What it does

Attach a Copilot Studio agent to the voice channel as an IVR before live reps. Handles speech and DTMF input, does lookups, routes by intent, transfers or ends calls. Supports classic topic-matching and generative AI.

Key facts

• 1-hour maximum for bot-handled voice calls: auto-disconnected if not transferred or ended within this window
• Input methods: natural language speech, DTMF keypad, constrained speech recognition (grammar-based, for alphanumeric inputs like account numbers)
• Supports SSML for voice customisation, barge-in (customer interrupts bot mid-prompt), silence detection, custom audio files
• SIP UUI and X-Headers for passing context in/out of calls (Direct Routing only: not available with ACS managed numbers)
• Two NLU options: standard NLU (processes in Copilot Studio) and NLU+ (processes within D365 boundary: relevant for data residency requirements)
• Multilingual IVR requires separate configuration in Copilot Studio: not configured solely in admin center

When to use / skip

Use IVR when the client wants automated triage, self-service resolution, or intent-based routing before human reps. Skip if all calls go straight to queue, pointless IVR just adds friction.

Configuration decisions

• Classic vs generative AI orchestration: classic for deterministic menu-style IVR; generative for natural language intent handling
• DTMF vs natural language input: DTMF is more reliable in noisy environments and for structured data entry; NLU is better for open-ended queries
• Constrained speech recognition vs free-form: use constrained for account numbers, dates, tracking numbers where exact input matters
• Escalation path design: define what happens when the bot can't resolve: which queue, what context passes to the rep

Gotchas

• Build the transfer-to-agent path first. Most IVR failures are teams that spend weeks on self-service flows without confirming escalation works. Get a live call to reach a rep before designing any self-service content.
• The 1-hour bot call limit affects complex IVR flows. For multi-step troubleshooting or long processes, test total call duration against this ceiling.
• SIP X-header context passing needs Direct Routing. If the client uses ACS managed numbers and needs legacy IVR context, you'll need Direct Routing for this.
• NLU+ keeps data inside the D365 boundary. For strict data residency, NLU+ is right. Standard NLU sends intent processing to Copilot Studio infrastructure.

Consultant notes

• Build the transfer-to-agent path first, not the self-service content. Most IVR projects get three weeks into bot design before confirming that escalation actually connects a call to a rep. Test the full handoff path end-to-end before designing any self-service topics: it's the foundation everything else depends on.
• NLU+ for data residency should be on the checklist question for regulated clients. If the client has data residency requirements, standard NLU sending intent processing outside the D365 boundary could be a problem. Raise it in discovery, not after go-live.
• DTMF vs natural language is a genuine design choice, not a default. Noisy call environments, older demographics, and structured data entry (account numbers, dates) all favour DTMF. Don't default to natural language just because it sounds more modern.

---

Source last updated: 2026-03-12 | Worth checking again when generative AI orchestration reaches GA, or if SIP context passing extends to ACS managed numbers