Status: preview, not yet confirmed GA
Microsoft's release plan lists public preview on April 30, 2026 and general availability in October 2026. Those are plans, not confirmation. Microsoft Learn's configuration page still labels the feature a production-ready preview, and it says conversation orchestration is available for voice and messaging channels only.
Microsoft's June 2026 Dynamics 365 blog announced public preview for voice and live chat channels. The roadmap's GA month is October 2026, the month this article is published. The documentation I reviewed does not confirm that GA has happened. Until Microsoft says so, treat this as a preview capability and check your tenant before relying on it.
The core idea: event, condition, action
Microsoft's responsible AI FAQ describes the model as an "event-condition-action" framework. The system monitors each conversation as conditions change, such as wait time rising, queues filling up or representatives going offline, and runs the relevant action from your configured playbooks.
Microsoft's example is "If a premium customer waits for more than three minutes and no senior agents are available, increase their priority and offer a callback." That is an illustration of the intent you can express. It is not a ready-made preset.
Unified routing is still the engine underneath. Microsoft describes the routing system as having two stages, classification and assignment. Orchestration adds event-driven behavior on top. It does not remove the need for queues, workstreams and channels.
The four scenario families
1. Dynamic prioritization. Priority is usually set once, when the conversation enters the system. Orchestration can update priority when a conversation transfers to a queue, and escalate priority based on wait time. Microsoft's troubleshooting guidance says to check that the queue has no custom prioritization rules, because those can interfere with the playbook.
2. Overflow handling. The overflow triggers include:
- the queue being outside operating hours
- all representatives being signed out
- a high estimated average wait
- an individual conversation's wait time
- the number of waiting conversations
- a direct inward dial call that a representative rejects, misses or can't take
The overflow actions Microsoft lists are transferring to another queue or external number, offering a callback or voicemail, keeping the conversation unassigned, or ending it.
3. Preferred or previously engaged representatives. The aim is to spare customers from repeating themselves. Playbooks can assign a conversation to a previously engaged or preferred expert, and can keep the same expert for asynchronous chats by checking presence status and active conversation limits. If that representative can't take the conversation, a configured fallback assigns the next available representative. Microsoft's troubleshooting notes say to verify the preferred-representative mapping and confirm that person is a queue member with an allowed presence status.
4. Bullseye routing. This expands the pool of eligible representatives in rings as the wait grows, using user groups. How it behaves, per Microsoft's bullseye documentation:
- Assignment starts at Level 0. Later levels are added after configured wait times.
- Earlier groups stay in the pool and keep higher priority.
- Wait time is measured cumulatively from the first attempt to assign to a human representative.
- The minimum wait time is 30 seconds. Level 1 can also be set to 0 seconds. Each later level either matches the previous wait time or adds at least 10 seconds.
- A transfer to another queue resets evaluation to Level 0.
- Among equally eligible users, the queue's assignment strategy decides. Custom assignments aren't supported.
Microsoft's sample setup uses an Expert group at Level 0, a Regional group at 30 seconds and a General group at 60 seconds. If one playbook covers several queues, it only offers the user groups those queues share. Where queues don't share groups, create separate bullseye playbooks.
Setup and prerequisites
You configure playbooks in Copilot Service admin center under Customer support > Conversation Orchestration (Preview). Microsoft's documentation lists these prerequisites:
- The System administrator or Omnichannel administrator role.
- Unified routing enabled.
- At least one queue and workstream set up for voice or messaging, plus a configured channel.
- The appropriate licensing. Microsoft says specific requirements apply and points to the Dynamics 365 Licensing Guide.
A playbook has queues, a trigger event, a channel and conditions. You can define up to 10 conditions, and each condition can use up to two context variables. Example variables include customer tier, VIP status, preferred language, customer region and account type. Saving creates a draft. Publishing activates it. A playbook can be moved between environments by adding it to a solution and using solution export and import.
Conflicts and validation
Validation is a real feature here, though it has limits.
- You can't have two active playbooks for the same scenario on the same queue. The attempt fails at publish with an error saying a playbook for that scenario is already published for the queue. This holds even if the intervals or priority values differ. The fix is to edit the existing playbook.
- Different scenarios on the same queue are allowed. For example, wait-time prioritization can sit alongside availability-based overflow. They run according to their triggering events.
- Cross-playbook checks happen only at publish time, and only against other active playbooks. A draft can look fine and still fail at publish.
Why you should still test
The roadmap says built-in validation lets admins publish without manually testing every edge case. Microsoft's technical documentation is more cautious. When you publish, the system converts your natural-language playbook into a structured runtime format. Microsoft says a language model does that conversion and the output may not fully capture the intended logic. It adds that deviations may show up only at runtime, and it tells administrators to test and monitor runtime behavior and revise the playbook if needed.
The responsible AI FAQ also says Microsoft will add a human review step. That would let administrators review the converted output before it takes effect. The FAQ describes this as planned, not shipped.
A WindowsForum report on the overflow templates gives similar practical advice: verify that the templates have reached your tenant ring, and test playbooks in a non-production sandbox before pointing them at live queues. This is general best practice, not a Microsoft requirement.
Limits and compliance
- Language: The responsible AI FAQ says the system currently supports English only.
- Employment decisions: Microsoft says the feature should not be used for decisions affecting employees, such as compensation, rewards or seniority.
- Legal duties: Customers are responsible for lawful monitoring, recording and storage of communications. That includes notifying end users and obtaining consent where the law requires it.
- Opting out: Administrators can delete saved or published playbooks and can turn off the feature toggle. Deleting an active playbook stops orchestration for its queue.
What this means for admins
For contact center teams already on unified routing, this is a meaningful step. Adjusting priority or overflow behavior no longer has to mean rebuilding rule trees. The practical gains are fewer customers stuck behind a stale priority, a defined path when queues close or empty, and better continuity with returning customers.
The caveats are equally practical:
- It is a preview with voice and messaging only.
- It is English only.
- A language model sits in the authoring path.
- Queue-level prioritization rules can interfere with dynamic prioritization.
A sensible rollout:
- Check licensing and confirm unified routing is enabled.
- Start with one queue and one scenario, such as overflow when all representatives are signed out.
- Test in a sandbox with realistic conversations.
- Use Microsoft's diagnostics queries to watch what the playbook actually does.
- Widen the scope only after the runtime behavior matches your intent.
Microsoft says these four scenarios are the first in a growing set. It names in-conversation messaging and escalation as future areas. Roadmap items change, so check the live roadmap before planning around any of them.
Update: Microsoft roadmap lists conversation orchestration as launched (November 7, 2026)
Microsoft’s AI at Work roadmap now marks conversation orchestration as Launched, with general availability listed for October 2026. That supersedes the earlier assessment that GA had not been confirmed.
The roadmap lists worldwide standard multi-tenant availability and says admins enable the feature for users. It does not establish that every tenant has received it, so administrators should still check availability and licensing in their own environments.