A well-known B2C app through the full-funnel lens: what's excellent, what misfires, what I'd test. Public data, no cheap shots.

The headline stats: downloads have slid from an estimated 32 million in 2020 to 7.3 million last year. Revenue peaked around $355 million in 2022; last year's estimate is $210 million. Subscribers: about 3.5 million, from a peak near 5 million. Topline numbers still solid, trajectory all wrong. Valued at $2 billion in 2020; no priced round since has tested it. Six years without a repricing is its own verdict.

What is excellent

The brand. Calm built the category, made Sleep Stories a cultural object, and put Stephen Fry on millions of pillows. Brand of that strength is a renewable asset almost no consumer app owns, and it's the only reason the funnel below still works at all.

Misfire one: the app store sells features, not outcomes

ASO 101. One screenshot headline gets it right ("Sleep more. Stress less. Live better."), then it's a wall of features: Sleep Stories, Music, Masterclasses. And the proudest badge on display is App of the Year 2017. Nine years ago. When your store front leads with a museum piece, acquisition is running on brand memory.

Part of the download slide is channel economics rather than creative: post-ATT, paid acquisition maths turned against every consumer app at once, a story we covered in the channel selection post. But that makes the owned surfaces matter more, not less. When paid gets expensive, the store listing is the channel you don't pay per click for.

Misfire two: the onboarding

I walked it myself, screenshots in hand.

First screen: a questionnaire about your stress and goals. A stress-management app whose opening move is homework and option-anxiety. The crazy part: reworked as an adapt-to-you feature (five sliders, a user type, tailored suggestions), the same questionnaire would be a wow moment instead of a dull form.

Second screen: create an account. So much for progressive disclosure. "Hi, I'd like to test your product." "First I'll need your name, your email, and an account." Major turnoff.

Third screen: the paywall. £49.99 a year, or £18.99 a month (£227.88 a year). Zero minutes of value delivered, and they want your card. Who does this?

The famous stat, from the founders themselves: a third of Calm's subscribers convert on day one. That's like a paid car park boasting a 99% conversion rate. Yes, because you can't leave without paying. A hard paywall harvests the high-intent arrivals the brand already convinced, and burns everyone else. When 32 million people a year walked through the door, burning the fence-sitters was affordable. At 7.3 million, it isn't. The maths that justified the toll booth died with the traffic.

Two teardowns ago: Duolingo. Value first, commitment second, still the reference onboarding. Calm runs the inversion. Commitment first, value second, and the response to a narrowing funnel has been to raise the toll (a price increase in February, a second subscription in September), not widen the gate. MyFitnessPal shows where that road ends: tax the loop that creates committed users for long enough, and someone builds the bypass.

What I'd test

One complete session before any paywall. A full Sleep Story, then the ask, with the questionnaire personalising instead of gatekeeping. Measured on cost per retained subscriber, not day-one conversion, because day-one conversion is the metric that built the toll booth in the first place. Optimise the toll and you get a better toll. You don't get more traffic.

My guess: that simple test lifts onboarding conversion 30-40% and paid conversion 10% within a quarter.

Steelman the toll booth

Because it is real: the hard paywall is not stupidity, it's a model. Day-one conversion harvests high-intent arrivals at near-zero incremental cost. Calm's content is expensive (celebrity narrators are licensed, not free), so giving full sessions away has a real bill attached. Free tiers invite trial-hopping in a category where the competitor is one download away. And the strategic centre of gravity may simply have moved: the Calm Health push into employer and payer channels suggests a company monetising B2B distribution while the consumer funnel coasts on brand.

But every leg of that case depends on the same input: traffic. Harvesting intent is a strategy for when the brand manufactures more intent than you can serve. The download curve says that era is ending. A model that burns nineteen fence-sitters to convert one believer is affordable at 32 million doors a year and ruinous at 7. The steelman was true in 2020. The funnel stopped agreeing with it.

If you run a consumer product: where in your funnel do you ask for the money before you've shown the goods? And do you know what that gate costs you, or only what it collects?

Fix the leaks. Fix your world.

Product onboarding is one of eleven modules in the Revenue Engine diagnostic, and the join between acquisition and onboarding is where a hard paywall quietly reprices your entire funnel. If any of this resonates, our diagnostic maps your entire growth funnel and shows you where the real leverage sits. No deck. No pitch. Just clarity on where the engine is misfiring.

FAQ

Why does Calm charge before showing the product? Because it converts a third of its subscribers on day one, a stat the founders have cited proudly. It works as long as brand-driven intent keeps arriving in volume. The teardown's argument is not that the paywall never worked; it's that the traffic that justified it has fallen by roughly three-quarters.

Is a hard paywall always wrong for subscription apps? No. It's a bet on intent volume. If your brand or channel generates more high-intent arrivals than you can monetise, harvesting them is rational. The failure mode is keeping the toll booth after the traffic dies, and reading day-one conversion as health while the funnel narrows.

What's the alternative to paywall-first onboarding? Deliver one complete unit of value first, then ask, with any questionnaire doing personalisation rather than gatekeeping. Measure the change on cost per retained subscriber and 90-day retention, not day-one conversion. Duolingo has run this playbook for a decade.