We Like Humans

Everyone gets a
Fairy Godmother.

An AI superconnector for every human, wrapped in a community people actually want to join.

Investor presentation · August 2026 · Singapore

Connections make things happen.

A referred candidate is about four times more likely to be hired than one who applied. Referrals are 7% of applicants and up to half of all hires.

Zippia; ERIN (2025)

But only in cultures of vulnerability and generosity.

What a platform rewards "Humbled to announce…"

What actually gets you help "I don't know what I'm doing. Can you help?"

Showcasing invites applause. It does not invite generosity.

And now AI

Both sides automated. The signal is gone.

Job applications on LinkedIn	+45%
Job postings, same period	-10.6%
Applications per opening	242
Employers screening with AI by end 2025	83%

Tell the chain, do not read the table. One. Agents make it trivial to apply everywhere. They research the company and write the tailored letter. That looks good for applicants and it destroys the one way you used to stand out, which was doing real work. Real effort was attractive because it was visible. Now it is indistinguishable. Two. Applications per applicant go up, so applications per hiring manager go up proportionally. 242 per opening, a 0.4% chance each. Three. Drowning in volume, hiring managers screen with AI. 48% today, 83% by the end of this year. AI interviews went from 10% to 34% in two years. Those filters only see on-paper fit. Four. So the weird, diverse, brilliant candidate no longer gets through, and the route they used to have, researching a manager and reaching out, has evaporated. Five. Same pattern everywhere a platform exists. Every pitch deck is polished now. Every dating profile is optimised now. The signal that let you spot the thing that was not perfect on paper but was really great is eroding fast. Sources: eWeek, The Interview Guys, Resume Builder survey of 948 business leaders.

The fourth wave

Reach went up. Relevance went down.

Three waves so far, and each traded relevance for reach. Wave one: everything travelled through people you knew. High trust, high relevance, tiny reach. Your prospects were set by the village you were born in. Wave two: mass media. Enormous reach, one direction, no relationship. It could tell a million people a thing. It could not introduce two of them. Wave three: platforms promised everyone would be connected. What we got was reach without relationship, and a professional network people use to broadcast at each other. Wave four is the first one that gets both. An agent per person that knows what you actually need, talks to every other agent, and brings back the one human who can help.

The product

You tell her once. She works the network for life.

One

She gets to know you

Including the things you would never put on a profile.

Two

She talks to the others

Every Fairy Godmother compares notes with every other one, in private, at machine speed.

Three

She brings you a person

An introduction, or a seat at a table. One yes-or-no at a time.

Our data

People will not say out loud what they actually need.

Said at the table "Volunteering opportunities."

Said to her, in private "I'd like to meet a Muslim man, 44 to 50, based in Singapore. Emotionally mature, kind, intellectually curious."

60%

of attendees will not name what they asked for, even in writing, afterwards, in private.

11.8 : 1

private asks to public ones. 662 against 56.

This is the slide. Slow down. We capture asks in two places. Out loud at a gathering, and in private to a Fairy Godmother. So we can measure the difference, and nobody else can. 662 active private asks from 127 members, which is 5.2 each. Against 56 substantive public ones. And the 60%: of 141 people we asked afterwards, in writing, in private, what did you ask for at the table, only 56 gave a real answer. The rest said none, nothing, can't remember. In a room explicitly designed to force asking. That is the ego-risk tax, measured, in our own product. It is also precisely the information a matching engine needs and a platform can never get. One more thing if there is time: public asks are vaguer too. 50% of them are too unspecific to classify at all, against 18% of private ones. Rarer AND lower resolution.

Our data

The gap is widest where saying it costs something.

Share of asks about	Out loud	In private
Friendship	5.4%	21.5%	4.0x
Wanting a mentor	8.9%	23.6%	2.7x
Dating	5.4%	12.8%	2.4x
Raising money	1.8%	4.2%	2.3x
Looking for a job	19.6%	25.5%	1.3x
Looking for customers	12.5%	9.8%	0.8x

Scale

No human was ever going to read 3.2 million of anything.

2,541 members

3.2m possible pairs

22,053 pairs scored so far

Go to market

Nobody wants to be sold an AI product.

What we marketed	Cost per senior acquisition
A professional-sounding AI agent	$60+
Fairy Godmothers, on their own	$2-3
A community, with a Fairy Godmother inside it	under $2

Same product. 30x the cost of acquisition. And more than 10% of the senior C-level leaders we cold-invite in Singapore accept.

Why that is true

AI now polls worse than ICE.

AI	-20
ICE	-18
Donald Trump	-12
AI, among 18 to 34s	-44

And in the same poll, 56% had used an AI product in the last few months. People use it and dislike it. So do not sell it to them.

NBC News, net favourability, 27 Feb to 3 Mar 2026

The brand

We pitch the opposite of every other community.

Everyone else	Us
Likeminded	Unlikeminded
Curated	Anticurated
Exclusive	Inclusive

Being less exclusive doubled the share of the most influential people joining.

We ran the professional, exclusivity-flavoured brand first. It was called Atlas House. It worked okay and it did not spread. We Like Humans spreads. And the result nobody expects: switching to the more inclusive brand doubled the proportion of 9s and 10s on our internal influence score. Less exclusive attracted MORE senior people. The explanation is ad blindness. "Exclusive", "invite-only", "Atlas House" sound like they should appeal to senior people, and senior people get so many of them that they glaze straight over. "Fairy Godmothers", "We Like Humans", "Unlikeminded" cut through. If pressed on the doubling: it is an internal score, measured before and after the brand change, and I can walk you through the methodology offline. Be ready to give the denominator. The names are not set in stone. The vibe is right.

Where we are

Real people, real tables.

70 gatherings held

375 people at a table

4.65 average rating, out of 5

94% rated it 4 or 5

Replace with the real grid of members and hosts

Revenue

When she vouches for it, 30% say yes.

Getting an offer in front of the right person	Accepts
LinkedIn sponsored content, median click-through	0.5%
LinkedIn ads, typical conversion	2-4%
LinkedIn lead gen forms, their best mechanism	6-10%
A proposition her own Fairy Godmother approved	up to 30%

Three to five times the best thing LinkedIn sells. The filter is why.

The honest bit

Demand is priced. Supply is short.

One partner, one product

Will pay $50+ per lead, below what they spend on LinkedIn. Would spend $1,000+ a month.

The problem

Only about 5% of our audience fits their criteria and is interested. The audience is small.

pilot companies running boosts today. All willing to pay thousands a month. One in the tens of thousands.

Add the five logos, and a bounded monthly total

Pricing

We price the work, not the outcome.

We could charge

Recruiter fees

Thousands per introduction. It would work, briefly.

But

Anyone can undercut it

A fee unrelated to the cost of producing the match is a fee waiting to be beaten.

So we charge

For the effort

What the two Fairy Godmothers actually did. Today, low single-digit dollars.

More advertisers means more competition for her attention, which means more effort per match. The price rises on its own, and we never change the principle.

Do not say the words "cost-plus". It is accurate and it sounds like a regulated utility. The bit nobody works out for themselves, so say it: when acquiring a customer is this cheap, advertisers pour money in, because the returns are extraordinary. More money means more propositions competing for each Fairy Godmother's scarce attention. More competition means more attempts per success, which means more effort per match. So the price of a positive match climbs to whatever the market will bear, by itself, while we carry on charging for work done. If asked why we are leaving $47 on the table today: because the prize is the layer everyone routes through, and a standard only wins if it is too cheap to bother routing around. That is the next slide but one.

The race

Boardy is ahead of us. Their architecture is the problem.

Boardy · $11m raised

One superconnector

Easy to understand, and it means a single agent decides who gets the interview, the date, the pitch. Enormous incentive to lie to it, and concentration risk for everyone downstream.

One each, and a protocol

Everyone has their own, negotiating with the others. No kingmaker. We are building the protocol and the first superconnector on it.

They are also North America centric, and fundraising is 4.2% of what our members privately ask for.

Say "Boardy is ahead of us" out loud. Everyone in this room already knows, and naming it buys credibility for the rest. $11m raised, Creandum led the seed, founded by Andrew D'Souza who built Clearco. Serious people. Voice-call onboarding, and they have made 10,000-plus calls. Three differences, and the third is the real one. Niche: they focus on investors and founders. Our own data says fundraising is 4.2% of private asks. Mentoring is 23.6%, friendship 21.5%, dating 12.8%. They have picked the narrowest slice, and we can prove it is narrow. Geography: North America. The rest of the world is open. Architecture: a single superconnector is a kingmaker. If one agent decides who gets a tier-1 pitch, everybody has a reason to lie to it, and everybody carries concentration risk. A protocol between many agents is the fix. That is why we are building the protocol. Others: Profound.me is a career agent, "actors have agents". Gigi and a few more, all tiny. Note the pattern: almost all of them are selling professional-sounding AI representation, which is the positioning we measured at $60 a head.

The race

Why not LinkedIn. Why not the labs.

Their revenue comes from recruiters and advertisers, and both need a performed profile to sell against. Becoming a place where people admit what they need means devaluing their own inventory.

The AI labs

A lab cannot ship an anti-AI-fatigue product. An AI lab is the AI brand, and their distribution sits in the one channel we measured at 30x worse.

And if a lab does come: they will need a protocol to negotiate through. We would rather own that than race them.

LinkedIn, second leg: it is a performance culture, and vulnerability and generosity are the two inputs we need. Culture is slow; the revenue conflict is structural. The labs, in order of strength: One, the brand. Already covered, and it is our own data. Over $60 versus under $2. Two, the asset is disclosure, and disclosure is a function of trust, not model quality. Telling a Fairy Godmother from a community you have had dinner with that you are leaving your job, or lonely, or ill, is a different act from typing it into a general assistant whose parent is in the news about training data. Three, cold start is physical. Density in a real city means hosts, venues, gatherings. Local operations are the opposite of a lab's margin structure. Four, model conflict. The core feature is refusing to surface most propositions. A lab selling ads or seats has no reason to build that, and introducing strangers carries safety and liability surface they will take on slowly. Five, if they come, we want them routing through the protocol, so entry becomes adoption. Be honest about the counter. If a lab bolts "connect me to someone" onto 800 million users, the superconnector layer commoditises fast, and that is exactly when the protocol layer gets more valuable. Our nearer-term risk is not a lab at all, it is a well-funded fast follower who copies the human-brand insight.

The prize

Two layers. We want the one underneath.

Layer one

Superconnectors

Many companies will build these. We intend to be the largest, and a world with only one would be worse.

Layer two

The protocol

How agents negotiate privately, prove they have their human's permission, and pay to offset the compute another agent spends considering their proposition.

Every superconnector that ever launches has to route through something. That is where the money is.

Team

Built things that went viral. Ran things at scale.

CEO

Janakan Arulkumarasan

Apps and games used by 100m+ people, as CEO. Top producer of Facebook games two years running, five #1 chart-topping apps, one exit. His biggest game took 2m answers a day, which is the whole problem here.

COO

Arun Makhija

Former COO and CFO at foodpanda. Managed 250,000 riders and $5.5bn of revenue. Active angel investor.

Plus five colleagues, including a head of partnerships who ran partnerships for King Charles' charity, and engineers who have worked with Janakan for up to 15 years.

The credential that matters most is the least obvious one, so say it out loud: a question-and-answer game taking two million answers a day. This entire business depends on getting people to answer questions honestly, and he has done that at consumer scale before. Right to win is the intersection of three things, and most teams have one. Deep tech, especially efficient matching. UX, to get people to trust it and answer. And access to senior leaders. On that third one: Janakan and our partnerships lead also run Leaders.org, a global leadership network. Reaching senior people is battle-tested, not aspirational. We talk to heads of state. And we built all of this ourselves with agent workflows. Member app, admin console, matching engine, native iOS and Android. Which is also why the burn is small.

The ask

$500k in angel tickets, and a $1.5m pre-seed.

Where it goes
Team, including one senior AI hire	~50%
Member acquisition	~50%

What it buys

5 cities, saturated. 100k members in Singapore alone would be disruptive. At $4 per member that is 250k. At today's rate, 500k.

We plan against $4, not against the sub-$2 we see today.

The number we are chasing is not sign-ups. It is asks per member, because that is what she runs on.

Deliberately plan against a degraded CAC. $1m of acquisition spend at today's sub-$2 is exactly 500k members, and anyone can do that division in their head and conclude the plan depends on CAC never degrading at 250x scale. So say $4 and 250k, and let today's number be the upside. Naming your own weak point buys more than defending it. Then the harder truth, before they find it. Acquisition is not our real risk, activation is. 127 of 2,541 members have made a private ask. A Fairy Godmother needs asks, not sign-ups. So the number that governs the outcome is asks per acquired member, and that is what this round proves. Depth in five cities rather than a thin global spread. 100k in Singapore, or Dubai, or London, or New York, is enough to be genuinely disruptive on its own.

The bet

For all of human history, a great connector was someone you were lucky enough to know.

We are giving one to everybody, and wrapping it in an evening they would have come to anyway.

Appendix

Where the private asks come from.

The table

Six semi-random people, ninety minutes, life stories, and one round where everybody has to ask for something.

The card deck

Yes or no cards, three minutes. 2,879 answered so far. The no's carry as much signal as the yes's.

Her, directly

Conversation, and a nudge when an ask is too vague to act on. 5.2 asks per asking member.

Appendix · projection, not data

Consumer willingness to pay.

10-15%

of members we project will pay to boost at least one ask. This is a projection and not yet measured.