WORKFLOW PERFORMANCE · Published · Updated
Measuring session bandwidth in AI medical scribes
Uploading audio is only part of the data an AI scribe moves during a visit. The app also loads its own assets, keeps connections open while recording, and may send or receive more data while it finalises the note after the microphone stops.
Measuring from browser launch to browser close
Each measurement covers the whole isolated Chromium session, from launch to close. This captures products that upload during recording and products that move more data at finalisation, without choosing a time window that suits either.
Averaged per consultation across all 12 consultations, traffic was 9.1 MB for Hanah, 17.1 MB for Freed, 19.7 MB for Lyrebird, 30.8 MB for PatientNotes, 31.5 MB for CliniScripts and 44.6 MB for Heidi.
These figures describe the sessions we configured, and other visits will differ. They include background browser traffic, and profile caching, app updates and visit length can all change the result.
WebSocket traffic
We read socket and UDP byte events from Chromium's NetLog. Counting bytes at the transport layer includes WebSocket traffic without counting the decrypted payload a second time. If a NetLog is missing, the run is recorded as missing data.
The figure will not match an internet provider's bill exactly, because packet headers and TCP retransmissions are not included.
Bandwidth and speed
Lower data use can help on a slow or capped connection, but this test does not measure performance under bandwidth limits. Latency, processing time and connection quality also affect speed, and we report data use and timing as separate measures.
Frequently asked questions
How much data does an AI scribe use per consultation?
Averaged across 12 consultations, traffic was 9.1 MB for Hanah, 17.1 MB for Freed, 19.7 MB for Lyrebird, 30.8 MB for PatientNotes, 31.5 MB for CliniScripts and 44.6 MB for Heidi. Each measurement covers the whole browser session, from launch to close.
Will the figure match my internet bill?
Not exactly. Packet headers and TCP retransmissions are not included, and profile caching, app updates and visit length can all change the result.
Does lower data use make an AI scribe faster?
Lower data use can help on a slow or capped connection, but this test does not measure performance under bandwidth limits. Latency, processing time and connection quality also affect speed, and we report data use and timing as separate measures.
Figures in this article come from the ScribeStandard benchmark. See the full results, each test and how we test.