How we calculate our numbers.
Every cost and deflection figure on this site comes from a model, not a slogan. Here are the assumptions behind it — so you can pressure-test them against your own operation.
≈ £3.20
A representative fully-loaded cost per handled call in a UK call centre. Your number depends on wages, occupancy, and average handle time (AHT) — swap in your own.
- Loaded agent cost
Salary + employer NI + pension + benefits
- Supervision & QA
Team leads, workforce management, coaching
- Facilities & tooling
Seat, telephony licence, CRM, office overhead
- Shrinkage
Breaks, training, absence, attrition
≈ £0.18
A representative production cost for an automated tier-1 call in 2026, summed from the per-call components below. Unlike the human figure, this number keeps falling as models get cheaper.
- Speech-to-text (STT)
Per-minute transcription of the caller
- LLM tokens
~£0.04 per call (frontier model, ~4-min average)
- Text-to-speech (TTS)
Per-character synthesis of the agent
- Telephony / SIP
Inbound minutes + carrier / SBC
- Infrastructure
Orchestration, RAG store, logging, evals
Where the headline numbers don't apply.
- Per-call unit cost ≠ blended saving
The ~94% gap between £3.20 and £0.18 is the unit cost of an automated call. Your blended saving is lower because you only automate the calls the AI can actually handle.
- Realistic deflection is 40–60%
Tier-1 (short, common, well-defined) calls deflect well. Complex cases still route to humans — and you still pay for the calls where the AI escalates.
- ~60% operational cost cut
The −60% figure we quote across the site is a blended operational number at realistic deflection — not the raw per-call unit gap. It scales with inbound volume.
We'll run this model against your real AHT, wage, telephony, and volume numbers in a scoping call — and show you the working, not just the conclusion.