User demand tells you a feature is wanted; unit economics tell you whether you can afford to sell it. At a consumer caller-ID app, around 25% of surveyed users said they would want VoIP calling, and we still didn’t build it. A flat “unlimited” price couldn’t cover per-minute billing, costly call destinations, and Google Play’s fee, and many AI products sold for a flat monthly fee face the same mismatch today.
The Feature That Made Perfect Sense
The app was already built around phone calls, so users were already thinking about calling inside the product. When we started exploring paid VoIP calls to regular phone numbers, it didn’t feel like a stretch. It felt obvious.
A survey confirmed the signal. Around 25% of users said they would want a VoIP feature. At the scale the app was operating, that was a real market inside an existing product.
The competitor plans we looked at were flat monthly bundles: one price, many countries, and “unlimited” minutes. Every surface-level signal said: build this.
Where Did the Unit Economics Break?
They broke in three places: per-minute billing increments, destination costs, and the app store’s cut.
The WebRTC calling layer cost €0.0028 per minute, billed 60/60, which means every call is rounded up to the next full minute. A 10-second call costs the same as a 60-second one, so ten 10-second calls would be billed as ten minutes, not 100 seconds. That’s a small number until you model it at scale with heavy users.
Then came destination costs: what it costs to complete a call on the receiving country’s phone network. The US looked manageable. Bangladesh looked manageable. Moldova did not. Users choose destinations by who they need to call, not by what the call costs the app, and in our model the heaviest callers, the people most likely to pay for a calling plan, could easily be the ones calling the most expensive places.
Add Google Play’s fee on top of that, and a low-priced “unlimited” package that looks attractive to a user looks very different in a margin model.
The problem wasn’t the feature. The problem was the promise. “Unlimited” is a word users love and unit economics hate. Matching those offers on the app’s cost base meant taking on a risk the team couldn’t close.
Is User Demand Enough to Justify Building a Feature?
No. Demand is the starting point. A feature becomes a business case only when the price you can charge covers what your users will actually cost you, heaviest users included. Before you promise a flat or “unlimited” price on a feature whose costs grow with use, run five checks:
- Model the heavy users, not the average one. An average hides the small group who will use the feature most, and they may be the people most likely to buy it.
- Read the billing increments. Per-minute rounding, minimum charges, and per-request fees change the real cost of every use.
- Price the mix users will actually choose. Destinations, regions, or models aren’t picked at random. Model the case where your best customers choose the expensive ones.
- Subtract every cut. App-store fees, payment processing, taxes, and refunds all come out of the price before it can cover your costs.
- Decide what you can honestly promise. If “unlimited” fails the model, test a capped plan or a higher tier, or don’t build. Not building is a valid outcome.
It’s the same discipline we use in paid user acquisition, where we set the numbers that end a channel test before spending the first dollar.
Why Does This Matter for AI Products Now?
Because many AI products face the same mismatch: token costs rise with use, while users often expect a flat price. On August 19, 2026, Stripe announced it had agreed to acquire OpenRouter, a gateway that helps businesses route AI requests across more than 400 models and manage token usage, for an undisclosed price that The New York Times reported at about $7.5 billion.
The gap between a variable cost and a fixed price is where margins disappear. It was true of calling plans and of cloud hosting, and it is true of AI features sold for a flat monthly fee. Every usage-based business that offers simplicity on top of complexity has to decide who pays for its heaviest users. The teams that come through it model the cost structure honestly before they make the promise.
What We Took From It
Sometimes the most important product meeting isn’t the one where you decide to build. It’s the one where someone opens the spreadsheet and the room goes quiet. We didn’t build the VoIP feature, and that was the right call: users wanted it, but the math said the promise would cost more than the product could bear.
User interest is a starting point. Unit economics are the finish line. You don’t get to skip the distance between them just because the demand signal is real.
If you are weighing a paid feature or a subscription change in an app, our mobile app marketing page shows how we approach monetization alongside acquisition and retention.
Sources: Stripe Newsroom, “Stripe agrees to acquire OpenRouter to help businesses optimise token routing and usage,” August 19, 2026. The Next Web, “Stripe seals its OpenRouter deal for a reported $7.5bn or more,” August 19, 2026, reporting The New York Times figure. The survey result and cost figures come from our team’s work with the app.
Updated October 23, 2026: we rewrote this post, re-checked its facts, and added sources.
