What it does

Puts a two-way AI conversation inside a journey. The journey calls a proactive engagement configured in Dynamics 365 Contact Center, a Copilot Studio agent holds a voice call or an SMS exchange with the customer, and the journey branches on what actually happened. The SMS side was added in the September 2026 update; voice came first.

Key facts

• Turn on the Contact Center integration feature switch under Integrations. Customer Insights – Journeys then detects Dynamics 365 Contact Center automatically if it's in the same environment.
• Contact-based journeys only. Leads and Customer Insights – Data profiles can't be the audience.
• The journey doesn't hold the conversation script. It selects a proactive engagement defined in Contact Center; the AI agent, its prompts and its outcomes all live there.
• Voice steps pick a compliance profile, a Purpose (and Topic where used) and a quiet-time option on the tile. SMS steps don't pick a compliance profile on the tile — it comes from the text message record itself, and SMS reuses the ordinary text-message compliance profiles rather than the separate voice ones.
• Three variables describe an outcome, and you often need all three: Disposition codes (set by a human representative), Result (returned by Contact Center for the proactive engagement request), and Attributes (set by the AI agent, only populated when a conversation actually connected).
• Branching is two-stage. First a Wait for trigger step on Previous message gets an interaction, splitting into attempted / not attempted / blocked; then an Attribute branch on the detail. For SMS the create branches link builds the whole structure for you.
• The three top-level outcomes mean different things operationally: attempted failed for a temporary reason (voicemail), so a retry may work; not attempted failed for a permanent one (bad number), so change channel; blocked means the journey never called Contact Center at all — no consent, or quiet time left no window.
• Quiet times apply and matter more here than for email, because voice outreach runs into rules like the US TCPA that restrict calling hours by the recipient's local time.

When to use / skip

Use it where the next step genuinely depends on what the customer says: qualification, appointment confirmation, renewal intent, collections triage. The value isn't the outbound message, it's that the journey path is decided by a real answer rather than by whether a link was clicked.

Skip it for anything a template message and a click would settle. It costs a Contact Center licence footprint, a Copilot Studio agent to build and maintain, and a consent model that is materially harder than email. And skip it entirely if the audience is leads — the integration doesn't support them.

Configuration decisions

• Same-environment deployment of Contact Center and Customer Insights – Journeys. This is a prerequisite, not a preference, and it drives environment strategy well before anyone writes a journey.
• Which team owns the AI agent. The journey is marketing's; the proactive engagement and its outcomes are the contact centre's. Somebody has to own the seam.
• Which of disposition code, result and attribute each branch reads. Mixing them arbitrarily produces branches that look right and fire wrongly.
• Compliance profile design for voice, which is separate from the SMS and email profiles and usually needs its own Purposes.
• Retry policy per outcome — what attempted leads to, and which channel not attempted falls back to.

Gotchas

• Same environment or nothing. There's no cross-environment option. If Contact Center already lives in a separate environment from marketing, this feature is a migration conversation, not a configuration one.
• "Blocked" is not a failure of the call, it's a failure of your consent or quiet-time setup, and it never reaches Contact Center at all. Teams debugging low connect rates look at the agent first and miss this for days. Branch on it explicitly and count it.
• Attributes only exist for conversations that connected. A branch keyed on an attribute silently takes the default path for every unanswered call, which looks like the agent misbehaving.
• Voice and SMS handle compliance profiles differently — tile-level for voice, message-level for SMS. Reviewing a journey's consent posture means looking in two places.
• Contact Center owns the disposition codes. Someone renames one there and the journey branch stops matching, with no warning on the journey side.
• Leads not being supported bites late, because most outbound campaigns that want a qualifying call are lead-based by nature.

Consultant notes

• Scope this as two workstreams, not one. Every project that treats it as "a new journey tile" underestimates it, because the interesting half — the agent, its prompts, the outcome model, the disposition codes — is Contact Center work with a Copilot Studio build behind it.
• Nail the outcome taxonomy on a whiteboard before anyone opens the journey designer. Disposition code, result and attribute are three different sources with three different owners, and journeys built by guessing which one to read are the main reason these things behave oddly in UAT.
• The consent conversation needs legal in the room early, particularly for voice. Quiet times, TCPA-style calling windows and a separate voice compliance profile are not things you retrofit the week before go-live, and getting them wrong is a regulatory problem rather than a defect.
• Get the same-environment requirement in front of the architecture team on day one. It's the single constraint most likely to kill the design, and it's cheap to check.
• Instrument the blocked branch even if the client sees no reason to. It's the fastest way to find out that consent data is thinner than anyone claimed.

---

Worth revisiting if lead-based audiences become supported, or when the SMS conversation step gets its own release-plan entry