The 5-upload mistake we almost made permanent
Three weeks after Scribr launched, a university researcher emailed to say she'd hit her monthly cap on day eight. Not angry. Just resigned. She'd recorded her lab meetings, transcribed them locally on her phone without paying us a penny, and then run out of the one thing she wanted to do more often: send audio to the cloud for our AI to pull out action items. That email changed how I think about the free tier entirely.
We designed it for the wrong person
The five-upload cap made sense in a spreadsheet. We were protecting our infrastructure. We thought we were discouraging the "free users who would never convert" from wasting our resources. In reality, we were filtering out the exact people most likely to become paid users.
Here's the pattern we missed. A consultant records a client call on her phone using our Quick Record widget. Whisper transcribes it instantly, on-device, no network needed, no privacy risk. She reads through it in two minutes. Job done. Free tier complete. But then she thinks: "What if Scribr pulled out the action items automatically?" That's when she hits the paywall. Or doesn't. She just moves on.
The five-upload limit wasn't stopping abuse. It was stopping curiosity. It was stopping the moment right before someone became convinced they needed to pay.
What free transcription actually costs us
When we built Scribr, we made a deliberate choice. Free users get on-device transcription via Whisper or Apple Speech. The audio never leaves their phone. It's private. It costs us bandwidth: zero. It costs us storage: zero. It's a gift that scales perfectly because it runs on their device.
The AI features are different. Summaries and action-item extraction run on Deepgram infrastructure. That costs us per call, per month. Pro users get 500 calls; Team users get 1,500. Enterprise gets unlimited. We have to be careful about that spend. But transcription itself? The thing we're most proud of? We can afford to be generous.
So why did we cap uploads at five? Honestly, I think we were overthinking the funnel. We wanted to make free feel like a trial, not a product. And then we realised: for some users, transcription isn't a trial. It's the whole product. A student recording lectures. A therapist recording session notes. A researcher recording field observations. They don't need AI summaries. They need the raw text. On their phone. Private. And as many times as they want in a month.
The gap between what we measure and what matters
We track conversion metrics like everyone else. I can tell you our Pro upgrade rate. I can tell you which users stay active after week two. But I can't tell you how many people tried Scribr, hit the five-upload wall, and decided it wasn't worth their time because we made something feel limited that had no reason to be.
The researcher's email made me ask a different question. Not "How do we make free users convert?" but "What does a free user actually need to decide we're trustworthy?" The answer isn't features. It's friction. Does the app work? Does it keep my audio private? Can I use it as many times as I want without hitting an arbitrary wall?
Five uploads a month felt like a safety valve. In practice, it was a speed bump on the path to trust. And trust is the thing that matters most when you're asking someone to record their thoughts, their meetings, their work. We were making it harder to discover what makes Scribr different: that your audio never has to leave your phone if you don't want it to.
What we're actually thinking about now
We're not removing the free tier. We're not removing all limits. That would be naive and it wouldn't be sustainable. But we're rethinking what free means. It means you can transcribe as much as you want locally. It means you get Quick Record and the Siri shortcut. It means your notes are yours, biometric-locked, never sent to a server unless you explicitly choose to upgrade and ask for AI.
The five-upload cap is still there, but now it's just for cloud uploads - the thing that actually costs us. And more importantly, it's presented as a constraint on the feature that requires our servers, not as a punishment for using Scribr at all. There's a difference. One says "try it out"; the other says "you're not important enough to use this for free."
We're learning that people want to use Scribr a lot more than we thought. Researchers conducting interviews. Freelancers recording client calls. Students capturing lectures. Therapists documenting sessions. They're not trying to build a habit. They're trying to do their jobs better. And they'll pay for the AI features once they trust that Scribr will handle their audio with respect.
The lesson isn't just about uploads. It's about understanding what free means to the person using it, not what it means to your spreadsheet.
When you design a product, you're always choosing whose friction to reduce and whose to increase. We chose wrong at first. The question now is whether we're choosing better. What would you keep free if you could afford the infrastructure and still sleep at night?
Ready to try Scribr by MRVL?
One tap to download. No sign-up wall.