How Teams call recording works, step by step

The recording workflow at a glance

Every recorded Teams call moves through the same five stages, and all of them run automatically once your tenant is configured. The recording engine is a certified Microsoft Graph Communications recording application— a bot that Microsoft's compliance framework invites into calls — running as a managed .NET service on your dedicated server. There is nothing for your users to click, and nothing for your IT team to babysit. Here is the full pipeline, start to finish:

• 1. Join. The bot joins the call as a participant, triggered by a compliance policy, a manual request, a calendar event, or a direct call.
• 2. Capture. It records a combined mix of the call and a separate, unmixed track for each participant.
• 3. Transcribe. Azure AI Speech-to-Text converts the audio to searchable text, labelled by speaker.
• 4. Store. The recording and transcript upload automatically to a SharePoint library in your own Microsoft 365 tenant.
• 5. Access & monitor. A live dashboard shows active recordings in real time, and finished evidence sits in your tenant under your controls.

The rest of this page unpacks each stage. If you want the regulatory side — which rules require recorded communications and for how long — see our compliance recording overview instead. This page is about the mechanics.

Step 1: The bot joins the call

Nothing gets recorded until the bot is in the call, so the join step is where the whole workflow starts. Because the recording application is built on the Microsoft Graph Communications platform, Microsoft Teams treats it as a legitimate participant and hands it the call's media the moment it joins. The bot can enter a call four different ways, which together cover both policy-scoped users and one-off scenarios:

• Compliance policy (the primary method). When an administrator assigns a Microsoft compliance recording policy to a user, the platform automatically invites the bot into every call that user makes or receives. Teams shows its standard recording banner to participants, and the user cannot opt out — which is exactly why policy-based joining is the backbone of compliance recording.
• Manual join. Using the Graph API, the bot can be added to any specific call or meeting on demand, for a one-off session you want captured deliberately.
• Calendar auto-join. The bot monitors Exchange Online calendars for a defined group of users and joins their scheduled meetings automatically as those meetings start — useful for capturing recurring meetings without assigning a per-user policy.
• Direct 1:1 calls. Direct calls placed to or from in-scope users are captured too, so conversations that happen outside of scheduled meetings do not slip through.

The policy-based path relies on Microsoft's own framework; you can read Microsoft's documentation on compliance recording policies for the platform details. Each method is covered in depth on our recording bot and automatic recording pages.

Step 2: Capturing audio

Once the bot is in the call, it captures the audio in two forms at the same time. The first is a combined recording — a single mixed track of the whole conversation, which is what most people picture when they think of a call recording. The second is per-participant, unmixed audio: a separate track for each speaker, isolated from everyone else on the call.

The separate tracks are not a nice-to-have. On a busy multi-party call where people talk over one another, a single mixed track makes it genuinely hard to prove who said what. Unmixed per-speaker audio removes that ambiguity — each voice sits on its own track, so attribution is clean and defensible even when the conversation is crowded. For regulated firms, that clarity is the difference between a recording that answers an auditor's question and one that raises new ones. Video teleconferencing (VTC) endpoints are supported as well, so calls that involve room systems are captured the same way.

Step 3: Transcription

After capture, each recording is transcribed automatically using Azure AI Speech-to-Text. Microsoft's speech service is well suited to this because it applies speaker diarization — it separates the audio by speaker and labels each segment of the transcript accordingly. Combined with the per-participant tracks captured in the previous step, that produces a transcript where you can see not just what was said, but who said it.

The practical payoff is search. Audio alone is slow to review — you have to listen through it. A speaker-labelled transcript turns every conversation into readable, searchable, auditable text, so finding the relevant moment across hundreds of calls becomes a text search rather than an afternoon of listening. If you want more detail on how Azure speech models handle this, see Microsoft's Speech-to-Text documentation.

Step 4: Archiving to SharePoint

When capture and transcription finish, the recording and its transcript upload automatically to a SharePoint document library in your own Microsoft 365 tenant. This is a deliberate design choice: the evidence lives inside your environment, governed by your access controls and your retention rules, rather than on a third-party platform you do not fully control.

Because the archive is your SharePoint, you decide who can open recordings, how long they are kept, and how retention maps to your regulatory obligation. Retention is configurable to match whatever your regulator requires — whether that is the multi-year preservation windows common in financial services or the access-controlled handling that healthcare data demands. There is no manual export step and no separate vendor portal to log into for day-to-day access; the files simply appear in a library you already own. The SharePoint storage page covers the archiving model in more detail.

Step 5: Monitoring & control

The final stage runs continuously rather than call by call. Your dedicated server hosts a live web dashboard that shows active recordings and calls in real time, so you can see exactly what is being captured at any given moment. If a call needs to be stopped — for example, a session that should not be recorded for a compliance reason — it can be terminated directly from the dashboard.

Just as importantly, this monitoring is managed by our team on your dedicated, US-hosted server. We host the server, keep the recording service running, apply patches, and watch for problems, so the recording pipeline stays healthy without your IT team operating it. The single-tenant model means your recordings, transcripts, and configuration are never co-mingled with any other organization's data. You can read more about that on our compliance recording service page.

What onboarding looks like

Getting to that automated steady state takes a short, well-defined onboarding. It is a collaboration: we handle the server and the recording engine, and your Microsoft 365 administrator handles the tenant-side actions that only they can perform. The sequence looks like this:

• Provision your dedicated server. We stand up a single-tenant, US-hosted server for your organization and install the managed recording service on it.
• Register the recording application in your tenant. Your administrator registers our Microsoft Graph recording application in your Microsoft 365 tenant and grants it the permissions it needs to join and record calls.
• Assign the compliance policy to in-scope users. Your administrator applies the compliance recording policy to the users who need to be captured, which is what enables automatic policy-based joining.
• Confirm SharePoint archiving. We connect the pipeline to the SharePoint library in your tenant where recordings and transcripts should be stored, and confirm the upload path works end to end.
• Run test recordings before go-live. We record and transcribe test calls to verify that capture, transcription, and archiving all behave correctly, and only then move the service into live use.

We do not quote a fixed number of days, because the honest answer is that onboarding timelines depend on your tenant administrator's availability to complete the Microsoft 365 steps. Once those are done, the rest moves quickly. To see the managed model in full, visit the compliance recording service overview.

FAQ