Almanac

Consultant-focused KB for Microsoft Dynamics 365 Contact Center: implementation notes, gotchas, and configuration decisions beyond the official docs — across voice and digital channels, routing, agent and supervisor experience, Copilot & AI, workforce engagement, analytics, administration and security.

feature-consent-based-recording.mdv1 · history
CurrentApplies to BothUpdated 2 months agoSource Microsoft Learn

What it does

Gives organisations a mechanism to capture explicit customer consent for call recording at the start of a voice interaction with a voice agent, and honours that consent decision through any escalation to a human rep. If a customer declines, recording and transcription are suppressed for the entire conversation, including after transfer.

Key facts

  • GA'd April 2026; opt-in: requires admin enablement in Copilot Service admin center
  • Consent is captured by the voice agent (IVR/Copilot Studio bot) at the start of the call
  • If consent is declined: recording is off, transcription is off, and the rep cannot start, pause, or view transcripts
  • Consent decision travels with the conversation through escalation: the rep inherits the customer's choice, not their own preference
  • Targeted at GDPR and similar voice recording compliance obligations

When to use / skip

Use this on any deployment where regulatory compliance requires documented consent before recording, common in UK, EU, and other jurisdictions with two-party consent laws. If your legal team has flagged recording compliance as a requirement, this is the solution.

Skip it if you're in a jurisdiction where recording disclosure is achieved by a static IVR prompt ("this call may be recorded") and no opt-out is required. The consent mechanism adds flow complexity; if you don't need it, don't add it.

Configuration decisions

  • Enable in Copilot Service admin center: the exact path isn't yet fully documented in GA docs, but it's in the recording/compliance section
  • Design the consent prompt script: what the voice agent says, what key press or voice response counts as consent, and what happens next if the customer declines
  • Decide what the rep experience looks like when recording is suppressed: reps should know why they can't access transcripts so they don't raise a support ticket

Gotchas

  • The feature works end-to-end only when you have a Copilot Studio voice agent at the front of the call. If callers can reach a rep directly (no IVR), the consent capture step doesn't fire. Don't deploy this as your sole compliance mechanism if you have any direct-to-rep paths.
  • Declined consent means no transcription for the entire session. AI features that depend on transcription (Copilot summaries, sentiment analysis, real-time suggestions) will be unavailable for those calls. Factor that into your quality and reporting expectations.
  • "Consent travels through escalation" sounds clean but requires the voice agent to pass the consent flag correctly in the context handoff. Test this explicitly in UAT: the context variable name matters and varies by Copilot Studio implementation.

Consultant notes

  • This is a legal-led feature. Before designing the consent flow, get the legal requirement documented: which jurisdiction, what the consent basis is, whether verbal consent is sufficient or a DTMF key press is required, and what the consequence of declining is. The technical implementation is straightforward once the legal brief exists.
  • The biggest architecture oversight is assuming consent capture works on all call paths. If any callers can reach a rep directly (no IVR, direct dial, Teams escalation), the consent step doesn't fire. Map every inbound path before declaring this feature covers your compliance obligation.
  • Test the consent context variable handoff in Copilot Studio explicitly: the variable name must match exactly between the bot topic and the D365 recording configuration. A mismatch means recording runs despite the customer declining, which is the compliance failure you're trying to avoid.

Worth revisiting as regulators clarify AI-generated transcript requirements: the consent model may need extending to cover AI-derived data, not just the raw recording.

Was this accurate?