Dynamics 365 Contact Center Customer-First Callback: Setup, Profiles and Queue Risks

What changed, and why it matters​

In the classic direct callback experience, the system assigns the callback request to a representative before it calls the customer. If the customer doesn't answer, hits voicemail, or the number is unavailable, the representative waits while the attempt finishes. That is a polite way of saying you pay a salaried person to listen to ringing.

With customer-first callback, the system calls the customer first when the request reaches the front of the queue. Once the customer answers and completes any required verification, the system connects them to an available representative. The roadmap entry describes the business value as keeping the customer's place in line while representatives join only when the customer is ready.

The lifecycle, step by step​

According to Microsoft Learn:

1. The request stays in the queue and moves according to routing priority.
2. When it reaches the front and at least one representative is available, the system starts the callback.
3. An AI agent can engage the customer, collect information, or perform verification.
4. The system transfers the customer to an available representative, who sees a direct callback notification.

Microsoft's documentation also says the system automatically closes direct callback requests after seven days.

Release timing: a small mess​

The sources disagree on the date.

• Microsoft's release-plan page lists general availability for Aug 24, 2026. An earlier snapshot of the same page showed only "Aug 2026".
• A third-party roundup (M365 Admin) reported July 31, 2026, and another listing showed July 2026.
• The roadmap record in the submitted story shows a month-only August 2026 date and a status of Launched.
• The release-plan page now carries a notice that it was archived as of October 2026 and may not reflect current guidance.

Treat the Learn configuration article as the authoritative source for how the feature works today. Don't rely on any one date. The roadmap entry lists availability across Worldwide, GCC, GCC High and DoD clouds, which matters for public-sector deployments.

Setup walkthrough​

These steps follow Microsoft's documentation.

Prerequisites

• Provision the Microsoft Copilot Studio agent in your environment.
• Configure phone numbers with outbound calling enabled.
• Create any Copilot Studio agents you plan to use for verification or information collection.

Enable the feature

1. In Copilot Service admin center, go to Routing > Callback management.
2. Enable the Copilot AI agent for callback management.
3. Confirm the enablement prompt. The setting is irreversible.
4. Wait for the Callback Agent to provision, and check that it did so without errors.
5. Select the Customer-first direct callback toggle.
6. Create and save your first callback profile. Saving it enables the feature.

Two caveats from Microsoft belong in your change ticket. AI agent use can consume Copilot or Microsoft service credits, and the licensing and usage requirements that apply to other AI agents also apply to callback agents. The documentation also says you can't disable the feature once it is on. During initial setup, the first profile automatically applies to queues that already use direct callback as an overflow action. Expect existing queues to change behavior immediately.

What a callback profile controls​

• Callback flow. Choose whether to ask for an alternate number, edit the initial offer message, and set how often the message repeats until the customer responds. A preconfigured Copilot AI agent captures the alternate number. Microsoft warns against editing or deleting it, because changes might stop callbacks working as expected.
• Callback offering window. Offer callbacks only during certain hours, stop near closing time, or limit them at peak periods. The time zone follows the queue's operating hours, and UTC is used if none are defined.
• Duplicate prevention. Check within the current queue or across all queues. Then choose to inform the customer and end the call, or inform them and let them stay in the queue. Detection uses the customer's phone number and runs only when overflow triggers the direct callback experience.
• Queue allocation. Set a maximum number of active callback requests. At the limit, callbacks stop being offered and customers use the standard queue. Offers resume automatically when capacity frees up.
• Dialing. Use the dialed-in number, with a fallback if it can't make outbound calls, or always use a specific number.
• AI agent and fallback. Pick the Copilot Studio agent that handles the customer after they answer. Use it to verify customers, collect information, present callback context, route to representatives or escalate. If the agent can't engage, choose "Prompt and hang up" or "Prompt and escalate".
• Rules. Set a target average wait. Microsoft's example is five minutes over the last 30 minutes. If recent callbacks averaged eight minutes, new attempts are delayed three minutes. At six minutes the delay drops to one minute, and at five or below calls go out immediately.
• Allow callbacks outside operating hours. An optional setting.
• Retry behavior. Configure retries for no answer, failed calls and answering machines. Set the number of attempts and the wait between them. The minimum wait is five minutes.

The same-platform trap​

This is the setting most likely to cause trouble. Inbound calls and outbound callbacks must use the same phone platform. Microsoft's example is a customer calling an Azure Communication Services number while the callback uses a Microsoft Teams number. In that case the system doesn't initiate the callback and automatically closes the request.

If your estate mixes telephony platforms, audit it before you flip the switch. Otherwise customers could be offered callbacks that quietly close.

Attaching a profile to a queue​

For a voice queue, select Direct callback as the overflow action for a queue condition. Then choose a callback profile and save. Microsoft's documentation says you can configure one callback profile per overflow condition for a single queue.

Profiles can differ by queue or scenario. Microsoft's example is one profile that requires verification and another that plays an informational message before transfer. A billing queue and a general-inquiries queue don't need the same experience.

Analysis: what to weigh​

The upside. The design targets a real inefficiency, which is representatives sitting through unanswered attempts. The wait-time rules also address a quieter problem: a customer who answers a callback and then waits again for a representative. Delaying dialing to match capacity is a sensible safeguard.

The caveats.

• Microsoft's documentation explains the design rationale but gives no quantified savings or service-level guarantees. Measure the effect in your own environment.
• AI agent consumption has a cost. Model the credit use before enabling it broadly.
• Enablement can't be undone. Test the first profile carefully, because it applies automatically to queues already using direct callback.
• Unattended outbound dialing needs a verification agent that is well built. That includes the fallback path, which should cover customers who don't engage with the AI.

A related duplicate-prevention roadmap item exists separately. It describes combination-based logic using caller number, channel number and queue ID. That is distinct from the profile-level duplicate setting documented here, so check which one applies to your tenant.

Pre-flight checklist​

1. Verify outbound calling and a single phone platform for inbound and callback numbers.
2. Review queue operating hours and time zones.
3. Decide queue allocation limits and duplicate handling.
4. Leave the alternate-number capture agent untouched.
5. Build and test your verification agent, and set the fallback action.
6. Set retry rules, remembering the five-minute minimum.
7. Estimate credit consumption and get licensing sign-off.
8. Attach a profile to every direct-callback overflow condition.

This is a configuration-heavy feature with an irreversible switch. Plan the rollout like a production change, not a toggle you try for fun.