The Transcription feature steers both live captioning and after-call voice transcription capabilities in Nimbus. It uses Speech-to-Text (STT) technology for converting spoken words into written text. Areas of use are:
- Showing transcription / summarization contents within My Sessions and Attendant Console.
- Providing transcription, summarization, live captioning, and AI-steered Codes and Tags suggestions for both inbound and outbound audio calls.
- Transcribing and collecting “silent” interactions between Customers and Virtual Users, collecting the data for further processing within Trigger Events and Flow Actions within the Nimbus Power Automate Connector.
Live Captioning in the UI
Preconditions
- To use Live Captioning functionality, Transcription (as part of the Companion license) must be set up first → See general preconditions above.
- Note that Live Captioning is not part of the Companion base license and will require a Virtual Assistant license in future.
Live captions are audio transcriptions generated real-time during call sessions. When a call is accepted, a bot is invited to the call in order to start the audio transcription of the session. As a user, you see your spoken conversation with the caller in the Live Caption widget. Live Caption is only visible during ongoing calls, visible in either My Sessions as "Companion" widget and within the sidebar of Attendant Console .
Transcription in the UI
When the call session is ended, the live caption is saved as a transcript. This process may take a bit, depending on how long your call with the customer was.
⮑ Once the final transcription is processed, it appears in the Companion widget.
💡Good to know:
Transcription access
- Within My Sessions you can also access past session transcripts by clicking on a concluded session. If available, the transcript of the session is then opened in the "Companion" widget. Note that a transcript will only be created when the feature was enabled before the Nimbus session started.
- Transcription features are also available in Attendant Console within the sidebar.
- When you have After-Call Work (ACW) enabled on this service, you may see a message: "Your transcription is currently being processed and will be shown shortly" because the session is still technically not "done" yet. This message may also be shown on very long transcripts that are still being processed.
- Note that once the feature is enabled on service level, Transcription is always enabled for Nimbus users, as long as they have the "Companion" license.
Data access and Retention
- AI Service Settings steer the Transcription feature availability to users, as well as the data storage retention time.
- You can also opt-in to and show a Summarization of the transcript. The same access and storage rules apply.
- All data generated by Transcription can be retrieved and handled by Nimbus Power Automate Connector, using respective "Companion" Flow Actions and Trigger Events.
Transcription for Outbound Calls
💡 Outbound call transcription requires the same preconditions as inbound - see Preconditions above. No additional setup is needed beyond enabling outbound calls for the service. Transcription, summarization, live captioning, and codes & tags AI suggestions are also available for outbound audio calls. The same AI Service Settings that apply to inbound conversations automatically apply to outbound calls as well — no additional configuration is required.
This applies to the following outbound call types:
- Call on Behalf
- Scheduled Outbound Calls, either generated with or without workflows.
🔍 Notes
- Same configuration, same behavior. If Companion > Transcription is enabled for the service, transcription runs automatically for outbound calls on that service. The same applies to summarization, live captioning, and codes & tags AI suggestions. Data storage retention settings configured for inbound conversations apply equally to outbound conversations.
- Power Automate. The "When the Companion has an update" trigger and the "Get Companion task data" action also fire for outbound conversations, enabling the same Power Automate flows you use for inbound calls to work for outbound calls as well.
Customer Opt-Out of Transcription
The Opt-out of Transcription feature allows callers to withhold consent before the conversation is transcribed. When consent is not granted, Nimbus skips the entire transcription routine for that call — no live caption, no transcript, no summary, no codes & tags suggestions — and carries that decision across transfers so it is honored for the whole call journey.
🔍 How is consent expressed? Through the Record Consent Conversation Handling workflow activity, which you add to your IVR before the call reaches a Nimbus user. The rest of this page covers how to set that up, what the Nimbus user sees, and what stops happening.
✅Preconditions
- In addition to the standard Transcription preconditions (to enable the feature), the opt-out flow configuration requires an “IVR” or “Input Customer” step in your Workflow that asks the caller for consent.
- The consent outcome is then stored using the Record Consent Workflow Activity. This activity keeps the consent choice, which also persists through transfers.
How it Works
The opt-out flow comes down to four moments in the life of a call:
- The IVR asks the caller for consent — using an Input Customer or Input Customer (Advanced) Workflow Activity that informs the caller their conversation may be transcribed and captures their yes/no response.
-
The workflow records the answer using the Record Consent activity — set to Granted when the caller agreed, Not Granted when they declined.
☝️If the activity is not included in the workflow at all, Nimbus treats the call as consent granted by default. -
The call is distributed to a Nimbus user. When the Nimbus user connects, Nimbus checks the consent outcome:
- Granted (or no Record Consent activity in the workflow) → transcription, summarization, and codes & tags suggestions all run as usual.
- Not Granted → none of them are started for this session.
- The decision travels with the call. If the call is transferred, parked/unparked, or re-distributed to a different Nimbus user or service, the consent outcome is carried along automatically. A "not granted" stays in effect for the whole journey, unless the workflow on the receiving service explicitly re-asks the caller and records a new outcome.
Setting Up Opt-out in Your Workflow
The opt-out flow is configured entirely in your service Workflow. There is no dedicated admin page or tenant-level toggle — the consent decision is per call and lives in the workflow.
- Navigate to Nimbus Administration > Configuration > Workflows and open the workflow you want to add the opt-out to.
- Accept the call as usual.
- Add an Input Customer or Input Customer (Advanced) activity that informs the caller their conversation may be transcribed and asks them to confirm.
💡 Example: "This call may be transcribed and summarized for quality and follow-up. Press 1 to agree, press 2 to decline." - Based on the caller's input, add a Record Consent activity and set the outcome:
- Caller agreed → set to Granted
- Caller declined → set to Not Granted
- Continue the workflow as usual — route the caller to the correct queue or service.
Configuration Scenarios
✅ Default is permissive. If you do not include the Record Consent activity in your workflow, every call is treated as consent granted. You only need to add the consent step for services where opting out should be possible.
- Re-asking on transfer: If a call moves between services with different sensitivity levels (e.g. general hotline → HR or financial), the destination workflow can ask the consent question again using a new Record Consent activity. A caller can switch from Granted to Not Granted at the new touchpoint — and the other way around.
- Overriding consent without asking the caller: You can add a Record Consent activity with a hardcoded outcome to control consent for specific routing paths — for example, to always block transcription on certain escalation paths, or to grant consent automatically when it has been obtained through other means (e.g. a web form).
Show some scenario examples
| Scenario | What happens |
|---|---|
|
Caller declines transcription A✖️ |
The caller is informed at the start of the call that the conversation may be transcribed and chooses to decline. Nimbus disables transcription for the entire session, including any subsequent transfers. Typical examples: medical or health-related questions, financial topics, or HR cases where sensitive information will be discussed. |
|
Caller agrees to transcription A✔️ |
The caller agrees. Transcript and summary are produced as usual and can later be stored in the CRM or used for quality assurance. The caller may also request a copy through the channels your organization provides. |
|
Caller agrees, then revokes consent on transfer A✔️→B✖️ |
The caller agreed at the general hotline, but the conversation is transferred to a department where sensitive information will be discussed. The destination workflow re-asks the consent question and the caller declines — the new Not Granted outcome overwrites the previous Granted and transcription stops for the rest of the call. |
|
Caller declines, then gives consent on transfer A✖️→B✔️ |
The caller declined at the reception line but is transferred to a service where they would like a record — for example, step-by-step technical guidance they may want to re-read. The destination workflow re-asks the consent question and the caller agrees — transcription starts from that point onward. |
What the Nimbus User sees
The Nimbus user's experience changes only inside the Companion widget and the Transcription tab. The Transcription behavior will be as follows:
| Capability | ✔️ Consent granted (or no Record Consent activity in the workflow) | ✖️ Consent not granted |
|---|---|---|
| Live caption | Live caption appears in the Companion widget during the call. | Not shown. The Companion widget displays the "no consent granted" message instead. |
| Transcript | After the call ends, the transcript is processed and shown in the Companion widget and the Transcription tab. | Not generated. Nothing is stored against the session or retained for the tenant. |
| Summary | Summarization runs and the summary is displayed under the transcript. | Not generated. The Summary section displays the matching no-consent message. |
| Codes & Tags AI suggestions | Codes & tags suggestions are offered to the Nimbus user after the call. | Not offered. The Nimbus user can still set codes & tags manually. |
| Transfers / Hand-overs | Default behavior is “Consent” on the first service. When the call is transferred, the default stays in effect. | The consent outcome travels with the call. A "not granted" remains in effect at the next Nimbus user or service, unless the receiving workflow re-asks the caller using a new Record Consent activity. |
| Routing & Distribution | Unaffected. Calls are distributed exactly as the service settings dictate. | |
| Voicemail Transcription | Unaffected. The customer voice is recorded when “Voicemail” activities are reached in the workflow. When “Voicemail Transcription” was enabled in the Modalities Service Settings, an additional transcript is provided alongside the recorded Voicemails. | |
Limitations
- Set at IVR time, not mid-call. Consent is captured by the workflow before the agent connects. There is currently no in-call control for the caller or the agent to change the decision once the conversation is underway. The closest workaround is a transfer to a workflow that re-asks the question.
- MS Teams calls only. The consent check is enforced for calls routed through the Microsoft Teams driver. Other drivers are not in scope.
- Distribution is not consent-aware. Routing and distribution policies do not change based on the consent outcome. Consent only affects whether transcription, summarization, and AI suggestions run.
- Default is permissive. If the Record Consent activity is not included in the workflow, every call is treated as consent granted. To make opt-out available to callers, the workflow must include the consent step.
- Virtual User transcription respects the same consent outcome — see the related Virtual User transcription feature for details.
Transcription for Virtual Users
Whenever a caller interacts with Virtual Users, Nimbus can produce a full transcript of the conversation once the Virtual User or Service session ends. The transcript captures both sides of the interaction — what the caller said and how the Virtual User responded — in the same structure as a Companion transcript.
✅Preconditions as for Transcription apply, plus the following:
- Virtual User Transcription is enabled in the service's AI Service Settings.
- Optionally: Caller consent to transcription can be retrieved via “Record Consent” workflow activity. As the default is “Consent granted”, this step is optional.
- ☝️Note: Virtual User sessions are not shown on the My Sessions page. A Power Automate flow must subscribe to the Virtual User trigger event. Setup is done by an administrator or service owner → refer to the Power Automate chapter below.
Power Automate Transcription (and Summarization)
Virtual User transcription is intended for downstream automation rather than live viewing. There is no live caption during a Virtual User call, and thus no task distributed to any (human) Nimbus user.
When a Virtual User session ends and the transcript is ready, Nimbus triggers the Virtual User Power Automate Connector flow and provides the full transcript — for example to update a CRM record, create a case, route a complaint, or sent a message to a specific channel.
Setup Steps
-
Subscribe to the Trigger Event When a Virtual User has an update and pick one of the Virtual User events:
- Virtual User Transcription Ready — fires immediately when a single Virtual User session ends and its transcript is ready. If a service session runs more than one Virtual User, this event fires once per Virtual User.
- Virtual User Transcription Summarization Ready — same timing, but fires once the Summarization of that Virtual User session is ready. Requires Summarization to be enabled in addition to Transcription.
- Service Session Terminated — fires once when the overall service session ends. Use it to retrieve all Virtual User transcripts of that service session at once.
-
Add the Flow Action Get Virtual User Data to fetch the transcript.
- Provide the
Service Session ID(mandatory) and set Data Type to Transcription.
Tip: If you set the Set Data Type to Summarization to receive the summary rather than the full transcript. All other fields and scoping rules are identical. - Leave
Virtual User Session IDempty to return every Virtual User transcript of the service session, or supply it to return one specific Virtual User's transcript.
- Provide the
☝️ Important Note
- Virtual User (AI agent) transcripts are delivered through the Virtual User trigger only.
- The Companion trigger events fire for agent (human agent) conversations.
- A flow built for one interaction does not affect the other. If your workflow contains both Virtual User and regular human interactions (e.g. via Queue), 2 separate transcription events are recorded. Detailed instructions and notes can be found on “Virtual User” the Trigger Event and Flow Action.
🔎 Optionally, a transcript can also be condensed into a Summarization — a short recap of the same session, delivered alongside the transcript through the same respective Companion / Virtual User triggers and actions.
Transcription in other areas
Automated Voicemail Transcription
You can use transcription features to automatically transcribe incoming Voicemails. Once a caller is asked to leave a voicemail via workflow, a transcript can be generated and added to the Adaptive Card or shown in a Voicemail channel in MS Teams .
✅Optional, separated feature:
- "Voicemail Transcription" can be enabled within Modalities Service Settings and is NOT tied to any Companion or Virtual User settings or preconditions. Instead, a configured Speech Recognizers is used to convert voice to text. More information can be found on the Voicemails page.
- Your workflow requires a "Voicemail" Conversation Handling activity and to trigger the generation of a voicemail.
 |
 |
Known Limitations
☝️ KNOWN TRANSCRIPTION & SUMMARIZATION LIMITATIONS
💡We are actively working on further improvements to the following limitations:
- Transcription data is not part of service/user transfers. Data is kept within the current customer/user session. This capability will be provided with an upcoming release.
- Virtual User Reporting: Virtual User sessions and their transcripts will be included in Virtual User session reporting as part of a separate, upcoming capability. This is not available yet.
- Power Automate Connector: New “Virtual User” data fields in the Trigger Events / Flow Action may still be erroneously labeled as “Companion”. The terminology will be corrected in a separate update. The Virtual User / Companion connectors separate their data as intended.
🔎 BY DESIGN
💡The following points are not limitations, but important to note:
-
Transcription Feature availability:
- Transcription is a prerequisite to Live Captioning, Summarization, AI suggested Codes & Tags
- Not for Copilot / Direct Line 3.0 Virtual Users. Transcription features apply Virtual Users built on GPT Realtime bots only.
- Transcription relies on external services and APIs. When the feature is disabled or unavailable (e.g. throttling or service impediments) info messages are shown in the frontend UI widget. Nimbus task handling and user-customer interactions themselves are not affected by this and will continue normally.
-
Virtual Users:
- Not for Copilot / Direct Line 3.0. Transcription features apply Virtual Users built on GPT Realtime bots only.
- Post-session only. The transcript is produced after the session ends; there is no live/real-time transcription during a call.
- No (default) My Sessions view. Virtual User transcriptions do not appear on the Nimbus UI — they are consumed through Power Automate.
- No audio storage. Only the generated transcription data is processed in Power Automate. Audio recordings are discarded after automated processing.
-
Transcription scope:
- 3rd-party participants: Third parties, consultants and call conference attendees are not part of the transcribed content. Only the transcription data between Nimbus users / AI agents and the calling customer is kept.
- Supervision - As long as Supervisors remain in Listen or Whisper mode during a conversation, their voice is not being transcribed. Only during ”Barge In" they are part of the conversation and transcription is active.
- Summarization items generated from the Transcription might be missing when there is not enough data to draw from.
Troubleshooting