Case study
Unified communications across a multi-site organisation
A multi-site voice platform where the dial plan, not the technology, was what would age badly, and where the call quality complaints traced to the data network rather than to anything anyone calls voice.
The dial plan first, with room built in and site codes that map to routing, then CUCM clusters tested by failing the publisher, local breakout at each site, and QoS enforced from the access edge.
A branch that loses its WAN keeps local and external calling instead of going silent, new sites have been absorbed without restructuring, and the call quality fault was closed by correcting a switch, not the platform.
A unified communications deployment across a multi-site organisation: CUCM clusters, voice gateways, SIP trunking and a dial plan designed to still make sense after the next ten sites are added.
The problem
Most dial plans are built for the organisation as it is on the day of the migration. Two years later, extensions collide, three sites share a prefix by accident, and adding a branch requires touching translation patterns in four places. The structure, not the technology, is what ages badly, and by the time it shows, the platform is already carrying every call the organisation makes, so the cost of restructuring is measured in change windows nobody will approve.
The other half of the problem is that a voice rollout is judged on call quality, and call quality complaints after one are almost never a voice problem. They arrive as a report that calls break up on some routes and not others, they are raised against the new platform because that is what changed, and the cause is usually somewhere in the data network that nobody thought to look at.
What I did
The dial plan came first: a numbering scheme with room built in, site codes that map to routing rather than to history, and partitions and calling search spaces that express policy instead of accumulating exceptions.
- CUCM cluster with publisher and subscribers sized for the user count plus growth, and redundancy tested by failing the publisher deliberately.
- Voice gateways at each site, with local breakout so a WAN failure degrades a branch to local and external calling rather than silence.
- SIP trunking to the provider, with the codec and transcoding policy decided explicitly rather than left to negotiate.
- End-to-end QoS: marking at the access edge, trust boundaries enforced, and queuing on every WAN link that voice traverses.
The part that is never in the plan
On this project the worst of the call quality complaints traced to an access switch where the trust boundary had not been applied, so marked traffic was being re-marked to best effort on the way into the network — three hops from anything anyone would call a voice system.
Finding that kind of fault is why the network and the voice platform should not be two separate conversations. Nothing in the voice configuration was wrong, and no amount of work inside it would have found the cause.
The result
A voice platform that survives a site losing its WAN, a dial plan that has since absorbed new sites without restructuring, and a documented QoS policy that new switches inherit rather than miss.
The call quality complaints were closed by correcting a switch configuration rather than by changing the voice platform, which is the outcome worth aiming for: the alternative is rebuilding something that was never at fault and still having the problem afterwards.