The two minutes after the meeting: how Scribr turns talk into action

Why the problem exists in the first place

Most of us spend our days in back-to-back conversations. Meetings, calls, voice notes to ourselves, recordings of lectures or interviews. We leave each one thinking we'll remember the important bits. We won't. The human brain doesn't work that way. It's too busy processing the next thing, or the Slack message that just came in, or what you're going to have for lunch.

So we try to take notes during the call, which means we half-listen and half-type. Or we take notes after, when we've already forgotten 60 percent of the detail. Or we skip notes entirely and hope something sticks. None of those outcomes are great. The worst part: even when you do capture something, it's in sentence fragments and your own shorthand. You have to decode it later.

The real problem isn't capturing the audio. It's turning that audio into something you can actually use.

How Scribr handles the transcription piece

Scribr starts simple. Hit record on your phone, have your conversation, stop. The transcription happens automatically. There are two routes here, and which one you use depends on who you are and what you need.

On the free tier, transcription happens on your device itself, using the same technology that powers Apple's voice-to-text. Your audio never leaves your phone. This matters for people handling sensitive information, or anyone who just doesn't want their conversations floating around in the cloud. We chose to build this way first because we knew some users would need it non-negotiable.

For longer recordings, or if you're on a Pro or Team plan, we use Deepgram for cloud transcription. It's faster and handles background noise better than on-device. You get to pick which you prefer. The key thing: once you have text, you have something searchable. You can skim it. You can quote it. But you still have to actually read it.

Where the summaries and action items come in

This is where the second layer kicks in, and it's the part that changes the actual workflow. A transcript is useful. A transcript with a three-line summary and a bulleted list of who's doing what by when is transformative.

When you enable AI summaries on a Pro plan or above, Scribr reads the full transcript and generates two things: a concise summary of the conversation (what was discussed, what matters), and an explicit list of action items (who said they'd do something, and what was that thing). The summary lives at the top of your note. The action items get their own widget on your home screen, and you get a push notification so you don't lose them to the sea of other notifications.

The process isn't magic, but it is useful. It's the difference between having text and having a usable note. Between remembering you talked about something and remembering the three specific tasks that came out of it.

This is only available to paid users because it requires compute on our end. You're not just storing something; you're asking us to process it. Free users get unlimited private transcription on their device. If you want the summaries and action items, that's a Pro feature (500 AI calls a month), or Team (1,500), or Enterprise (unlimited).

The consent question we had to get right

Early on, we had to make a choice about how obvious we made the fact that AI processing was happening. We could have made it completely transparent in the background, or we could have made it a deliberate choice. We chose the latter.

When you upload an audio file or use cloud transcription, you're asked to explicitly opt in to AI processing. You see it coming. You're not surprised later to find out your meeting was analysed by something you didn't know about. For therapists, journalists, legal professionals, this matters. For most consultants and sales people, it's fine. The point is: you know.

We also built Vault Mode (on Pro and above), which encrypts your notes end-to-end. If you record something truly sensitive, that's the setting you use. It means Scribr can't process it for summaries and action items, because we can't see it. But you can.

What actually happens when you use it

Let's talk about a real use case. A sales rep at a mid-sized software company records a discovery call with a prospect. Twenty-eight minutes of conversation about their pain points, their budget, their timeline, who the decision-makers are. She stops recording. Five seconds later, a summary appears: "Prospect has three teams affected by the problem. Budget approved but not allocated. Two key stakeholders involved. Timeline: Q2 go-live." Below that, three action items: "Follow up with pricing proposal (Rep)", "Get data on current system (Prospect's IT lead)", "Schedule technical demo (Rep)".

She can now skim that in five seconds instead of re-listening to the call or spending ten minutes reading a transcript. She can paste the summary into her CRM. She can see what she's responsible for because it's flagged with her name. If the team is on a Team plan, her colleagues can see the note and the action items, and Contact Intelligence will flag when the prospect is mentioned in other calls.

This isn't hypothetical. Customers have sent us screenshots of this happening. It saves them time. It catches things they'd otherwise forget. It changes how organised they are when they show up to the next call.

The limits we're honest about

Summaries aren't perfect. If the audio quality is poor, or the accents are heavy, or the conversation jumps around wildly, the transcription suffers first, and the summary suffers second. Action items are extracted from what was actually said, so if someone agreed to something verbally but never said it clearly, it might not appear. You still need to skim the summary and the action items yourself. We're not replacing human judgment. We're replacing the part where you sit there rereading a wall of text.

There's also a volume limit. Pro gets 500 AI calls a month. That's roughly 16 a day, or about 15-20 hours of audio a month depending on file length. For a solo freelancer or someone in a few calls a week, that's plenty. For a large sales team, that's Team territory (1,500 a month). We built it this way because processing isn't free, and we wanted to be transparent about what each tier gets rather than pretend it's unlimited.

One more thing: Android is scaffold only right now. We're a mobile-first app, which means we started on iOS, where we can ship what we promised. Android will come, but it comes when it's ready, not when some roadmap says it should.