Interested in hearing more from me? |
The best consumer founders know the game. 13K+ of them read Consumer Startups every week.
Stay ahead. Get the playbook behind today’s breakout startups.
Read time: 4 min 52 seconds
—
What if I told you a startup has figured out how to make dating apps 50,000x more effective?
While Hinge and Tinder are busy optimizing for endless swiping, Jake Kozloski has quietly built an AI matchmaker called Keeper that's achieving what seemed impossible: 1-in-10 of Keeper’s matches result in a long-term relationship.
For context, traditional dating apps? Well… good luck swiping on 5 MILLION profiles to find the one.
Jake on Fox News after going viral
Quick teaser for today’s playbook:
Tactics Keeper used to grow from 0 to 1.5M sign-ups without paid marketing
How to build and optimize an AI matchmaking engine
Keeper’s product journey
If you're building in consumer tech, there's a growth playbook here you need to see—regardless of whether you care about dating apps.
Let's break down exactly how they did it…
A word from our partner
In 2 years you will be working for AI
Or an AI will be working for you
Here's how you can future-proof yourself:
Join the Superhuman AI newsletter – read by 1M+ people at top companies
Master AI tools, tutorials, and news in just 3 minutes a day
Become 10X more productive using AI
Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.
For Jake, this mission is deeply personal. His parents got a divorce when he was just nine months old, leaving him with a core childhood memory that would shape his future.
"I remember as a kid, just always having a strong feeling when I was around a family that didn't have a divorce, and just thinking I wish my parents were together and I don't make the same mistake that they did in the future."
Fast forward to college, he was among the first generation of young people to use dating apps. While he didn’t have too many issues getting dates from dating apps, he couldn't find someone he wanted to spend his life with.
This frustration sent him down a rabbit hole.
He read books, interviewed executives and engineers at Match Group, and became obsessed with solving the problem for himself.
Through his research, Jake discovered several critical insights about this market.
1/ Dating apps are ineffective at finding you your soulmate. They are optimized for engagement and perhaps casual relationships…
Let’s just look at the math below:
Overall, only 0.00002% of offered matches on dating apps lead to marriage. Yikes…
2/ Traditional matchmakers work.
They are orders of magnitude better than dating apps at finding you a good match, but they just don’t scale.
3/ The ideal solution should suffice three key criteria:
Accurately identify soulmate matches
Instantaneously
Within a large pool of people
When GPT3 came out, Jake had a lightbulb moment.
He realized he could help people find love at first match by creating an AI matchmaker that has the matching precision of a traditional matchmaker with a much bigger pool of people.
"The thesis is if you use language and vision models, you can automate what the traditional matchmakers do because all they're doing is aggregating as much data as they can in terms of who you are and what you're looking for. And then making matches based on that. At a fundamental level, that's what Keeper is doing."
In 2022, driven by this unique insight, Jake went all in on building Keeper, an AI matchmaker.
To get familiar with this industry, Jake and his team first built Keeper as a manual matchmaking service.
Day 1, they threw up a landing page, connecting it with a Typeform and a spreadsheet. They would make a profile on Notion for every user.
MVP
This manual phase allowed them to test and validate the core hypotheses about matchmaking, including one particularly contrarian: most people know what they want.
"The popular notion is that people don’t know what they want, but we’ve found that’s not really true. If you get people to be honest and take a moment to reflect, and then introduce them to someone who matches what they say they’re looking for, chances are, they’ll fall in love."
This insight, validated through manual matchmaking, became foundational to their AI approach.
The goal is to get people to provide their honest preferences and find the people that match those stated preferences the most.
As they gathered more data, Jake and his squad started creating a suite of internal AI tools to assist with that manual matching process, while also automating more steps on the user end.
This hybrid approach created a tight feedback loop between human matchmakers and the engineering team:
"This hybrid approach kept the feedback loop tight because matchmakers could correct and give feedback to our engineering team on where the AI was making mistakes." - Jake
The ultimate goal is to create a zero-shot matchmaking product that can help people find their soulmate with their very first match. They are not there yet… but not that far off.
1/ AI Performance Benchmark
Keeper’s internal AI benchmark
Keeper tracks how well their AI is performing with an internal benchmark.
The benchmark they use shows the size of the dating pool where Keeper’s AI, without any human help, can match you to your soulmate with a 1-in-10 success rate (the same hit rate a human matchmaker can achieve), assuming both you and your soulmate are in the pool.
Why focus on pool size?
If you’re in a pool of just two people (think: Garden of Eden), finding your soulmate is easy.
But if you’re trying to find them in a world of 8 billion people, it’s a lot harder.
In December of 2024, Keeper’s model could help you find your soulmate in 10 matches within a pool of 2,306 people. This number has been growing 2x-3x consistently every month, and in April 2025, the number has reached 158K+ 🤯.
“To put this in perspective: if this benchmark number is 2,500,000, then with a pool size of 250K, the model can get it right on the first try every single time.”
2/ Keeper’s AI secret sauce
Keeper's current AI system isn't a single model—it's an orchestrated collection of specialized models designed to scale matchmaking to millions of users. This architecture is designed to optimize matching performance but also decrease cost since LLMs can get expensive.
Under the hood, Keeper uses AI in two phases: first, for enriching and processing user profiles, and second, to use that data to find high-quality matches at scale.
The profile enrichment and processing system includes:
Multimodal LLMs: For enriching user profiles by inferring implicit details from profile text and photos, as well as AI sourcing of public user information
Computer Vision Models: For analyzing physical traits like facial attractiveness and body shape – physical attraction might be the most important factor for many people
LLM Preference Extraction: For producing a list of user preferences and associated importance levels
The matching system includes:
Basic Algorithmic Matching: For straightforward attributes like age and height
NLP Recommendation Models: For creating vector representations of user profiles to score matches at very large scale
LLM Matching: For final refinement of top matches
LLM Match Offering: For generating questions to resolve uncertainties about match quality, and crafting personalized profiles for offered matches
One interesting insight regarding LLMs is that GPT-4.1 is outperforming Claude and other reasoning-focused models for matchmaking specific tasks.
Combined max pool size based on the model
3/ Keeper’s business model: find love, pay later
Keeper’s business model flips traditional dating economics on its head.
Instead of charging users endlessly for swipes and subscriptions, Keeper makes money only when you find real love.
Users have two paid options: either pay a fee for each first date, or sign a “marriage bounty” contract promising to pay a bigger fee only if they find a life partner through Keeper (amount varies).
There is a free version as well. Free members are matched only with premium subscribers, whereas premium members can be paired with anyone—free or premium. Matches take a bit longer to appear for free users, but they generally do land one in the end.
This model aligns Keeper’s incentives with users’.
Keeper grew mostly via traditional Twitter marketing when they first launched the product in April 2022.
For example:
Keeper’s early tweets
By fall 2022, Keeper secured their first meaningful press coverage, which led to two key developments:
A significant uptick in organic traffic
Improved SEO rankings for key terms (e.g. AI matchmaking)
Twitter and early press coverage pushed them to 10K sign-ups towards the end of 2022.
In 2023, the team started experimenting with viral tools as a way to drive leads.
They spotted a viral tool on the internet called the Female Delusion Calculator, which pulled in hundreds of thousands of organic views each month by estimating a woman’s odds of meeting her ideal partner based on filters like minimum income, height, and other criteria.
The Keeper team realized there was an opportunity to create a much better version of that viral tool to get new users.
"We saw that website and thought that there was an opportunity to add more data to this tool and make the branding a little nicer.”
They launched the first version later that year called Standards Calculator and saw decent success, getting 5-10K new users.
Initial UI
Not satisfied with the results, they dove into deep user testing and conversion analysis, pinpointing and refining every churn point.
A few months later in early 2024, they did another launch… and it blew up!!
Second launch UI
Some metrics from the second launch:
10 million hits in a single week
Tim Ferris tweeted about it
Drew Barrymore discussed it on her TV show
1.5 million total sign-ups
200,000 completed profiles
Around the same time, they also started running UGC videos on TikTok and Reels and found success running street interview style content.
Keeper’s Reels
*Btw If you are looking to learn more about UGC, I highly recommend joining my friend Joseph Choi’s Viral App Founders Club where you will get more in-depth courses, community support on this specific topic.
Today, they have over 250K active profiles on the platform with close to 2M sign-ups.
Keeper’s breakout growth didn’t just catch users' attention — it turned investors' heads too. Jake and the team have raised $4M so far, closing their latest round just last year. They've already turned down acquisition offers from two of the biggest names in the space, staying laser-focused on building something bigger.
The current priority for the company is to launch their fully autonomous AI matchmaker app in the next few months and help more people find their soulmate.
For the 80% of Gen Z that wants to get married and the millions of others still looking for love, there might finally be a tool that can help them beat the odds.
Oh by the way, Jake is happily married now!!
Jake and his wife
Start with personal pain: Jake’s obsession with fixing dating came from growing up in a divorced family and struggling to find a life partner through traditional apps. The best startups often start by solving a founder’s own problem.
Do things that don't scale (at first): Before building AI, Keeper manually matched users through Typeform and Notion profiles. This scrappy phase gave them deep domain expertise and key insights that shaped their AI.
Find the broken incentives: Dating apps make more money when users stay single longer. Keeper flipped the model by optimizing for marriages, not endless engagement.
Innovate on existing viral tools for lead gen: Keeper did not reinvent the wheel on the viral tool - they simply made it 10x better
Relentlessly iterate until it clicks: Most founders give up too early. Keeper kept refining their viral tool and went viral with the second launch.
—
See you next Tuesday,
Leo
Thoughts on today's email? |
📥️ Want to advertise in Consumer Startups? Learn more.