I'll shape the sections: a "case at a glance" summary with number bullets, a secondary keyword H2 explaining what's being sold, a paywall break, a CTA linking to the full issue mentioning founding membership, and three related case links with plausible slugs — keeping the tone direct and teaching-focused throughout.
---
title: "How an Offline iOS Transcription App Made $7,591 in 28 Days"
meta_description: "An offline iOS transcription app did $7,591 in 28 days — but only $1,295 recurs. Why the on-device engine was really a pricing instrument."
slug: "/free/offline-ios-transcription-app-7591-28-days"
canonical_issue: "/issue-073.html"
keywords: ["offline iOS transcription app", "on-device Whisper Core ML", "indie iOS app revenue case study", "solo developer MRR"]
---
One developer, one offline iOS transcription app, 258 active subscriptions, and zero cloud API bills. The thread argued about whether the engineering was actually hard. Wrong argument — the interesting thing is what the engineering did to his pricing page.
The case at a glance
- 28-day revenue: $7,591
- MRR: $1,295 — meaning roughly $6,296 came from one-time purchases that won't repeat
- Active subscriptions: 258, or about $5 per subscriber per month
- Marginal cost per hour transcribed: $0
- Paid channels running: none (Apple Search Ads tested, killed as unprofitable)
- Team: 1 — development, support, updates, marketing
He entered a category that was already crowded, which meant the easy path — wrap a cloud API, polish the UI — was taken. So he built his own Core ML inference engine to run Whisper Large V3 Turbo on-device, refusing to downgrade to a smaller model just to fit it on a phone. That bought a window: phones got fast enough to run a frontier-grade speech model locally, and almost nobody wanted to do the memory, heat, and battery work to prove it. Product: on-device transcription, speaker separation, and structured notes from meetings, lectures, and interviews. Pricing: subscription plus a one-time lifetime license.
What he's actually selling (not the on-device engine)
The product isn't the transcription. The product is a permission slip.
Someone recording a therapy session, a client call, or a divorce mediation is not shopping for word error rate. They're shopping for "this file never leaves my phone." He spent months on inference and sold it as speed — then discovered it converts as consent. Nobody buys Whisper Large V3 Turbo. They buy a readable transcript of Tuesday's call with names attached to the lines.
The mechanism underneath is cost structure. Every cloud-API competitor has a meter running against it: a customer drops in a four-hour deposition and the founder loses money on that account. So competitors write usage caps and hedge their promises. He has no meter, which means he can promise unlimited and mean it — and, more importantly, he can offer a lifetime purchase at all. A subscription asks a stranger to commit to a relationship. A one-time price asks them to make one decision. On the App Store, where visibility rewards listings that convert browsers into buyers, that second option isn't a pricing nicety. It's a ranking instrument that his competitors can't copy without bleeding.
His inference engine isn't a feature. It's the thing that pays for his distribution.