2025
Live Transcription for Calls and Conferences
eXpress is a corporate messenger, video conferencing platform and super app used by employees at large enterprises including Nornickel and Rosatom. It let you run and record conferences, but not work with what was said afterwards.
- 3M+
- Users on the platform today
- 88%
- Comprehension rate in testing
- Web · iOS · Android
- Platforms designed for
Product Designer. I owned the feature end to end, from concept to final layouts across all platforms, working alongside an analyst.
context
eXpress had no conference transcription at all. No call could be turned into a written record automatically, and there was no way to find a specific moment after the fact. The request came from two directions at once: from the product team, which saw the gap against the market, and from users, who needed to capture what was agreed without taking notes by hand. The feature was built for every platform at once: desktop, web and mobile.
problem
Users either took notes manually during the meeting or went back to the recording and hunted for the right fragment. Neither is a reasonable way to leave a meeting with a record of what was decided.
research
Competitive analysis
We looked at how Zoom, Google Meet and Microsoft Teams handle transcription. Three principles came out of it, and each one became a constraint on the design.
Recording and transcription are independent
In Microsoft Teams you can stop transcription while recording continues. Google Meet controls both through separate settings. Zoom treats transcription as a standalone saved meeting artifact.
A change of state has to be visible to everyone
Google Meet notifies participants when a meeting is being transcribed; Teams notifies everyone automatically when recording or transcription starts.
The output exists after the call ends
Across competitors, transcription is a post-conference artifact: saved, processed, and made available once the meeting is over.
The matrix of roles and states
The new function had to slot into an existing conference model that already carried roles, permissions and recording state. A conference has two independent states (recording and transcription), two roles with different rights over each, and several points where state can change: before the conference starts, during the call, and when a participant leaves. I mapped every combination to the notification or dialog it produces.
| Admin | Enable transcription | Disable transcription | Stop recording | Leave conference |
|---|---|---|---|---|
| Transcription off · Recording on | Notice: transcription enabled | None | None | None |
| Transcription off · Recording off | Notice: transcription enabled | None | None | None |
| Transcription on · Recording on | None | Notice: transcription disabled, link sent in direct messages | Dialog: keep transcribing the meeting? (continue / stop) | Dialog: end for everyone · leave and end recording and transcription · leave |
| Transcription on · Recording off | None | Notice: transcription disabled, link sent in direct messages | None | Dialog: end for everyone · leave and end transcription · leave |
| Participant | Transcription enabled | Transcription disabled |
|---|---|---|
| Transcription on or off | Dialog: a user started recording the meeting for transcription. Options: agree, or leave the conference. Then a notice that transcription is on. | Notice: that user turned off recording for transcription |
design
Planning
Transcription can be switched on ahead of time, while scheduling the conference. Participants joining such a conference are told transcription is active before they enter.
Admin controls
The admin controls both recording and transcription, and the status makes it unambiguous which combination is live right now.
Participant controls
A participant's rights are deliberately separated from the admin's. They can control their own availability to the transcript, but they do not get control over the conference's global recording state.
Session boundaries
Two cases were worked through separately: leaving with both recording and transcription active, and leaving with only transcription active. In both, it had to be clear what happens to the data and what notification each person receives.
Delivering the result
After the call the transcript arrives as a message. The output of the feature is as simple and predictable as receiving any other message in the messenger.
results
The main outcome was a systemic model for the new function. Validation was built around a single metric, comprehension rate: the share of test participants who correctly identified the current recording and transcription state in a given scenario, unprompted.
We tested 20 users across 5 scenarios: recording off with transcription on; both on; admin turned recording off while transcription continued; joining a conference already being transcribed; and a participant turning transcription off themselves. Each user had to answer two questions: what is being recorded right now, and is transcription running?
88% correctly identified the recording and transcription state without prompting. The errors clustered in the scenarios where recording was off but transcription was still running.