October 23, 2026 is the first date that changes the migration conversation. Microsoft begins restricting onboarding and expansion for customers without an ACS resource created before that date. The final retirement date is September 30, 2028, but treating that day as the start of the project leaves too much work in one queue.
Migrate in dependency order. Inventory every ACS capability, start with services that are retired outright or have no migration path, validate the underlying calling path before replacing its user interface, then move surviving workloads through a staged release. The order should be adjusted for the application's dependency graph and customer cutover risk.
Microsoft's categories give you the first sorting rule. Retired capabilities will be removed after September 30, 2028. Breaking-change capabilities remain available, but continued support requires modification and use with a supported Teams-aligned service. The current categories and dates are summarised in the ACS retirement and cost breakdown; the migration order starts by turning that status list into an engineering inventory.
1. Build the capability inventory before choosing an order
Begin with product behaviour rather than ACS package names. For each user-facing feature, record the ACS capability underneath it, the data it creates, the identity and token flow, external integrations, client surfaces, operational alerts, and the customer promise attached to it.
| Inventory question | What to record |
|---|---|
| What does the customer use? | Chat, rooms, calling, SMS, email, recording, captions, diagnostics, or automation. |
| What does the product depend on? | ACS SDKs, UI components, APIs, webhooks, media workflows, storage, identity, and provisioning. |
| What must survive the cutover? | Active sessions, phone numbers, recordings, transcripts, conversation history, permissions, and audit data. |
| What can be released independently? | Backend services, client UI, calling or messaging paths, admin tools, and reporting. |
| What is the rollback? | The old route, retained data, reversible configuration, and the signal that triggers a return. |
Use a row for every capability even when several rows appear to belong to one feature. A customer support screen may combine calling, recording, captions, diagnostics, identity, and a custom UI. Those parts can have different retirement categories and different replacement work.
This inventory also exposes the work that a simple endpoint swap hides. The video-provider switching inventory covers SDK objects, tokens, events, rendering, and recordings. Reuse that shape for ACS, then add the migration date, dependency owner, customer impact, and rollback state to each work package.
2. Move retired capabilities and missing paths to the front
Retired capabilities should enter design and replacement work first because Microsoft's guide places them on a removal path. The list covers the messaging services, Rooms, Job Router, number management, Direct Routing and both UI Library SDKs. They remain available during the transition but are listed as unavailable after September 30, 2028.
Rooms needs special treatment. Microsoft says ACS Rooms and Teams meetings through the Microsoft Graph API exist in different environments and do not offer a migration path. That makes Rooms an early redesign constraint. The work is to define the target meeting model, map participants and permissions, preserve any required records, and test the customer journey before committing to a cutover date. The specific constraint is covered in the ACS Rooms migration analysis.
Missing migration paths also change the shape of the schedule. A capability with a direct replacement can be isolated behind an adapter while the replacement is tested. A capability with no direct path needs product decisions, data decisions, and customer acceptance before implementation can be considered complete.
Put these work packages at the front of the programme, then examine what they sit underneath. A replacement that changes identity, permissions, or shared records can affect later calling and administration work.
3. Validate the calling path before replacing the interface
Microsoft lists ACS Voice and Video Calling SDK, Call Diagnostics, Call Automation, Audio Streaming, Call Recording, and Closed Captions as breaking changes. These capabilities remain available after the retirement date, but Microsoft says continued support requires a supported Teams-aligned service and may require a new major SDK version. Eligibility does not guarantee feature parity.
That category calls for a compatibility workstream before a visual rewrite. Map the current calling path from authentication to session setup, media negotiation, events, recording, diagnostics, automation, and failure recovery. Then test the target Teams-aligned scenario or alternative route against the behaviours the product actually promises.
The UI Library deserves its own checkpoint. Microsoft lists the Web and Mobile UI Library SDKs as retired while allowing them to remain available through the retirement period with limited break-fix support. Microsoft also says customers that need ongoing evolution or dependency modernisation should fork and maintain the library independently. The UI Library replacement analysis covers the rebuild surface.
Do not schedule a visual replacement as if it were independent of the calling path. In your dependency graph, identify which UI states come from the current calling implementation, which are product-owned, and which are produced by recording, diagnostics, or automation. Validate the target calling flow first. Then replace or fork the UI against a known set of events and states. This sequence reduces the chance of presenting a finished shell over an untested media path.
4. Leave surviving workloads until their target is proven
Existing ACS Calling workloads have no immediate changes and continue to operate until September 30, 2028, according to Microsoft's guide. That gives a team room to test a target route before moving a stable workload. It does not provide a reason to leave the dependency unexamined: post-retirement support requires an eligible Teams-aligned scenario.
Microsoft names Teams Phone Extensibility, Teams Meeting Interoperability, and Teams Click-to-Call for Teams Voice Apps as supported Teams-aligned services in its guidance. It also identifies Teams interoperability or an alternative partner solution through Microsoft Marketplace as directions for ACS Calling migration. These are documented routes; each workload still has to be tested against them.
Use the remaining ACS period to prove the target against real product requirements. Test authentication, call setup, media, reconnect, recording, diagnostics, automation, captions, administration, and support tooling. Include the cases that are easy to miss in a happy-path demo: a returning user, a changed permission, an interrupted network, an active recording, and a session that begins before the cutover and ends after it.
Commercial review belongs alongside technical validation. The supported route may introduce service or licensing implications, and the applicable model depends on the identities, participants, Teams-aligned service, and contract terms in use. The Teams licensing analysis covers that branch without applying one per-seat price to every deployment.
5. Sequence the work in releaseable stages
Once the inventory and dependency graph are complete, turn them into stages that can be tested and reversed. Microsoft's migration guidance recommends early planning, dependency and commercial assessment, testing, and phased deployment. The sequence below applies those principles to an ACS workload.
- Freeze the baseline. Capture current SDK versions, API calls, events, identity flows, data stores, recordings, operational dashboards, and customer-facing behaviour. Record the ACS resource creation date because the October 23, 2026 restriction treats customers differently based on whether a resource existed before that date.
- Separate retired work from breaking-change work. Give every capability a category, target route, owner, dependency list, and customer impact. Retired services and missing paths receive the earliest design attention.
- Redesign the hard constraints. Start Rooms and any other capability without a direct migration path. Agree the product behaviour, data handling, permissions, and support model before building a replacement.
- Prove the underlying service path. For calling-related workloads, validate the target scenario and its event, media, recording, diagnostic, automation, and caption behaviour before replacing the client shell.
- Build parallel adapters and test fixtures. Keep the old route available while the new route handles controlled traffic. Compare events, state transitions, media outcomes, records, and operational alerts rather than relying on visual similarity.
- Release by customer or workload group. Choose a group with a clear rollback path, monitor the agreed signals, and retain the old route until the observation period and acceptance criteria are complete.
- Retire the old dependency deliberately. Remove credentials, SDKs, webhooks, storage paths, dashboards, runbooks, and provisioning logic only after the replacement is operating and the required records are retained.
The order inside each stage should follow dependency depth and cutover risk. A small feature with a missing path may deserve earlier attention than a large but well-supported calling workload. The dates set urgency; the dependency graph sets the implementation order.
Build, Buy, or Deploy the Replacement
Sequencing does not decide the replacement architecture by itself. Once the capability inventory is complete, choose the path that matches the workload's control, media, compliance, and maintenance requirements.
Build on open source
Building on open source can make sense when the team needs control over signalling, media handling, deployment location, or long-term protocol choices. It also transfers responsibility for security updates, capacity planning, observability, interoperability, client behaviour, recording, and incident response to the product team. The migration schedule must include that operating capability as well as the code that replaces the ACS calls.
Buy a cloud API
A cloud API can shorten the first replacement cycle when the workload fits the provider's supported primitives and the team accepts provider-specific abstractions. The evaluation should cover the exact calling, recording, automation, caption, identity, data retention, and regional requirements in the inventory. A short integration can still mean a long migration if the product's current behaviour sits outside the new API's model.
Deploy a commercial platform
A commercial platform can provide embeddable SDKs, OEM and white-label depth, and a managed-service option. Samvyo is one option in this category. It is based on SFU architecture and can fit a team that needs those SDK and product-ownership characteristics while retaining control of the media path, TURN, and recording processing and storage. It is a poor fit when the workload is tightly coupled to Microsoft Teams-aligned services or when the team wants to remain entirely within Microsoft's supported migration route.
Whichever path you choose, keep selection separate from sequencing. The replacement decision should be made against the inventory and acceptance tests; the rollout should still start with the earliest constraint and proceed through validated dependencies.
Bottom Line
Start an ACS migration with an inventory, then move retired capabilities and missing migration paths into design first. Treat Rooms as an early redesign constraint, validate the calling path before replacing its interface, and use the remaining ACS period to test surviving workloads before a phased cutover.
The final retirement date creates urgency, while the dependency graph determines the safe order. A staged migration with explicit rollback is more reliable than a single provider switch at the end of the transition.
What's Next
Sequencing tells you what to move first; the next decision is what you move to. Choosing an ACS Replacement: What to Check Before You Commit Again sets out the checks to run before committing to a new provider. Once the new stack is chosen, Mobile Web Video Beyond iOS: What Android Chrome Does Differently covers the Android Chrome test plan for its browser clients.
FAQs
When should I start migrating from Azure Communication Services?
Start with the capability inventory and target design before the final retirement date. Microsoft begins restrictions for some customers on October 23, 2026, while retired services are scheduled to be removed after September 30, 2028, so waiting until the final year compresses design, testing, and rollout.
Which ACS services should be migrated first?
Start with capabilities Microsoft lists as retired and with services that have no migration path, especially ACS Rooms. Then address dependencies that sit underneath customer-facing UI or workflows; surviving calling workloads can follow after their target route and feature behaviour have been validated.
Can I keep using ACS Calling until 2028?
Existing ACS Calling workloads continue to operate until September 30, 2028, according to Microsoft's retirement guide. Use that transition period for target validation and phased deployment, because continued operation through the date does not establish standalone support after retirement.
Should I replace the ACS UI Library before migrating calling?
Validate the calling path and its target scenario before replacing the UI shell. The UI work should be mapped against the calling events, state, recording, diagnostics, and automation behaviour in your own application; Microsoft's current guidance lists the Web and Mobile UI Library SDKs as retired and allows a customer-maintained fork during the transition.
Does ACS Rooms have a direct migration path?
No. Microsoft states that ACS Rooms and Teams meetings through the Microsoft Graph API exist in different environments and do not offer a migration path. Treat Rooms as a product and data redesign, then test the chosen meeting model before scheduling customer cutover.
What should I test before switching away from ACS?
Test authentication, provisioning, calling or messaging, media, reconnect, recording, captions, diagnostics, automation, permissions, data retention, monitoring, and rollback. Test customer journeys and failure cases as well as a successful session, and compare the replacement's events and records with the baseline captured before migration.
Can Samvyo be used as an ACS replacement?
Samvyo can be considered when the requirement calls for embeddable SDKs, OEM or white-label depth, and a managed-service option around an SFU-based architecture. It should be evaluated against the application's exact Teams-aligned, media, recording, TURN, identity, and deployment requirements rather than treated as a universal replacement.