What is a cloud PBX?

A PBX is the switchboard that connects your company's phones to each other and to the outside world. A cloud PBX is that same thing run as software on servers you do not own, reached over the internet, so there is no hardware in your office and no engineer visit to change a setting. Your desk phones, softphones and mobile apps register to it from anywhere, and your phone numbers point at it rather than at a building.

Karan Sharma · Updated 2026-10-29

What the box used to do

Before anything was in the cloud, a PBX — Private Branch Exchange — was a physical unit in a cupboard. Lines from the phone company came in, extension wiring went out to each desk, and the box decided which internal phone rang when a call arrived, let people transfer calls to each other, and made internal calls free because they never left the building.

Everything a phone system does today is still that list. The cloud did not add the features; it moved where they run and changed who maintains them.

Which matters more than it sounds. When the switchboard was a box, every change was a visit: a new starter meant wiring an extension, a changed opening hour meant reprogramming hardware, and a second office meant a second box with a leased line between them. The features were similar. The cost of using them was not.

Five words you need before comparing anything

Most confusing vendor conversations come from these being used loosely. They are not interchangeable, and pricing usually hangs off two of them.

TermWhat it actually is
ExtensionAn internal identity — a person, a desk or a department. 201, 202, reception. Usually what you are billed per.
DIDA real phone number from the outside world that points at your system. Also called a virtual number. You may have far fewer DIDs than extensions, or more.
TrunkThe connection between your phone system and a carrier, over which calls reach the public network. A SIP trunk is that connection carried over the internet.
ChannelOne simultaneous call. Thirty staff who are never all on the phone at once might need eight channels, not thirty.
SoftphoneA phone that is an app rather than a handset, registering to the same system as a desk phone and behaving identically.

What moving it to the cloud actually changed

Three things, none of which is "more features":

  • Changes became configuration instead of visits. Adding a person, changing a menu, moving a number to a different team — all are now a form, and that is most of the saving in practice, not the licence fee.
  • Location stopped mattering. An extension registers over the internet, so a home desk, a branch office and a mobile are the same thing to the system. This is why remote work did not break phone systems that had already moved.
  • Capacity became elastic in one direction. Adding channels is a setting rather than a physical line, so a seasonal spike no longer needs provisioning weeks ahead.

What did not change: the call still has to reach the public telephone network, and somebody still has to have a commercial relationship with a carrier to make that happen. That relationship is the part buyers most often do not realise is separable — see the guide on bringing your own carrier.

Cloud, on-premise, and the honest trade-off

The usual comparison is a sales table where cloud wins every row. The real trade-off is narrower than that, and it is mostly about who carries which risk.

On-premiseCloud
Up-front costHardware, installation, often a leased lineLittle or none
Ongoing costMaintenance, and someone who understands itPer extension, monthly
Making a changeOften a visit or a specialistA setting
If your internet dropsInternal calls usually keep workingEverything stops
If the provider has an outageNot applicableEverything stops
Where recordings liveYour buildingWherever the provider says

Read the two "everything stops" rows together, because they are the honest cost of the model. A cloud PBX concentrates your failure modes into connectivity and a supplier. That is usually a good trade — a single well-run provider is more reliable than a cupboard nobody has patched in four years — but it is a trade, and the mitigation is boring and specific: a failover route for your numbers that does not depend on the same system.

How to work out what size you need

Two numbers, and people usually get the second one wrong by a factor of three.

  1. Count extensions. One per person who needs to be reachable individually, plus one for each shared endpoint — reception, a warehouse handset, a meeting room. This is almost always just your headcount in roles that answer calls.
  2. Count concurrent channels. Not headcount. Look at your busiest hour and ask how many conversations were happening at the same moment. For most offices this is between a quarter and a third of extensions; for a sales or support team it is higher.
  3. Add headroom to the channel number, not the extension number. Running at exactly your peak means the next caller above peak gets a busy signal, and that is the one call you cannot afford to lose.

If you have an existing system, it already knows both numbers. Pull a report of simultaneous calls over a month rather than estimating — the estimate is nearly always too high, and you will be quoted against it.

What actually goes wrong

Not the things the brochures address. In rough order of how often they bite:

  • Audio quality blamed on the provider when it is the network. Voice is unforgiving about jitter and packet loss in a way that video calls and file downloads are not. A connection that feels fine for everything else can still be bad for calls.
  • Number porting taking longer than anyone promised. Moving existing numbers to a new provider is an administrative process involving your old carrier, who has no incentive to hurry. Plan for it to be the slow part of any migration.
  • Nobody owning the dial plan. Six months in, calls route through rules three different people added and nobody can say why. This is a documentation problem, not a software problem, and it happens on every platform.
  • Discovering the recording retention policy during an incident rather than before one.

Questions worth asking any vendor

  • Can I keep my existing numbers, and what is the realistic porting timeline?
  • Am I billed per extension or per user, and what counts as a user?
  • Can I bring my own carrier, or must I buy calling from you?
  • Where are recordings and call records stored, for how long, and can I export them?
  • What happens to a call in progress during an outage, and where do my numbers fail over to?
  • Is this multi-tenant, and if so, what isolates my account from another customer's?

The third and the fifth separate vendors more than any feature list will. A provider that cannot answer the fifth has not thought about the failure mode that matters most to you.

See it on OwnPBX