Justin McLeod

The through-line

Across more than a decade of talking about his work, Justin McLeod keeps returning to one argument: a dating product should be judged by whether it gets people into real relationships, not by how long it holds their attention. That conviction hardened into a company-defining choice in 2015, when he concluded that the app he had built was not aligned with his own values and rebuilt it from scratch [1]. The formulation he uses today is blunt: "we are the app that's designed to be deleted. We're not about engagement and retention and gamification. We're really about getting people who want to find their person off our app as quickly as possible" [1].

The mechanism he trusts to deliver that has changed while the goal has not. The earliest version of the idea was social: replicate the way people meet through friends, using second- and third-degree Facebook connections, with the real-name social graph supplying both quality and accountability [3]. The current version is computational: AI that understands daters deeply enough to match on values and long-term compatibility, and that coaches people through the parts of dating they are bad at [2]. He frames the AI shift as larger than the last platform change, saying "it's going to be a bigger transition than even what what happened with the transition to mobile" [2]. What connects the two eras is a distrust of superficial signals and a belief that the app's job is to reduce friction between real people rather than to become the destination.

On the 2015 reboot and knowing when to start over

The trigger was a Vanity Fair article, "The dawn of the dating apocalypse," about how apps had created hookup culture and ruined dating. McLeod says it was hard for him to deny much of what was in it: "this isn't really the company that I wanted to build. It's not having the success that I imagined it having. It's not aligned with me from a values perspective anymore" [1]. There was a business case too, and he was clear that the fix could not be incremental: growth had petered out and what was needed was "a kind of a reset" rather than a marginal iteration or a growth hack [1]. The push to act came from his Chief Brand Officer, Katie Hunt, at a lunch before Thanksgiving, when he said "I just want to start over" and she asked what was stopping him. He thought about it over the holiday, came back and told the board and his senior team the company had to start from scratch [1].

The execution was brutal and deliberate. He called founders who had been through something similar, including TaskRabbit, for advice. On announcement day the people being let go came in the morning and the people staying were asked to arrive around noon, so he could deliver the layoffs and then turn around and give an inspiring presentation about the future in the same building [1]. The company was 30 people; half were let go to cut burn rate, and every remaining person was offered the identical severance package if they did not want to come along. One took it and fourteen stayed [1]. To show strength, he took the whole team to a house in Connecticut for a few days to reimagine the product and the culture and to run a full diagnosis of where the company had gone wrong. He calls that the birth of the Hinge that exists today [1]. Not everyone agreed; some people, including board members, told him it was the wrong path and that he should simply start monetizing users [1].

He generalises the lesson into his sharpest statement about founders: "entrepreneurship is really the art of being both hopelessly idealistic and ruthlessly practical at the same time" [1]. Holding a big vision and refusing to settle for an app with 500,000 users has to coexist with admitting when things are not working and constantly iterating. He also says he would have welcomed an employee raising the alarm before his own realisation, and that Katie did exactly that by reflecting his thinking back at him: "any good leader would welcome that kind of big push and big thinking" [1].

On ignoring the competition

Watching rivals is the mistake McLeod says he made the first time around. Before the reboot, everyone at Hinge had every dating app downloaded, and each time a competitor shipped a feature the team scrambled to ship their own version. His conclusion: "if you're the smaller player, copying the bigger player is never going to, like, you're never going to be able to break out like that" [1]. So one of the foundational principles coming out of 2015 was a ban: "no one's allowed to have any of the other dating apps on your phone. We're only paying attention to our customers." If a competitor did something, Hinge would deliberately not do it [1].

He no longer enforces the ban, but the posture survives, because the differentiation is now structural rather than disciplinary: "we don't look at the competition because in a sense, like we are playing a different game" [1]. Asked about Toto Wolff's habit of having people print out a picture of their direct rival to stare at every day, he said he is "the complete opposite of that" [1]. Sitting inside Match Group alongside competitors he describes as "sibling rivalry" — some bumping, but the same family, all wanting each other to succeed, and a fair amount of genuine independence [1]. His broader read is that the space is not zero-sum: there is so much opportunity to serve customers, especially with what AI is opening up, that carving up territory makes no sense [1].

On culture as a document people argue with

McLeod treats culture as something written, contested and revised rather than declared. After the reboot he opened a Google Doc and began writing down every principle he thought should define the company, from office policies to how he made decisions to strategy, and left it open for comment so anyone could join long debates in the margins [1]. That same document kept accumulating comments for years as the company scaled, and only when Hinge reached roughly 100 to 150 employees did it stabilise enough to be locked, designed and printed as a book to hand to people [1].

The direction of travel is compression. Roughly 150 principles were distilled repeatedly down to five and then four, on a counterintuitive rule he learned with scale: "as we get bigger, I've learned that you actually need fewer principles for people to, so that everyone in the whole organization knows them and they're really clean and simple so that people know how to apply them," with sub-principles nested underneath [1]. Hiring is built to test that culture directly. Hinge runs three interviews: a skills interview for the role, a narrative interview covering history and track record, and a culture interview asking whether the candidate has demonstrated the company's values in prior roles [1]. His method is to take the attributes you actually want and ask behavioural questions about them; the authenticity question is "when have you actually spoken up when it was uncomfortable to do so?", because the company wants people to say what is on their mind even when the moment is uncomfortable [1]. The conversation notes a 94% employee retention rate at Hinge alongside this approach [1].

On planning cadence

He is not a continuous-planning advocate. Hinge runs an annual planning process, flowing into Match Group's own cycle for headcount, resourcing and strategy, and is not afraid to update it mid-year [1]. The value he sees is the interruption itself: planning continuously means you "kind of miss the moment to step back and think like, are we really heading in the right direction?" A semi-annual to annual cadence works as a forcing function, and the team always maintains a roughly three-year vision of where it is trying to go [1].

On what daters actually respond to

Pressed on the cynical view that online dating is 90% whether someone looks cute in their photos, McLeod concedes looks matter and are the easiest thing to snap-judge, but argues "a lot of people will make snap judgments based on photos but that's just foot in the door" [2]. Everything that determines whether you actually show up, get along and want a second date sits after that. He credits Hinge's success precisely to not being photo-based: you cannot simply like someone, you have to go through their profile and choose something specific about them that you liked [2]. The same design choice is what makes the opening exchange less generic than the match-and-go pattern elsewhere [2].

He also thinks hard about how much self-disclosure a format can carry, describing an "efficiency Frontier of vulnerability" [2]. A question like your go-to karaoke song is one everyone will answer and one that never leads to a date; a question about what you wish you could change about your relationship with your mother might produce a real conversation about values but is far too much to post publicly. The craft is finding the sweet spot where people are willing to share and what they share actually lets others in [2]. Format follows the same logic: video is too much, text is not enough, and voice prompts land in between, giving rich texture about someone's personality without demanding they record a video dating profile [2].

On AI as the next shift in dating

He sees two vectors. The first is personalized matching: getting to know daters much more deeply, going "two or three or four or five levels deep" so people do not discover three months in that they are misaligned on major values, goals or expectations [2]. Technically that means moving away from the collaborative filtering that drove the 2015 algorithm, which was based less on what Hinge knew about you than on your relationship to other users, and towards content-based models that actually look at your photos and read your prompts [2]. It also means being able to ask nuanced questions that never fit a multi-select dropdown like religion or height, covering backstory, what matters to you and relationship history [2]. The layer he is most excited about goes beyond taste-matching: using relationship science, the existing research on which personality combinations do well together and what leads to long-term relationships, to predict long-term compatibility rather than just the likelihood of an initial match [2].

The second vector is coaching, aimed at the large number of daters who struggle to get a first match or first date and do not know whether the problem is their photos or their prompts. His framing is forgiving: "none of us are born good at dating apps" [2]. He expects coaching to run the full gamut, from tips and tools at the start all the way to helping people navigate the emotional ups and downs of dating and keep a sense of confidence and hopefulness [2].

On coaching without writing for people

The line McLeod refuses to cross is generating content on a user's behalf. The principle is coaching, not answers: the product does not say "use this one instead," it says "hey could you say a little bit more about that" [2]. Early implementations include a photo finder built on models of what kinds of photos do well on Hinge, and prompt feedback, which he says tripled the incidence of high-quality prompt answers and cut low-quality one-word answers by more than a third [2]. The main thing that feedback pushes for is specificity, on the empirical finding that the more specific an answer is, the more it lets people in and the more likely it is to lead to a conversation and a date [2]. He was planning to start testing conversation starters the following quarter, again as nudges rather than scripts: pointing out something in the other person's profile that might be worth asking about, not supplying the message [2].

He takes the flattening objection seriously and rejects it on the grounds that the tool's direction is a choice: "AI is a tool it can be used to flatten it can be used to make everyone the same and tell everyone to put the same thing in the in the box or it can be used to like really find out what's unique about you and bring that out" [2]. His evidence is anecdotal but consistent with his product view: he knows dynamic, interesting people whose profiles completely miss who they are, and he does not think a badly built profile should be the barometer for whether someone would make a good partner [2]. On the accusation that this is personality airbrushing, he draws the same distinction: a small nudge to be a bit more vulnerable and a bit more specific leaves people in their own voice [2].

On designing for decency and safety

Some problems, he argues, do not need AI at all. Analysing responsiveness showed that much of the unresponsiveness on Hinge came from a small number of people who match with many others and then never reply, so the team shipped Your Turn Limits: once about eight conversations are waiting on your reply, you are blocked from sending likes or making new matches until you respond [2]. He reports responsiveness rising 20 to 30% as a result, and notes the rare pleasure of a positive Reddit thread and a one-star App Store review revised upward after the feature landed [2]. Romance scams he treats as a trust and safety problem where AI genuinely helps: a large dedicated team uses IP addresses, prior photos and behaviour patterns to catch bad actors, and he says it is not a big problem on Hinge [2].

An older version of the same instinct ran through the original product. Requiring real social connections and pulling identity from Facebook created transparency and accountability that anonymous profiles could not: your name, school, workplace and mutual friends were visible, and pulling multiple photos from Facebook made it harder to represent yourself with two or three unusually flattering pictures and waste someone's evening [3]. On data, he said Hinge kept user data in-house, restricted internal access to customer service and the people working on the algorithms, and did not sell it [3].

On the social graph and the stigma of online dating

The original idea came in his second year at Harvard Business School, when he was heading into management consulting and realised how hard it is to meet people when you travel constantly, while also having no appetite for long profiles and the rituals of traditional dating sites [3]. What he wanted instead was a product shaped like the way he actually met people: through friends and acquaintances. Hinge started with friends of friends and second- and third-degree Facebook connections, on the finding that relevance drops off sharply beyond the third degree, and that with an average of 500 friends the resulting network is still very large even after filtering for city, age and gender preference [3]. The system then learned your taste as you used it, the way a good friend who knew you would refine their suggestions [3]. He was explicit that this depended on a network where people map real-world connections, which is why Twitter-style follow graphs were not a candidate: following a celebrity does not mean you are connected [3].

He argued that this social layer made proximity secondary. Being within a two- or three-mile radius in a city like Mumbai tells you almost nothing about whether someone is right for you, even though people do need to be within reasonable distance to actually date [3]. He also saw the social graph as a substitute for the defensive machinery other services were building, the long exhausting forms and quality-control gates used to make the experience tolerable for women, because accountability arises naturally when everyone can see how you are connected [3]. Retention, he noted, behaved like a social network's: people leave when they get into a relationship and come back when they are not in one [3].

On the stigma, his view was that it came from the old model rather than from meeting people through technology. Traditional online dating was something you paid for and went through long forms for when you were struggling in the real world, which is what attached the stigma; using an existing social graph and location to make lightweight introductions changes the paradigm entirely, so it is no longer "for desperate people" [3]. He described early growth of roughly 20 to 30% a month, expansion from Washington DC to New York to 34 US markets plus London and Toronto, and marketing that stopped after about two months in a new market because word of mouth took over [3]. Hinge began as a web application in 2011, and around mid-2012 he concluded mobile was not the eventual future but the present, tore down the existing web infrastructure and relaunched in February 2013 [3]. India was always on his mind as a market, both because of its scale and because he believed the role of social connections and relationships in dating there suited Hinge's model particularly well [3].

Takeaways

• The decisive signal for the 2015 reboot was values, not metrics: a Vanity Fair piece on the "dating apocalypse" convinced McLeod that Hinge was no longer the company he wanted to build, and he chose a reset over further iteration [1].
• Run reboots with symmetry and speed: layoffs and the rally presentation on the same day, the same severance package offered to everyone including the people he wanted to keep, then an offsite to rebuild product and culture from scratch [1].
• His definition of the job: "entrepreneurship is really the art of being both hopelessly idealistic and ruthlessly practical at the same time" [1].
• Copying a larger competitor is a losing strategy for a smaller player, which is why Hinge banned rival apps from employees' phones after the reboot and now competes by playing a different game entirely [1].
• Culture principles should get fewer as headcount grows: roughly 150 principles were distilled to five and then four so everyone can recall and apply them, with hiring split into skills, narrative and culture interviews [1].
• The two AI vectors he is betting on are deeper personalized matching, including long-term compatibility informed by relationship science, and coaching that helps people who cannot get a first match [2].
• Coaching means nudges, never ghostwriting: prompt feedback that pushes for specificity tripled high-quality answers and cut one-word answers by more than a third while keeping people in their own voice [2].
• Product rules can fix behaviour that algorithms cannot: Your Turn Limits, capping unanswered conversations at about eight, raised responsiveness by 20 to 30% [2].