The hardest part of leaving Cisco UCCE or Avaya is not choosing what comes next. It is moving a live operation whose worst possible maintenance window is the exact moment a customer needs help. The phones cannot go dark; payment calls must still be recorded; the VIP queue must still recognize a VIP; yesterday’s dashboard must still reconcile with tomorrow’s. That is why a large installed base can remain on legacy infrastructure long after its operators have stopped defending it.
The escape route is not a weekend of crossed fingers. It is a controlled transfer of traffic. Run two environments, expose the new one to progressively harder work and let evidence—not a launch date—open each gate. A practical sequence has four phases: prove a subset of queues in parallel, move digital channels, migrate voice by business unit, then decommission the old estate.
First, map the system nobody actually has
Most mature contact centers have an architecture diagram. Few have the whole operating truth. A holiday prompt may live in one admin’s calendar. A fraud queue may depend on a CRM field set by an overnight job. A toll-free number may be billed to a subsidiary acquired six years ago. Historical reports may count transfers differently from the executive scorecard. Migration begins by finding these discrepancies while they are merely awkward, not while customers are waiting.
Build an inventory at the level of a customer journey: entry number or digital endpoint, identity check, menu path, routing decision, agent group, transfer, recording, disposition, downstream data and report. Add volume, peak pattern, owner, compliance class and rollback route. This becomes the migration ledger. Every item needs a destination, a test and someone willing to say it is correct.
The safest migration is not one giant leap. It is a series of small, reversible promises.
Phase one: make a small promise
Stand up the cloud platform beside the legacy stack and choose pilot queues with real traffic but contained consequences: an internal help line, a modest regional service desk or a product group with predictable hours. Avoid the temptation to pick a queue so quiet that it proves only that login works. The pilot needs enough volume, transfers and exceptions to reveal the shape of production.
Keep number control conservative. Temporary forwarding or carrier routing can steer calls while permanent ports are prepared. Confirm caller ID, toll-free ownership, callback presentation, international reachability and the emergency reroute before the first live call. A port date is not a resilience design. It is merely the moment a carrier changes where traffic lands.
Routing deserves translation, not photocopying. Legacy logic accretes: nested skills compensate for old staffing choices; priorities fix yesterday’s outage; after-hours branches outlive the team that requested them. Write down the business intent—who should answer, under what conditions, after how long—then implement the simplest rule that preserves it. Test overflow, no-agent, disconnect and transfer failures as seriously as the happy path.
The pilot meets agreed service, audio, transfer and data thresholds across normal and peak periods—and the team has successfully rehearsed a rollback.
Phase two: let digital find the cracks
Chat, SMS, email and messaging usually move with fewer carrier dependencies than voice. That makes them an ideal second wave, not an afterthought. They force the new environment to prove customer identity, CRM lookup, transcript storage, concurrency rules, escalation and supervisor visibility while the legacy voice platform remains available.
Channel continuity is the test. A customer who begins in chat and escalates to a call should not have to perform the story again. UJET says its voice and digital suite can blend channels, accept customer photos and video, and exchange CRM data in real time. In a migration, those capabilities are not brochure items; they become acceptance criteria. Does context arrive with the interaction? Does the transcript land in the right customer record? Can an agent request richer information without creating a shadow workflow?
Phase three: move voice where the business lives
Voice is the consequential wave because it joins telephony, people and policy. Move it by business unit, geography or customer segment—groups that share leaders, schedules, service levels and failure consequences. A coherent unit can own a go/no-go decision. A technically convenient assortment of queues usually cannot.
Train agents close to their wave. A brief sandbox session followed weeks later by production is theater; memory fades and local workarounds return. Use real scenarios, include transfers and disposition codes, then give the first shifts a staffed support channel with operations, telecom, CRM and vendor experts in one place. Supervisors need separate practice in monitoring, queue intervention and exception handling.
Reporting must run in parallel too. Preserve raw legacy exports, metric definitions, time zones and hierarchy mappings. During overlap, publish three views: old-platform traffic, new-platform traffic and a reconciled enterprise view. Never splice two “average handle time” fields together until analysts confirm that both start, pause and end the clock the same way. A beautiful trend line built from incompatible definitions is worse than a visible break.
Speed changes the economics. UJET says its average implementation is two months, compared with an industry range of six to 18 months for enterprise deployments. A shorter build can reduce duplicate licensing, split administration and the fatigue of living between systems. But it should shorten the overlap, not erase the proof required to leave it. Calendar speed is useful only when operational confidence keeps pace.
Phase four: make the old platform unemployed
Decommissioning begins before the last voice wave. For every dependency in the migration ledger, capture an owner and an exit condition. Recordings and transcripts need accessible archives with tested search and legal retention. Workforce tools, quality systems, bots, payment flows and data warehouses must stop calling legacy interfaces. Finance should know when carrier trunks, maintenance and data-center costs can end—but operations decides when the platform has no work left.
After the final traffic move, hold a zero-traffic observation period. Watch for calls arriving on forgotten numbers, nightly jobs throwing errors and analysts reaching for old reports. Archive configuration and routing logic. Remove privileged accounts and network paths in a controlled sequence. Keep defect ownership alive beyond the celebration; migrations create delayed failures when a monthly process or rare journey runs for the first time.
The five controls that travel through every phase
- Numbers: ownership, port batches, caller ID, forwarding and emergency reroutes.
- Routing: documented business intent, edge cases, overflow and rollback behavior.
- History: raw exports, metric definitions, retention, access and reconciled reporting.
- People: just-in-time training, supervisor drills and first-shift floor support.
- Failure: thresholds, named decision makers, rehearsed reversal and a written incident path.
The essential discipline is almost unfashionably simple: do not move the next slice because the project plan says Tuesday. Move it because the present slice is healthy, observable and reversible. That rhythm turns a frightening replacement into a series of operating decisions. Eventually Cisco UCCE or Avaya is still powered on, still technically available—and doing nothing. That is the quietest possible ending, which is exactly what a contact center migration should earn.
Sources and further reading
Product capabilities and deployment figures are attributed to UJET. Migration controls and sequencing are category-level operational guidance.