Business IP telephony

A telephony migration is judged on the first inbound call that fails to connect.

Parent division: Networks and communications

The problem

Telephony is the only service whose every defect is observed by someone outside the organisation.

Slow email is noticed internally. A call that does not connect is noticed by the citizen, the supplier or the ministry that was calling. That is what makes a telephony migration politically riskier than it is technically difficult.

The risk almost never comes from the software. It comes from a numbering plan carried over unchanged without anyone checking which numbers are actually published, from guard posts and emergency lines missing from the survey, and from the system’s behaviour when the carrier link goes down.

Well-designed telephony keeps working inside a site when the carrier link is cut, and still routes emergency calls. Both properties are decided in the architecture; added afterwards, they usually mean reworking the connection.

A telephony migration and its two breaking points, schematic

01Survey of what exists: Handsets, lines, diversions, hunt groups, and the numbers published externally — including printed ones.
02Numbering plan: Consistent across sites, mapped to public numbers, historic numbers retained.
03Call core: Redundant or hosted, automated handset provisioning, administration console.
04Handsets and power: Power over the network: the switches feeding the handsets are backed up, or telephony falls with the mains.
05Carrier access: Gateway, links, number portability and contractual exit terms.
06Survivability and emergency calls: The site keeps calling when the link fails, and the emergency call carries the exact location.

Schematic. The accented path marks the two properties decided in the architecture, which cannot be added afterwards.Typical figures. The end-to-end delay is the conversational quality threshold of ITU-T recommendation G.114; the power figure is that of a first-class Power over Ethernet port.

Scope

What the service covers

Survey of what exists: Inventory of handsets, lines, diversions, hunt groups and externally published numbers, including those printed on documents and signage.
Numbering plan: Internal numbering consistent across sites, mapping to public numbers, diversion rules, and treatment of historic numbers that must be kept.
Call infrastructure: Redundant or hosted call server, gateways to carrier links, handsets and softphones, automated handset provisioning.
Network and quality of service: Logical separation of voice, marking and prioritisation of flows, link sizing, power over the network for handsets, and backed-up power for the switches.
Continuity and emergency calls: Site survivability when the carrier link breaks, a fallback path, and routing of emergency calls carrying the exact location of the calling handset.
Operation: Administration console, creating and removing handsets, call statistics, recording where it is required, and documentation that lets the in-house team hold the service.

Situations

The most frequent situations

Replacing an end-of-life exchange: The hardware is no longer maintained and spares are bought second-hand. The constraint is keeping published numbers and running both systems side by side during the cutover.
Branch network: Several sites, one numbering plan, internal calls that must not transit the carrier, and each branch remaining reachable if its link fails.
Public-facing administration: Switchboard, queues, transfers to departments, and monitoring of the answered-call rate. The service delivered is measured, not asserted.
Remote site: A single link, sometimes cut. Internal telephony must continue without the carrier link, and emergency calls must still be routed by a separate path.

Requirements

What to require, of us as of anyone

These requirements hold whichever supplier is appointed. Written into a tender, they filter out the responses that will not hold.

A complete survey of externally published numbers, before migration, including those on printed material.
The system’s behaviour when the carrier link fails, and when the call server fails. The two cases are distinct.
Emergency call routing and the location transmitted, handset by handset, on every site.
Backed-up power for the switches that feed the handsets: without it, telephony falls with the mains.
The cutover plan, with its window, its point of no return and its rollback procedure.
The ability for the in-house team to create, change and remove a handset without going through the provider.
Number portability and the contractual exit terms with the carrier.

Pitfalls

Common mistakes, and what they cost

Cutting over every site on the same night: A simultaneous cutover leaves no reference site to compare against and no credible rollback. A phased cutover costs a period of coexistence and avoids a general outage.
Forgetting the handsets that are not handsets: Lifts, gates, alarms, intercoms and fax machines use analogue lines. They appear in no directory and are discovered the morning after the cutover.
Neglecting quality of service on the network: Voice tolerates jitter and packet loss badly. A network that performs perfectly for office work degrades voice as soon as a file transfer occupies the link.
Treating emergency calls as a detail for later: It is the one call whose failure cannot be defended. It is addressed on the first day of design, with the exact location passed to the service that answers.

Questions

Questions asked before consulting

Host the call server or take it as a service?

On-site hosting keeps internal telephony working with no external link, and suits organisations whose continuity cannot depend on a connection. A hosted service reduces the operating burden and suits multi-site organisations with reliable links. The deciding criterion is the expected behaviour when the link fails, not the monthly cost.

Can existing numbers be kept?

Keeping public numbers depends on the carrier and on the applicable rules; it is verified before the schedule is settled, not after. Internal numbering can always be kept, and almost always should be: a changed plan forces every directory, every sign and every habit to be reworked.

How much bandwidth does voice need?

A single call needs little; what matters is the stability of the link and the number of simultaneous calls actually observed at peak. Sizing is done against that number, measured on the existing installation, and not against the number of handsets.

What happens during a power cut?

Handsets powered over the network go down with the switch that feeds them. Backed-up telephony therefore means backing up the access switches, the call server and the carrier gateway, for a duration settled at design stage. Otherwise telephony disappears at the moment it is most useful.

How should remaining analogue lines be handled?

Through gateways, once they have been listed. Lifts, alarms and gates carry obligations of their own, and their line must stay usable independently of the rest. They are the most frequent cause of post-migration incidents, precisely because they appear in no directory.

Can calls be recorded?

Technically yes. Recording processes personal data and subjects the organisation to informing the parties beforehand, to a declared purpose, to a retention period and to a restricted access regime. It is put in place for a specific reason, over a defined scope, rather than by default across every handset.

Evidence

Where we have applied it

Finance and fraud prevention2024
Branch network — interconnection and enterprise telephony
Twenty-two branches linked to head office, with unified telephony, backup paths and common supervision.
Branches interconnected
22
Branches on duplicated links
8
Reference KP-2024-019