Sanitized case study
A Phone Number Is Not a Phone Number
A sanitized Teams Phone Direct Routing case study about inventory, coexistence, failback, and the operational work that makes a migration succeed.
BLUF: Moving a legacy voice environment to Teams Phone is not primarily a phone-replacement exercise. It is a business-process migration with SIP attached. The work succeeds when discovery exposes the real call flows, coexistence is deliberate, failback is tested, and operations receives something it can actually support.
The result
The useful outcome was not simply getting calls to ring in Teams. It was establishing a migration method that surfaced hidden dependencies before they became cutover surprises.
That meant treating every number as a possible application, using the Oracle SBC as the session-border control point and the Oracle ECB as the SIP proxy and routing decision point, and judging completion by caller experience and operational readiness rather than by a changed configuration screen.
This account is sanitized. It contains no employer identity, production addresses, real phone numbers, customer information, or proprietary routing details.
The situation
The starting point was a legacy voice environment shaped by years of business requirements. Users and extensions were only the visible portion. The environment also included main numbers, queues, auto attendants, voicemail, overflow paths, analog devices, emergency-calling considerations, reporting, and carrier services.
The target was Microsoft Teams Phone with Direct Routing. The business still had to remain available while old and new platforms coexisted.
A one-sentence summary—“move the phones to Teams”—was technically true and operationally inadequate.
The first decision: inventory before design
The earliest design mistake would have been drawing the target architecture before understanding the existing service.
Instead, the inventory started with questions:
- Who owns this number when its behavior needs to change?
- What should a caller experience during business hours?
- What happens after hours, on holidays, or when nobody answers?
- Does the call overflow to another team, a queue, voicemail, or an external service?
- Is recording, reporting, or a compliance process attached to the flow?
- What is the expected behavior during a WAN, carrier, SBC, or platform failure?
- Which dependencies are intentional, and which merely exist because nobody has questioned them?
The answers turned a list of numbers into a list of business services.
A sanitized example
A department might describe a destination as “our main number.” Discovery reveals something else:
- The published number reaches an auto attendant.
- A menu choice sends the caller to a queue.
- An unanswered call overflows to a second group.
- After-hours calls take a different path.
- A holiday prompt changes the experience again.
That is not one number with a forwarding rule. It is a small application with owners, schedules, failure modes, and caller expectations.
Once that becomes visible, the migration discussion changes. The question is no longer “Can this number ring somewhere else?” It becomes “Which parts of this behavior must survive, which should be simplified, and who is allowed to make that decision?”
The options
Three broad approaches were available.
Rebuild everything exactly as it was. This preserves familiar behavior, but it can also carry years of accidental complexity into a new platform.
Simplify aggressively during the move. This sounds efficient, but it combines platform risk with business-process change. Strange legacy behavior is not automatically useless; sometimes it exists because a real operational problem needed solving.
Use inventory-led selective redesign. Preserve behavior that serves a current requirement, simplify what can be explained and approved, and defer decisions that need a business owner.
The third approach was the sound choice. It reduced unnecessary replication without using the migration as cover for unapproved process changes.
The architecture decision
Direct Routing made the Oracle SBC more than a piece of infrastructure passed on the way to Teams. It was the session-border control point where carrier connectivity, interoperability, and failure behavior had to be understood as one system.
Between the SBC and the two voice platforms, the Oracle ECB served as the SIP proxy and routing engine. That made it the decision point for whether a destination remained on legacy voice or moved to Teams Phone during coexistence. The SBC controlled the border; the ECB applied the routing decision.
- CallerPSTN
- CarrierSIP service
- Oracle SBCSession border
- Oracle ECBSIP proxy + routing
- Teams PhoneDirect Routing
- User or queueBusiness flow
Coexistence / failback: the Oracle ECB preserves a tested routing decision back to legacy voice.
The product names identify responsibilities, but they do not replace design. The important questions live between the boxes:
- Which platform owns each route during coexistence?
- How does a call reach a user or queue after migration?
- What changes when a destination is unavailable?
- Where can support observe signaling and establish what actually happened?
- How can traffic return to a known-good path if the new path fails validation?
Those are design and operations questions, not diagram decoration.
Coexistence and failback
A phased migration means the old and new environments share responsibility for a while. That needs explicit routing ownership. A number should not be “kind of moved,” with two platforms each assuming the other handles the failure case.
The coexistence design therefore needed a clear state for every migrated destination: legacy-owned, transitioning, or Teams-owned. The routing decision at the ECB and the operational record had to agree.
Failback received the same treatment. A useful failback plan identifies:
- The trigger for declaring that a migration attempt failed.
- The person authorized to make that decision.
- The exact routing change required to restore service.
- The validation calls that prove restoration.
- The evidence to preserve before troubleshooting continues.
A rollback that depends on remembering yesterday’s configuration under pressure is not a plan. It is an anecdote waiting to happen.
Testing the caller’s experience
Configuration validation was necessary but insufficient. The test matrix followed the caller through normal and abnormal conditions.
For each representative flow, testing covered business-hours routing, after-hours behavior, no-answer and busy conditions, queue or voicemail delivery, overflow, and the relevant failover path. Migration validation also confirmed that the intended platform owned the route and that support could see the expected signaling.
Emergency calling required its own explicitly owned validation process. It could not be treated as just another successful test call.
The most useful tests were often the ones derived from discovery: the unusual holiday path, the second overflow destination, the shared voicemail dependency, or the analog device everyone assumed had been retired.
Operational readiness
A migration is not complete when the last number moves. It is complete when the service can be run without depending on the migration engineer’s memory.
The handoff needed usable call-flow documentation, named ownership, monitoring, known-good reference behavior, and a troubleshooting path that began with evidence rather than random changes.
Support needed to answer basic questions quickly:
- Which platform currently owns this destination?
- What is the expected caller experience?
- Where can the call be observed?
- What is the approved escalation or failback path?
Documentation that cannot answer those questions is mostly a souvenir from the project.
What this changed
The inventory-first approach moved difficult conversations earlier, when they were cheaper. Business owners could discuss behavior before a cutover. Engineers could distinguish intentional requirements from inherited clutter. Support could participate in the target state before inheriting it.
It also changed the definition of success. The goal was not a dramatic launch day. The goal was a routine day afterward: callers reached the right destination, users could work, support could troubleshoot, and nobody needed to reconstruct the design from a chat thread.
What I would do differently next time
I would involve business owners even earlier. Technical discovery can reveal what a call flow does, but not always why it matters or who has authority to simplify it.
I would also separate “strange” from “unnecessary” more deliberately. Legacy workarounds deserve skepticism, not automatic deletion. Some are technical debt. Others are the scar tissue from a real outage or operational requirement.
The engineering job is to learn the difference before choosing the target design.
The reusable checklist
For the next voice migration, the starting checklist is short:
- Inventory the behavior, not just the number.
- Assign a business owner and a technical owner.
- Record normal, after-hours, overflow, and failure behavior.
- Decide what to preserve, simplify, or retire.
- Define routing ownership during coexistence.
- Document and test failback before cutover.
- Test representative caller journeys, including odd ones.
- Hand support a service it can observe and operate.
The phones still matter. They are simply not the whole system.