods. Some of you may know the stable marriage problem as a pet example of this. For those who don't know, in a 2 way matching situation such as dating, just from having each member order the members of the opposite group from favorite to least favorite creates a set of pairings in which every person is matched up with their best available choice that also has them as a best available choice. This algorithm guarantees no two people would both prefer each other over their assigned matches - it's provably stable.
This post is meant to just put down some of my thoughts on how one might choose to improve this problem. If you, like me, believe that love is towards the top of the list of important things, not using our mathematical and technological thinking to improve it is a shame. I am in a happy relationship, but I imagine a world in which all of my single friends get paired up better as being a much happier one.
The Payment Structure Problem
A network that only gets paid when you leave it would be a product whose profit is directly equal to your romantic success. Theoretically, if a service was structured based on the user paying a larger sum after a successful agreed upon conclusion, all dating app developers would be economically forced to use their best ability to make that happen. This is not dissimilar to the idea of a "marriage bounty" that has been floating around recently, ie "if you introduce me to the person that I marry, I'll pay you $X thousand dollars".
The main problem with this is how does it become enforceable? If the agreed upon claim is "I'll pay if I get married because of your app" a penny-pinching user could make the reasonable claim that they knew the person before the app, so the app was not the specific reason that they got married. They could enter into a life partnership without the legal designation of marriage. In all of these cases, a robust enough contract and good lawyers could realistically sort it out in the favor of the app, but then the company has to spend some amount of revenue on contract enforcement that could otherwise be put to other things. There is a possible solution here if you're bullish on blockchain, some sort of one click affidavit from both parties. But it is a big issue to consider.
##Interface Design and User Experience
On possible structures of an optimal app, an ordering function might be mathematically optimal but may be a horrible user experience. A hypothetical app might show you 10 people this week, and ask you to order them. Then at the end of the week you get matched with your stable marriage problem perfect partner. This may work, but I worry that specifically ordering people would lead to some strange interpersonal problems. Imagine a couple that met through this structure on a second date, when it turns out that you were their second to last choice. Even if it works out, it's a lot like someone telling you they thought you were ugly before they got to know you. Ultimately hurtful. Still, probably a better structure than current apps. An in between might be fair. Scoring users would give you more data than a simple yes no, and in this way starts to reflect some alternative voting methods in politics. Scoring someone on a 1-10 may also have some negative interpersonal effects, I think a nice middle ground of 4 inputs would be nice. Twice as much data per swipe interaction as typical swipe format, leveraging the familiar "swipe" motion of users. Tinder has a "superlike" function behind a paywall that is structured as a "swipe up" rather than left or right, removing the paywall and structuring a Right lean-yes, Left lean-no, Up strong-yes, Down strong-no, might be a good start.
Different styles of users might gamify this, using only the strong yes and no functions. While they may be attempting to game the system, this would actually only decrease the effectiveness of their sorting algorithm. Voting honest leads to better results.
Signal Processing and Algorithmic Matching
I don't know how dating apps work under the hood (other than that they used to use an ELO system), but if they are not already used, the same signal processing used in advertising tech could be leveraged here. A swipe in 4 directions is one thing, but how quickly that swipe happened, what image they stayed on the longest, etc, increasing the density and dimensionality that can be used to create an optimal match.
In addition to this, a calibration system may also be used. While one could go the unethical route and intermix faux calibration profiles into the normal stream of possible users, it would probably be most ethical to clearly label the calibration profiles at the beginning of the onboarding process, or every once in a while (although it would data-wise be best to just throw them in amidst real profiles so you can avoid the stated preferences problem). We have had some of the smartest minds in the world working on systems to show you the next most relevant thing, and not leveraging this for something good is a missed opportunity.
The Group Date Structure
We've touched on some of the optimal ways to show people, but what then? I know from the experience of myself and friends that most matches don't go anywhere. My first thought here is a blind date situation. You get told to meet at this area, you don't know which person it is that you matched with, you just sort of set possible times and locations and the app tells you when to go. This sounds structurally very good, but would probably freak people out a lot, especially women. Also, while it is important to get people to go on dates or meet in person as fast as possible, our theoretical matching of their profiles in the ordering system might not actually translate to in person attraction.
I propose we expand slightly out of just raw "dating app", and take a note from apps such as TimeLeft, which aim to do group blind dates but for friends. This is an excellent scaffold to build upon. 2 groups of 3 from each gender, who all have sufficient minimal possible attraction to the other 3 members of the opposite group, are scheduled to go on a big group date. Pre-date communication is opened through app messaging but only with the group you belong to. In this way, you're meeting with 2 possible friends and 3 possible romantic matches.
This can also mathematically translate to homosexual or bisexual relationships quite well, because functionally you are just trying to find 2 groups of 3 people who aren't likely to be romantically interested in their own ingroup, but are possibly attracted to all 3 members of the outgroup. The social sunk cost, the greater sense of safety in larger groups, the higher possibility of success, the lower stress of trying to carry a conversation on your own, all seem to be a greater instrument for connection than what we have now.
After a date, you could possibly rerank or just pick your favorite person you met there to better calibrate the algorithm. In addition, by adding in your ingroup equivalents, you can do what many streaming platforms do and find people sufficiently similar to you in taste and use that information to give you better results. In my mind you never actually get in app messaging with the people you go on dates with, as this incentivizes some trading (or refusing) communication routes at the date.
However, I can imagine a "missed connection" feature, which would be a limited resource where if you didn't manage to get someones number you can send them a request and if they accept you get a temporary message platform just to swap actual contact info. This creates a safety valve for when the algorithm gets it wrong but there was genuine chemistry with someone else in the group - it solves edge cases without undermining the forced in-person exchange that makes the format work.
Scaling and Revenue Model
Of course, not every city can handle this 3x3 structure. A graceful degradation into 2x2 double dates should be the case in groups/cities where the 3x3 just cannot happen. This forced blind date structure could theoretically allow for an alternative revenue stream as compared to subscriptions or ads. If you are putting 4 or 6 people at a venue weekly, through hundreds of pairs, a small kickback from venues could possibly fund this structure. Guaranteed butts in seats is more valuable than a quick pop up ad in between mindless gamified swiping sessions.
Venues may balk at 10-15% kickbacks initially, but consider: these are guaranteed customers during typically slow nights (weekday evenings), with higher-than-average spend per person since it's a date, and built-in repeat traffic. The economics work for both sides.
This model fundamentally isn't for people who treat dating as entertainment. It requires commitment - showing up, being present, engaging. That's a feature, not a bug. It's for people who actually want to meet someone. Of course the cold start problem is there. An initial start would probably be offering this for an initial 200-500 person beta, doing the swiping and ordering to get your 2x2 and 3x3 groups, and then renting out a venue and holding your dates there, with a possible party afterwards. Cover charge to offset some of your loss of this initial big party, but afterwards you're left with a starting user base, strong word of mouth, and provable numbers to show to venues to help with initial kickback contracts.
The first 50 people are the hardest. Start with personal networks - friends, friends of friends. Expand to one college or young professional community to build density in a specific demographic and location. Once you have proof of concept and testimonials, then open up more broadly.
The Efficiency Gap
To quickly compare this to something like Tinder: if we suggest 50-100 profiles per day, an average match rate of 1-2%, 10% match to meaningful conversation conversion, 10% conversation to in person date conversion, that results in .005-.02 dates per day of swiping. That's about 50-200 days of swiping for one date. These numbers feel inflated, so we might double or 3x them to illustrate that someone who is seriously interested in dating may get better outcomes, so say 16 to 100 days roughly to get a date.
As compared to this, where in the worst case where you can only get 2x2 dates every other week, you're still getting 1 date per 7 days of swiping, or in the best 3x3 once a week case, basically a date every other day.
If you believe love matters, building better systems for it matters too. The current model is broken by design. We can do better.