How do I validate an app idea?

The strength of a signal tracks the cost of giving it. A like costs nothing and proves nothing. Card details cost trust, attention and a small fear of being charged, which is why they predict revenue better than anything else you can collect before launch.

Interviews and a fake door test are complements, so run both. Ten interviews tell you which promise to put on the page. The fake door tells you how many strangers act on that promise once nobody is being polite to your face.

Nothing here needs our software. The whole thing is a landing page, a payment step that never charges, an analytics tool and $300 of ads. Two evenings of setup, five days of running.

• 01Write the promise as an outcome, not a feature list. One headline, one sentence of proof, one price. If you cannot say what the person has after using it, you are not ready to buy traffic.
• 02Build a one-page site. Carrd, Framer or a Next.js page you already have. Above the fold: the promise, an image of the thing, the price, the button. Keep the page under 2.5 seconds to load on 4G, because mobile visitors abandon slow pages before they read a word.
• 03Put a price on the page and mean it. Charge what you would charge. A page that says free tells you nothing you can bank.
• 04Make the button lead somewhere honest. Two options that stay on the right side of the line: a Stripe payment link that takes a pre-order you will refund if you do not build, or a Stripe SetupIntent that stores a card without charging it and says so on screen.
• 05Instrument four events. Page view, button click, checkout reached, checkout completed. Plausible, PostHog and Vercel Analytics all do this in an afternoon. Without the middle two you cannot tell a bad offer from a broken page.
• 06Buy cold traffic, not your own audience. Meta or TikTok, one campaign, three to five creatives, $30 to $60 a day for five days, targeted broadly in a country whose buying power matches your price. Send each creative to its own UTM so you can tell them apart later.
• 07Leave it alone for five days. Editing an ad set restarts its learning. Judge nothing before 1,000 impressions per creative.
• 08Read the funnel, top to bottom. Visitors, clicks, checkouts reached, checkouts completed. The stage where the drop is worst tells you what to fix.

Two funnels can show the same 2% and mean completely different things. The difference is how many visitors produced it. A percentage on 100 visitors is barely a rumour, and the interval below shows why.

Read that table before you set a budget. A funnel converting at 4% is called after 300 visitors. A funnel hovering at 1.5% will still be undecided at 1,000, because the honest range covers both a kill and a build. If your first read lands in the middle, the useful move is to change the offer or the audience and retest, spending more on the same undecided version.

• Testing on your own audience. Your followers convert at several times cold rates and there are only so many of them. Warm traffic tells you your audience likes you.
• Buying the cheapest clicks you can find. A $0.10 click from a country where your price is a week of groceries produces traffic that never intended to buy. Test where you plan to sell.
• Incentives. Giveaways, discounts on signup and prize draws all inflate the top of the funnel and destroy the meaning of the bottom.
• Counting clicks as intent. A click on your ad is curiosity about a sentence. Count the step where something is at stake.
• Slicing until something looks good. Cut 500 visitors ten ways and one slice will look brilliant by chance. Treat a promising segment as the next test, never as the verdict.
• Stopping the moment you like the number. Decide the sample size before you launch. Ending a test on the day it looks good is how a 4% becomes a 1.5% the following week.

These are the exact bars the verdict engine applies, imported here from the code that runs them.

Funnel starts before any verdict

30

Below this, don't build

1.0%

At or above this, build

2.0%

Visitors before a segment counts

50

Two rules sit behind the bands. No verdict is issued under 30 funnel starts, because a handful of visitors cannot decide anything. And confidence is statistical, so it reads high only when the 95% range sits entirely inside one band, which is the point where more traffic could not change the call.

VerifyToLaunch builds the test rather than the page. You describe the idea, it writes a six-screen funnel carrying your price, generates a few ad angles with creatives, and hands you a tracked link per angle. You launch those links in your own ad account, on your own card. Visits, funnel steps and checkouts land on one dashboard, bot traffic is filtered out, and the verdict is computed with the Wilson interval you saw above.

What it does not do: buy the traffic for you (there is no ad-platform API here, and we do not claim one), tell you whether people will still be using the app in a month, or rescue an offer nobody wants. If you would rather build the page yourself, the steps above are the whole method and they work.