I launched a dating product. Building it was easier than getting the first users.

Thank you! I definitely agree that distribution feels like a completely different job from building the product :)

I’m already experimenting with social media, and guest posts/SEO are also on my list.
The pre-registration idea is interesting. I probably wouldn’t close REZYCO now because the product is already live, and at this stage I need to observe what people actually do inside it, not only whether they’re interested enough to join a waitlist.
But you’ve made me think about a variation of that idea: keeping the product live while also building small pools of interested users through different communities, and bringing them in together rather than one person at a time.
I’m actually starting to explore that now across different countries. I want to see whether REZYCO really needs dense local adoption first, or whether users with broader search preferences can create value even when they’re geographically distributed.

Thanks for the ideas — this whole thread is making me rethink my original cold-start assumptions :)

That was my instinct too, and it’s actually why I started looking at dating events.
But the more I think about it, the less sure I am that geographic density is the right unit of density for REZYCO.
An event could bring 200 people into the product and the model might still find only a few strong matches between them — or none. I don’t want to relax the filters or scoring just to make an event look successful.
At the same time, those 200 users wouldn’t necessarily be an “empty room.” They could still be relevant to users in other cities or countries who have chosen a broader distance range.

So now I’m curious about testing the opposite hypothesis too: what happens if I bring in users from different communities and different countries instead of concentrating everyone in one city?
Maybe for REZYCO the important density isn’t simply people per city, but enough mutually eligible people inside the search spaces users are actually willing to consider.
I don’t know yet — but that’s exactly what I want to test next.

That’s actually a very relevant question :)
The first stage happened before REZYCO was launched.
I couldn’t find an existing open-source dataset that matched what I needed for the compatibility model, so I created a dataset specifically for it. I split that data into training and test sets and used it to develop and test the model before integrating it into the product.

After the model was implemented in REZYCO, I tested the actual product myself using multiple accounts with different profiles, preferences and questionnaire answers. I also asked a few friends to help by creating accounts with different combinations of answers so I could test the matching flow with real users rather than only my own test accounts.
That part was actually quite funny because of the nature of some of the questions and the preferences we were choosing for the test accounts :)
So before launch, I had validated that the model behaved as intended on the test data and that the implemented product could process different user inputs, apply the filters, surface eligible candidates and produce Rezy Scores as intended.

What I haven’t validated yet — and what I’m really testing now — is the market side.
Will strangers actually value this kind of compatibility signal? Will they start conversations with people REZYCO surfaces? And can I bring enough of the right users into the product for those connections to happen?

That’s why the cold-start problem has become much more interesting to me after launch. Building and testing the system was one problem. Validating how real people will use it is a completely different one.
So I’d say the technical/model validation came first. The real-world product validation is what I’m doing now :)

This is really helpful — “density of intent” is probably much closer to what I’ve been trying to describe than geographic density.
The “tribal constraints” point is especially interesting because REZYCO was deliberately designed a little differently. I don’t define the tribe in advance — users effectively define their own search space through their preferences, including how much distance matters to them.

Candidates first have to fall within those user-defined boundaries and pass the relevant filters. Only then does the compatibility model evaluate and rank them.
Geography still matters, but it doesn’t automatically outrank compatibility. If two relevant candidates have the same Rezy Score, the candidate who is geographically closer to the seeker is ranked higher. If their Rezy Scores are different, the candidate with the higher Rezy Score is ranked higher, as long as both are within the seeker’s chosen distance range.

So geography is part of the ranking logic, but within the boundaries the user has already chosen.
I think you’re absolutely right that 10,000 global users could still feel like an empty platform if there isn’t enough mutual eligibility between them. At the same time, a much smaller geographically distributed group could potentially create value if enough of their search spaces overlap.

What I’m curious to test is whether REZYCO can create that kind of “density” without me defining the niche in advance — by letting users define their own boundaries and allowing the compatibility model to work inside them.
That makes “density of intent” a really interesting way for me to think about the experiment. Thank you.

That’s exactly what I want to figure out. Signups alone wouldn’t mean much to me. I want to see whether people actually complete the experience, get relevant candidates, and then do something with those results.
One metric I can already track is how often someone clicks “Message” on a candidate surfaced by the model. For an event pilot, I’d also want to see whether the pre-event experience leads to real engagement around the event itself.
So for me, 100 signups isn’t the goal — 100 people actually finding value in the experience would be much more interesting.

€150 CAC definitely makes me want to test organic channels first 😅 I’m still very early, so right now I’d rather learn where I can get real engagement before spending heavily on ads.

Which worked best for you — Threads, cold DMs, or build in public?

This gives me a lot to think about, because I agree with the core point, but I think the “unit of density” in REZYCO may be even more granular than city × age × orientation × gender.

Users define their own search boundaries through their preferences before Rezy Score is calculated. So two people can be in the same city, age range and demographic group and still never become candidates for each other. On the other hand, two people in different cities — or even different countries — can become relevant candidates if their chosen search boundaries allow it.
That’s why I don’t have an honest estimate for the smallest viable cohort yet. I don’t know that N=80, N=300 or any other fixed number is the right unit. That’s one of the things I want to learn from the early experiments.

I also agree with you about leading with the organizer’s metric. I’m not approaching organizers with “help me get users.” My current pilot idea starts after someone has already bought a ticket, with REZYCO as an optional pre-event experience. And rather than assume what matters most to the organizer, I want to ask them directly what would make the pilot valuable to them — attendance, participant feedback, repeat attendance, or something else.
Your point about day 8 is important too. I don’t see events as the permanent container for REZYCO. They’re one controlled way to introduce an initial group of users. REZYCO itself isn’t event-bound or city-bound — users decide how far they’re willing to search.

One clarification on the “people you know” point: I’m not recruiting friends, coworkers or people I know personally as a test cohort. When I talk about existing social circles, I mean them as a distribution channel — for example, someone sharing REZYCO with people in their age group who can decide for themselves whether they want to join.
REZYCO also doesn’t require users to disclose or verify their real-world identity. Accounts aren’t tied to a real name, workplace or organization, and private questionnaire answers aren’t shown to other users. Of course, that also means people can misrepresent themselves, as they can on other dating platforms. REZYCO has a reporting mechanism for situations where someone discovers that another user’s profile or stated information doesn’t reflect reality, but I don’t treat that as identity verification.
So if two users independently join, REZYCO surfaces them to each other, they choose to start talking, and they later discover that they already know each other, I see that as a different issue from deliberately recruiting a socially connected group to answer intimate questions for a test.

On measurement, I’m keeping it simpler at this stage. One concrete behavioral signal I currently track is whether someone clicks “Message” after REZYCO surfaces a candidate. Since users can only message people surfaced by the model, that tells me they chose to initiate a conversation with that person. I’m not currently measuring the 72-hour reciprocated-conversation metric you suggested.
And to your last question: I’ve tested that the matching flow works operationally, but I don’t yet have enough real-user data to claim that Rezy Score alone consistently makes strangers start talking without an event or another external context. That’s exactly the kind of behavior I want to understand from the first real cohorts.

I think your “unit of density” question is the part I’ll keep thinking about. For REZYCO, it may turn out that the useful unit isn’t a city or even a demographic slice, but the overlap between the search spaces users define for themselves.

It’s still too early for me to name a winner. REZYCO only recently launched, so right now I’m testing different channels rather than scaling one.
Interestingly, Indie Hackers has probably given me the most valuable feedback so far — not necessarily as a user acquisition channel, but as a way to challenge my assumptions about the cold-start problem.
For actual users, I’m exploring a few different paths: singles events, smaller communities, social networks, and geographically distributed users.
One of the behavioral signals I can already track is what happens after REZYCO surfaces a candidate — specifically, whether the user clicks “Message” and chooses to start a conversation.
So I don’t have a winning acquisition channel yet. I’m still testing where users come from and what they actually do once they’re inside :)

That’s an interesting angle, but there’s one complication specific to how REZYCO works.
REZYCO isn’t just a teaser — if two people are surfaced to each other, they can actually start communicating on the platform. So if I showed someone before buying a ticket that a highly compatible person was already attending, I’d have to ask the opposite question too: could that actually reduce the reason to buy the ticket? They might simply start talking and decide to meet on their own.
That’s one reason my first pilot idea starts after the ticket purchase. The organizer has already made the sale, and REZYCO then becomes an additional layer of anticipation before the event rather than a potential alternative to it.

I do agree with your bigger point, though: I need to understand what concrete value the organizer wants from the pilot, not just assume that attendee engagement is enough.
And I think that’s something I should actually ask the first organizer: “What would make a pilot like this valuable for you? What would you want to learn from it?”
Their answer might be attendance, repeat attendance, participant feedback, ticket sales — or something I haven’t considered yet.

So you’ve actually given me another useful question to bring into the first pilot conversation. Thank you!

Thank you — I think the intersection of individual search spaces is exactly the part I’m most curious to observe now.
I wouldn’t say the marketing problem disappears :) I still need to bring enough people into REZYCO for those intersections to exist in the first place. But once they’re there, I want to understand what actually matters more: geographic concentration, the size of the group, or the amount of mutual eligibility inside it.

And yes — if I get enough data to see a meaningful pattern, I’ll definitely write a milestone post about it. I think that could be a very interesting follow-up to this discussion.

Thank you — the event hypothesis is definitely one I want to test, especially because it gives me a concentrated group of people who are actually open to meeting someone.

But this discussion has also made me realize that I don’t want to assume geographic density is the only kind of density REZYCO needs.
In REZYCO, users decide how much distance matters to them. So alongside local event cohorts, I also want to test more geographically distributed groups and see what actually happens.

I’m starting to think the more interesting question may be whether there are enough mutually eligible people within the search space users themselves are willing to consider — rather than simply enough users in one city.

Your point about IdeaCrawl finding lack of local density as the #1 early problem for social platforms is really interesting in that context. I’d be curious whether you see the same pattern in products where geography itself is optional rather than fundamental?

I really like the idea of thinking about the first users as an ecosystem rather than a number.

I’m starting to realize that the interesting question for REZYCO may not even be “is 50 enough?” but what has to be true about those 50 people for the product to become useful?
They need enough mutual eligibility for the compatibility model to actually have meaningful candidates to work with.
That’s why I want to test different kinds of early ecosystems rather than assume there’s one correct way to create density — a concentrated event cohort, a university/social circle, and eventually more geographically distributed users who are open to matching across distance.

50 registrations and 50 people who can actually create value for one another are two very different things.

I think that distinction is becoming one of my biggest takeaways from this discussion :)

I agree that hyper-local seeding is probably one of the strongest ways to test the cold start — and I’m planning to experiment with exactly that through singles events and smaller concentrated groups.

But REZYCO has made me curious about the opposite experiment too.

Distance is a user preference, not a platform limitation. Someone can search within 50 km, or they can decide that distance doesn’t matter at all.
So I’m genuinely curious what happens if I compare, say, 100 users concentrated in one city with 100 users distributed across different cities and countries.

The local group has geographic density. But the global group potentially has a much larger search space for finding someone highly compatible — for the users who are willing to look beyond geography.

I don’t know which will produce more meaningful connections yet. That’s one of the things I’d really like to learn from the early users.
Maybe for REZYCO the question isn’t only “How many people are nearby?” but also “How far are people willing to look when they find someone they really resonate with?” :)

That’s interesting, because I’m actually experimenting with part of this already — although not by asking friends to measure their compatibility with each other.
My daughter is 20, and I’ve asked her to start sharing REZYCO with people around her age at university and at work. I’m even making a Russian version of the short REZYCO video so she can simply send people the video and the link rather than explain the product every time.
So I’m already testing the “people bring people they know” part of what you’re describing.
The “match with your friends first” part is a little different for REZYCO 😄 A significant part of the compatibility questionnaire is intentionally intimate, so discovering intimate compatibility within your existing friend group could become a rather different product.

There’s another reason I don’t want to think about density only geographically. In REZYCO, users decide how much distance matters to them. Someone can search close to home, but they can also say that distance doesn’t matter.
So theoretically, three people could join today from three different cities or even three different countries, and two of them could discover that they’re highly compatible and start talking today. Whether being on opposite sides of the world is a barrier is then up to them, not REZYCO.
That’s actually one of the things I find most fascinating about the product. If REZYCO grows, I’d love to understand what people are willing to reconsider when they discover strong compatibility. Does someone expand their preferred age range? Does another person decide that a national border or a few thousand kilometers isn’t actually the barrier they thought it was?

So concentrated local cohorts are very important for my early experiments, especially around events, but I don’t want to lose the other side of the hypothesis: compatibility itself may sometimes create its own reason to cross the boundaries people normally use in dating.

You mentioned you’re testing this on a compatibility product too — I’m curious what you’re building?

I think tracking the early users by cohort rather than treating them as one big signup number makes a lot of sense.
There’s an interesting wrinkle with REZYCO, though: density doesn’t necessarily have to be purely geographic.
Distance is one of the preferences users can set. Someone can look within 50 km, for example, but they can also choose that distance doesn’t matter. In that case, REZYCO can potentially connect compatible people much more broadly — even if they’re on opposite sides of the world.
So I’m starting to think about “density” less as simply how many users are in one city and more as whether there are enough mutually eligible people within the search boundaries users themselves choose.
For the first experiments, though, I still really like the event approach because it gives me a controlled cohort: people connected to the same event, in the same time window. That should make it much easier to understand what’s actually happening.
And I really like your point about separating cohorts before scaling distribution. Otherwise a total signup number could hide much more than it reveals.

Thanks — you’ve given me another useful way to think about the first experiments :)

Yes — that’s exactly what I’m hoping to learn from it.
The more I think about it, the more I see the first event not just as an acquisition channel, but as a controlled test environment for REZYCO.
If a group of people are in the same place, around the same event, and using the product within the same time window, the results should tell me much more than scattered registrations ever could.
And if it doesn’t work, that’s useful information too — I’ll have a much better chance of understanding why.
That’s one of the biggest things this discussion has helped me clarify :)

Thank you! 😄 It definitely is — but that’s part of the adventure.

Thank you! Definitely distribution.

Before launch, most of my attention was on building REZYCO and making sure the product and compatibility model actually worked. After launch, I realized that getting users is a completely different problem — and for dating, it isn’t even enough to simply “get users.”
You need the right people in the same place at roughly the same time. A large number of registrations spread across dozens of cities can create less value than a much smaller but concentrated group.
That’s probably been my biggest challenge — figuring out not just how to get attention, but where to concentrate it so REZYCO can actually demonstrate its value.
Right now I’m exploring singles events as one way to solve that: start with one concentrated group, learn from what actually happens, and only then think about repeating it elsewhere.

I’m discovering that shipping the product was one challenge. Shipping the distribution strategy is apparently the next one :)

I started by testing REZYCO with a small number of people I know. At that stage I wanted to answer the most basic question first: does my compatibility model actually work with real people?
That initial testing showed me that it does. The model was able to process real users’ preferences and questionnaire answers, surface eligible candidates, and calculate their Rezy Scores as intended.

The reason I built REZYCO in the first place is that I think there are important compatibility questions people don’t always feel comfortable asking each other early on — especially around intimacy. Not everyone is naturally comfortable having those conversations on a first date, and sometimes you only discover much later that you have very different expectations.
I included practical expectations too. Even something as simple as who expects to pay on the first date can become awkward if two people arrive with completely different assumptions and only discover that when the bill arrives :)

So the question I’m trying to answer now isn’t whether these differences exist. It’s how useful REZYCO can be at helping people discover relevant compatibility before investing time in a connection.
One behavioral signal I can already track is the “Message” click. Users can only message candidates surfaced by the model, so when someone clicks Message, it tells me they didn’t just receive a Rezy Score — they actually saw someone they wanted to talk to.

The next step is to test this with a larger, concentrated group and see what the behavior looks like there. I’m not trying to measure “successful couples,” because once two people connect they can move to another messenger, meet offline, or continue completely outside REZYCO.
So I’ve already answered one question for myself: the model works. Now I want to understand how much value it creates when more real people use it together.

Mostly pragmatism :)
I built REZYCO myself, so I wanted a stack that would let me move quickly without creating something I’d have to completely rebuild if the product started growing.
React + TypeScript made sense for the frontend, and I chose Python/FastAPI for the backend because it worked well for building the API and integrating REZYCO’s compatibility model. PostgreSQL was a natural fit for the structured data, with Redis for the parts where fast temporary state matters.
I also wanted the architecture to give me enough control over privacy and security, because that was a requirement for REZYCO from the beginning rather than something I wanted to bolt on later.
So it wasn’t really “this is the perfect stack.” It was more: what can I build and maintain myself, move fast with, and still have room to grow?
So far, I’m happy with that choice :)

Exactly. That’s probably been the biggest shift in my thinking since launching REZYCO.
While I was building it, the product itself felt like the hard part. After launching, I realized that getting attention isn’t even enough — it has to be the right attention, concentrated in the right place.
That matters even more for a dating product, because 1,000 scattered users can be much less useful than a much smaller group of relevant people in the same place.
So now I’m learning distribution almost like I learned product development — by testing, getting things wrong, and adjusting as I go :)

This is extremely useful — and interestingly, part of what you described is already very close to how I’m thinking about the first REZYCO pilot.
I want the organizer’s workload to be as close to zero as possible: no technical integration, just a ready-to-send invitation and simple instructions for attendees, while I handle the rest.
But I hadn’t thought about giving the organizer a one-page summary after the pilot. I really like that idea.
On measurement, REZYCO already gives me one useful behavioral signal: users can only message candidates surfaced by the model, so a “Message” click tells me that someone didn’t just receive a result — they actually wanted to start a conversation with that person.
And your last point about the composition of the room is especially interesting. REZYCO applies users’ preferences as filters before compatibility is calculated, so raw headcount alone may not tell me whether there is enough potential density in a particular event.
That’s definitely something I need to account for when designing the first pilot.

Thanks — this is the kind of practical feedback that is genuinely helping me think through how to structure the experiment.

That’s exactly the problem I’m trying to solve right now :)

My current hypothesis is to avoid spreading the first users across too many locations and instead concentrate them around specific cities and, even more narrowly, specific singles events.
Rather than trying to build density everywhere at once, I want to test whether REZYCO can create enough activity around one real-world event for participants to actually find relevant people and get value from the product.
I don’t know yet how many participants it takes for that to happen consistently — that’s exactly what I want the first pilot to teach me.

If it works in one concentrated group, I’ll have something real to learn from and potentially repeat. If it doesn’t, I’ll still learn much more than I would from having scattered registrations across dozens of cities :)

This gave me a lot to think about — especially the retention angle.
I’ve been thinking about events mainly as a way to create enough density before people meet, but I hadn’t really considered the organizer relationship from the “after” side in the way you described it. That’s interesting.
I also agree that signups alone aren’t a meaningful metric. I’m much more interested in actual behavior. For example, I can track when someone clicks “Message” on a candidate surfaced by REZYCO. Since users can only message people surfaced by the model, that gives me a much more concrete signal of intent than a registration number.
There are also limits to what I can — and want to — measure. I can see things like where registrations are coming from and engagement with messaging, but following what happens to a couple after that is much harder. They might start talking on REZYCO, move to another messenger, meet at the event, or continue completely outside the platform. And honestly, no dating product can guarantee what happens between two people :)

On trust and privacy, that has actually been a core part of REZYCO from the beginning. Registration doesn’t require a phone number or email, REZYCO doesn’t request precise geolocation, private conversations are end-to-end encrypted, and questionnaire answers are protected and kept separate from public profile information. Even I don’t have access to users’ private conversations.

But I really like your distinction between selling an organizer a “feature” and giving them a reason for people to come back. I hadn’t looked at the event partnership from that angle, and I’m going to think about it.
Thanks for taking the time to write such a detailed comment.

Yes — one thing surprised me quite a bit.

I expected the hardest part after launch to be simply finding users. But I’m realizing that the more interesting challenge is finding the right context for the product.
For example, I’ve started talking to singles event organizers. Some immediately see why compatibility before an event could be interesting, while for others it goes completely against the experience they want to create.
I actually find that fascinating. The same idea can fit naturally with one organizer’s vision and go completely against another organizer’s philosophy — and both approaches can make sense for the experiences they’re trying to create.

So I’m learning that early distribution isn’t just “How do I get more people?” It’s also “Where does this product naturally belong?”
I didn’t think about it quite that way before launching :)

Oops — apparently I decided to answer that one in Russian first 😂 Here’s the English version:

Honestly, I think I’d do pretty much the same thing again :)
I deliberately wanted to build the product myself and have something real and working before trying to convince people that the idea made sense.
REZYCO didn’t require a huge financial investment to build — mostly a very large investment of my time (and probably an unreasonable number of hours staring at my screen 😄).
With an unusual idea, I think explaining it before it exists can sometimes be harder than actually building it. People hear the idea and naturally imagine their own version of it — which can be completely different from what you have in your head.
Now I can simply say: “Here it is. Try it. This is what I meant.”
And I actually love that.
Distribution is a whole new adventure now, and I’m learning as I go. But if I started again tomorrow, I’d still build the product first.
Ask me again in a year though — I may have a much longer list 😂

Если честно, думаю, я бы снова сделала практически всё так же :)
Я сознательно хотела сама создать продукт и иметь что-то реальное и работающее, прежде чем пытаться убедить людей, что эта идея имеет смысл.
Создание REZYCO не требовало огромных финансовых вложений — в основном очень больших вложений моего времени (и, вероятно, совершенно неразумного количества часов, проведённых перед экраном 😄).
Когда идея необычная, мне кажется, что объяснить её до того, как она существует, иногда сложнее, чем действительно её создать. Люди слышат идею и совершенно естественно представляют свою версию — которая может полностью отличаться от того, что находится у тебя в голове.
А теперь я могу просто сказать: «Вот он. Попробуйте. Вот что я имела в виду».
И мне это на самом деле очень нравится.
Теперь привлечение пользователей — это совершенно новое приключение, и я учусь по ходу дела. Но если бы завтра пришлось начинать сначала, я всё равно сначала создала бы продукт.
Хотя спросите меня ещё раз через год — возможно, тогда у меня будет гораздо более длинный список 😂

I haven’t locked in the first city yet. I did a very small first round of outreach — around 10 emails to event organizers — but didn’t get any responses. At that point I basically only had the REZYCO website to show them, so I don’t think that was enough to learn much from.
Now I’m thinking more carefully about the offer itself. Your point about what the organizer gets is exactly the part I’m trying to solve. For attendees, the value is easier to explain: after buying a ticket, they could use REZYCO before the event and potentially discover people there they’re compatible with.
But for the organizer, “this could be useful for your attendees” probably isn’t a strong enough reason on its own. So before doing a bigger outreach round, I want to make the pilot offer much clearer from their side too.

That’s actually one of the most useful things I’m getting from this discussion — it’s making me rethink how I approach the organizers.

I really like the “one room, one night” way of thinking about it.
My idea is actually to test this with an existing singles event rather than organize the event myself — attendees would get access to REZYCO before the event, and then I could see whether the compatibility results lead to actual interest and interaction.
If I can make one event work first, then I agree — that gives me something much more concrete to take to other organizers. Thanks, this is a really useful way to frame the first test.

Honestly, I’m not there yet :) REZYCO only recently launched, so I’m still at the stage of getting the first users and testing different ways to reach them. That’s actually why I wrote this post — building the product turned out to be only half the challenge. I’ll definitely share what works (and what doesn’t) as I learn.

That’s a good point. I honestly don’t know what “enough” looks like yet — I think that’s something the first pilot should help me figure out.
Right now I can track when someone clicks “Message.” Since users can only message people surfaced by the model, that gives me at least one real signal that they saw someone they actually wanted to talk to.
So I’d rather learn from that behavior than make up a number like “50 people is enough” before I have any real data.

And I really like your point about treating the event, not the city, as the test unit. I hadn’t thought about it quite that way. Thanks for the perspective — it gave me a different way to think about the first pilot.