Self-hosted PBX vs hosted telephony: the cost crossover for contact centres

01Where per-seat pricing stops working

Hosted platforms charge per agent per month, which is excellent value for twenty seats and a significant annual cost at two hundred. Because agent counts in contact centres fluctuate seasonally, you are also often paying for peak capacity year-round or renegotiating repeatedly.

The second pressure is functional rather than financial. Dialer behaviour, custom routing logic, deep CRM integration and call flows specific to your operation are where hosted products stop bending. Teams usually hit that wall before they hit the cost wall.

02What the infrastructure actually is

A production voice platform is usually a signalling layer and a media layer separated deliberately - a SIP proxy handling registration and routing, with a media server behind it doing call handling. That split matters because registration load and media load scale differently, and combining them is what makes platforms fall over as concurrency grows.

Then carrier trunks, a session border controller, and recording storage. None of this is exotic, but all of it needs someone who understands SIP - and that skill is genuinely scarce, which is the real constraint on self-hosting rather than the software cost.

• SIP proxy for signalling and registration.
• Media server for call handling and recording.
• Carrier trunks with failover and least-cost routing.
• Session border control and NAT traversal.
• Recording storage with retention and encryption.

03The cost nobody quotes: toll fraud

Toll fraud is the expensive failure mode in self-hosted voice. A misconfigured system discovered by an automated scanner can generate a very large bill overnight through premium-rate destinations, and carriers will generally expect you to pay it.

The controls are straightforward and must exist from day one: destination allow-lists, per-account spend caps, rate limiting, and alerting on unusual patterns. Any vendor proposing a voice platform without discussing fraud controls in the design conversation is not one to hire.

04Quality is a network problem

Most call quality complaints on self-hosted platforms are network issues, not switch issues - jitter, packet loss and poorly prioritised traffic. Monitoring quality per call leg rather than waiting for complaints is what separates platforms that get better from ones that accumulate unexplained problems.

Codec choice matters more than people expect. The instinct is to pick the most compressed option to save bandwidth, but on a network with any loss a less aggressive codec sounds noticeably better than a heavily compressed one struggling.

05The hybrid that works

Self-host the contact centre where per-agent economics and call-flow control matter, and keep standard business phones on a hosted service where they are cheap and nobody needs custom behaviour. There is no requirement to run one platform for both.

This is usually the right answer for organisations with a large agent operation and a normal office. You take on operational complexity only where it buys you something, rather than everywhere for consistency's sake.

• Self-host where per-agent cost or call-flow control matters.
• Keep office phones hosted - they are cheap and undemanding.
• Do not run one platform for both just for tidiness.
• Budget SIP expertise, which is the real scarce input.