The recording problem nobody warns you about.

Strategy · 6 min read

The client said something that stuck with me. "My clients go to the doctor. They come home. And they can't remember what the doctor said." Not because they weren't paying attention. Because medical appointments are stressful, information-dense, and often rushed. For people with cognitive impairment, the problem is worse. The information goes in and doesn't stick.

The obvious solution: record the appointment. Use AI to transcribe it. Save a text summary the person can read later or share with their support worker. Simple. Useful. Everyone wants it.

Except it's not simple at all. And building it without understanding the complexity could get your users, and you, into real trouble.

The legal question that stops everything

In Australia, recording laws vary by state. In some states, a conversation can be recorded if one party consents. In others, all parties must consent. Recording a doctor without their knowledge could be a criminal offence depending on where you live. The Office of the Australian Information Commissioner has clear guidance on the collection of personal information, and a recorded medical conversation is about as personal as it gets.

That's not something you can hand-wave in a feature description. If your app has a "record this conversation" button, you're implicitly telling the user it's okay to press it. If it's not legally okay in their state, you've got a problem. Not just a usability problem. A legal one.

The client and I spent a full meeting discussing this. Not the UI. Not the button placement. The legal and ethical implications of storing audio recordings of medical consultations. That's the kind of conversation you don't see in a feature list, but it determines whether the feature ships or not.

The storage and accuracy problem

Even if the recording is legal and consensual, you still have to store it somewhere. Audio files are large. A thirty-minute GP appointment generates a file that needs cloud storage, backup, and potentially years of retention. That costs money. At scale, it costs a lot of money.

Then there's transcription accuracy. AI transcription tools have come a long way, but they're not perfect. Medical terminology gets mangled. Accents cause errors. Background noise degrades quality. If a user relies on an AI-generated summary of their doctor's instructions and the AI got the dosage wrong, that's not a minor bug. That's a health risk.

The question we kept coming back to was: can we transcribe to text only, without saving the audio? Some AI tools work this way — process the audio in real time, generate a text summary, and discard the audio file immediately. That solves the storage problem and reduces the privacy exposure. But it means the original recording is gone. If the transcription was wrong, there's no way to verify it.

What we actually built

For the MVP, we didn't build the recording feature. We designed the medical notes section with manual entry: what the appointment was for, who it was with, what the outcome was. Simple fields. No audio. The user or their support worker enters the information after the appointment, while it's still fresh.

That's not the final answer. The recording feature is on the roadmap. But it's on the roadmap with all its complexity acknowledged: legal consent flows, transcription accuracy safeguards, storage decisions, and clear user guidance about when recording is and isn't appropriate. Building it right will take more time than building it fast.

And that's the point. The MVP approach isn't about cutting features because you're lazy. It's about recognising which features need more research, more legal review, and more careful design than a five-week sprint allows. You don't ship something half-baked when the consequences of getting it wrong involve someone's health.

The lesson for your app

Every app has a feature like this. Something that sounds simple, tests well in a pitch deck, and falls apart the moment you dig into the details. Recording. Payments. Location tracking. Biometric authentication. The features that touch legal, privacy, or safety boundaries are never as straightforward as they seem.

The right approach isn't to avoid them forever. It's to scope them properly and build them when you're ready. Launch with the simpler version. Prove the core product works. Then tackle the complex features with the time, budget, and legal guidance they deserve. That's not cautious. It's smart.

Sources
Office of the Australian Information Commissioner - Guidance on collection, storage, and use of personal information.
Australian Government Attorney-General's Department - Overview of surveillance and recording laws by jurisdiction.

Related blog posts:

What is an MVP? →

Why we removed the photo upload →

Case study: NDIS life management app →