Demand evidence for founders

Most founders spend a year finding out nobody wanted it.
You can find out on Thursday.

Give us three ideas. In 60 seconds you'll know which one people are already searching for, what a real answer costs, and which to kill.

No card, no call, no account. Then if you want the full answer and we can't resolve it, you don't pay.

3
questions answered before you commit
~100
real people to an answer
$243
median media cost per test
0
decisions made by the AI
What it does

The whole loop, not a piece of it.

Step one reads the market. Step two runs the campaign and measures the response — lists, landing pages, surveys and analysis included.

Research
  • Keyword & demand research
  • Real ad costs per query
  • Competitor language
  • Subreddit & community mapping
Reach
  • Target audience profiles
  • Built & verified contact lists
  • Cold email campaigns
  • Google Ads in build
  • Reddit & Meta ads in build
Measure
  • Hosted landing pages
  • Variant randomisation
  • Choice-task surveys
  • Open-text harvesting
  • Audience screeners
Decide
  • Effect estimates per message
  • Funnel drop-off diagnosis
  • Your market's own vocabulary
  • Copy checked against results

Channels and data

Google Adsin build Redditin build Metain build Direct email Google keyword analysis Claude
The real cost

Conviction is the scarce thing now.

A weekend gets you a working product. A year gets you an answer. Founders keep spending the year to learn what a test could have told them first.

Build and see costs a year, and a flat launch never says which part was wrong.
Ask your network and they're encouraging. People who like you aren't the market.
Interviews capture what people say. Saying and doing part company when it counts.
Cakewalk measures what real buyers do, then tells you which part of the message did the work.
How it works

Two steps. The first one's free.

They answer different questions from different evidence. Most ideas can be killed or confirmed at step one.

Step 1 — Market scanfree · 60 seconds

Does the problem exist?

We read what your market already does — searches, asks and pays for — before anyone has seen your idea.

Evidence
What the market already does, at scale
Answers
Is anyone looking for this, what it costs to reach them, which of your ideas to pursue first
Takes
About a minute, no account

Can't tell you whether your solution is the one they'd choose. Nobody has seen it yet.

Step 1 in detail

What the market scan reads.

No black box. Exactly what we look at, and exactly where it stops.

What we check
Google search volume

How many people type this problem each month, and in whose words — yours or theirs.

What advertisers pay

Real cost per click on those terms. High bids mean the buyer is contested — a fact about your market, not your idea.

Twelve-month trend

Rising, flat or dying. A shrinking problem is worth knowing before you commit.

The long tail

Hundreds of related queries — your market's exact phrasing, and which are cheap enough to buy.

Reddit and communities in build

Where the conversation happens when nobody is searching yet.

App Store and Play search in build

Store demand and review mining, for consumer apps where Google is the wrong place to look.

Where it stops

The scan tells you whether a problem is articulated. It cannot tell you whether your solution is wanted.

Nobody has seen your idea yet. Search data is a fact about the market — that people have this problem and go looking for help. It says nothing about whether your answer is the one they'd pick.

Ideas with strong demand fail because the framing was wrong or the buyer was someone else. Ideas with no demand work, because the category didn't exist yet.

For that you have to put something in front of real people and watch what they do — a landing page, a campaign, real traffic. That's step two, and it's the only part that produces evidence about you.

Step 2 in detail

Not "variant 6 won". Why it won.

A winning variant is a lottery ticket. Knowing which ingredient did the work is something you keep.

Problem named Outcome promised Why it works Emotional angle
ONE RESPONDENT SEES THREE MESSAGES PICKED UPDATES ALL FOUR ESTIMATES Every message is built from the same four ingredients, in a balanced design — so one choice is evidence about all four.
An A/B test spends every respondent on one comparison. This spends each of them on four at once, which is why roughly a hundred people is enough to separate the ingredients — and why the answer transfers to copy you haven't written yet.
problem — 104 respondentsresolved
Wasted engineering time
+0.86
resolved
Gut feel over real data
−0.08
no effect
Falling behind peers
−0.78
resolved

Naming wasted engineering time beat peer pressure decisively. Stop writing the second one.

Why it holds up

Built to survive "how do you know?"

You'll make a build-or-don't call on this, and maybe raise on it.

01The AI never picks the winner

It writes variants and codes replies. A statistical model over real choices decides. Any tool that both writes and judges is grading its own homework.

02Claims are capped at the evidence

A test that measured attention says so. It won't call your idea validated, and the interface won't let it be screenshotted as if it had.

03"Not enough evidence" is an answer

Most first tests leave something unresolved. We say which, and whether more budget would actually fix it.

04No leaderboard while it runs

Watching a live winner and stopping when you like the number pushes false positives past 50%. Results unlock at checkpoints fixed before launch.

Pricing

A junior dev is $8,000 a month. A wrong year costs your runway.

An answer costs less than a week of either.

Scan
Free
  • Every idea you have
  • Real demand and ad costs
  • Ranked by cost to answer
  • 60 seconds, no card

No behavioural evidence — and labelled as such.

Scan now
The answer guaranteed
$499 once
  • We build a 10,000-person target list
  • We write and run the campaign
  • ~100 real buyers, four angles tested
  • Which message, which audience, their words
  • Media included up to $400

One idea, answered end to end. Three weeks.

Get the answer
Keep testing
$249 /mo
  • 10,000 contacts a month
  • Your channel stays warm — later ideas run faster
  • Language library and copy check
  • Cancel any time

For working through a backlog, not one idea.

Start the cycle

If we can't give you an answer, you don't pay.

Every test reports whether the evidence actually resolved — and when it doesn't, we say which part and why. If a test finishes without resolving a single dimension, it's free. We can offer that because we measure instead of guessing; a tool that always produces a confident answer has no way to know when it failed.

Straight answers

How long until I get a result?

The scan is immediate. A first test runs about three weeks, because a new sending channel has to warm up to reach inboxes at all. After that it stays warm, so later ideas start at full speed. Written replies from real people arrive within days.

Does this tell me if my idea is good?

No — be wary of anything claiming otherwise. It tells you which description makes a defined audience act, and how far that evidence stretches. A message can win for something nobody will pay for.

What if nobody's searching for my problem?

That's a finding. You'd be creating a category rather than entering one, so ads can't reach buyers at any price and the message has to teach instead of match. We flag it in the free scan and route to communities instead.

I don't have a list.

Good — you shouldn't use one. Each idea needs an audience matched to it, so your existing list is usually the wrong one. We build and verify a new one per test.

Can I just do the free scan and stop?

Yes, and plenty of ideas should. If nobody's searching and you weren't planning to create a category, that's worth knowing before you open an editor.

Sixty seconds now, or a year finding out the hard way.

Three ideas. No card. Kill the wrong ones today.

Scan your ideas