How Much Does an AI Voice Agent Cost in the UK? A 2026 Price Guide

If you are pricing an AI voice agent for a UK business in October 2026, you will meet three very different kinds of number: a monthly plan that includes some minutes, a per-minute usage rate from a model provider, and a bespoke quote that says "scoped to your needs". They are not comparable until you put them on the same footing. This guide does that, using prices the suppliers themselves publish, and shows where the bill usually grows beyond the headline.

This article is about cost, not whether a voice agent is right for you. For that decision, start with AI voice agents for service businesses, and for a plain-English view of how they are built, read how AI voice agents are architected.

Three ways to pay for a voice agent

Almost every price you will see falls into one of three models.

Three cost models for an AI voice agent (prices as published on 10 October 2026)

Model	What you pay	Published examples	Main risk
Packaged monthly plan	A fixed fee with included minutes, then a per-minute charge	Voicecake £100 a month with 600 minutes, then £0.10 a minute. VoxHive £49 to £399 a month with 100 to 2,000 minutes, then £0.25 a minute. IntelliTeam £197 to £1,297 a month with 100 to 900 minutes, then 33p a minute.	Minutes are defined differently by each supplier. Setup and integrations may cost extra.
Bespoke or managed build	A scoped build fee plus a monthly fee	Voicecake bespoke from £800 a month. VoiceDart bespoke from £149 a month with the build fee scoped to the job. IntelliTeam setup from £500 to £3,000 or a bespoke quote.	Hard to compare until the scope is written down.
Assembled from model APIs	Usage rates from model and telephony providers, plus your own engineering	xAI speech-to-speech voice at $0.08 a minute. OpenAI GPT-Live 1 at $0.05 a minute, with backend model and tool usage charged separately.	The headline rate omits telephony, hosting, monitoring and the people who maintain it.

A few things are worth noticing. The packaged plans are cheapest at small volumes because the supplier has already built and hosts the platform. The bespoke options exist because a package cannot cover every system a business uses. And the raw API rates look tiny, which is exactly why they mislead: they are one layer of a larger bill.

A worked example: 600 calls a month

Prices are easier to judge against a concrete workload. Take an illustrative service business that handles 600 answered calls a month averaging three minutes each. That is 1,800 minutes of talk time. This is an assumption for the sake of the arithmetic, not a benchmark, so substitute your own call log.

Using only the published rates above, the monthly cost of the plan itself works out as follows.

Illustrative monthly cost of 1,800 minutes on three published UK plans

Supplier and plan	Calculation	Monthly cost
Voicecake Standard	£100 includes 600 minutes. 1,200 extra minutes at £0.10 is £120.	£220
VoxHive Scale	£399 includes 2,000 minutes, so no overage. Quoted excluding VAT.	£399
IntelliTeam £397 plan	£397 includes 300 minutes. 1,500 extra minutes at 33p is £495. Setup from £1,500 is separate.	£892

The spread, from £220 to £892 for the same minutes, tells you less about which supplier is better and more about how much the packages differ. One may include integrations, support or a named contact that another does not. One may measure a minute differently. The lesson is not to pick the cheapest line, but to find out what each line contains.

What the minute rate leaves out

Whichever route you take, a handful of costs tend to sit outside the rate card.

• Telephony. Phone numbers, call routing and carrier charges. Some packages include a UK number. If you assemble your own, you pay a carrier, and some carriers round short calls up to the next minute.
• Silence and ringing. Some services meter connection time rather than speech. A long pause on a call is still billed.
• Human handover. A transferred call can occupy a second telephone line and a member of staff at the same time.
• Integrations. Booking into your calendar or writing to your CRM is where most of the real work sits. Our piece on durable API integrations explains why this part deserves its own budget.
• Quality review. Someone has to listen to samples of calls and fix what the agent gets wrong. That is staff time, or a paid review feature.
• Recording and storage. If you record calls, ask about storage charges and the default retention period. One UK supplier states it deletes recordings after 90 days by default.
• Compliance. Telling callers they are speaking to an AI, and handling personal data properly, is a design requirement rather than an optional extra. Our guide to the EU and UK disclosure rules covers what to plan for.

A voice-agent consultancy, Softcery, publishes its own estimate that a production voice agent running on separate speech, model and telephony providers costs roughly $0.12 to $0.33 per connected minute in the United States once everything is added to the platform fee, and that fixed charges dominate the bill below about 5,000 minutes a month. These are one company's estimates in US dollars for US calls, not a UK quote, but the pattern they describe matches what the packaged plans show: the per-minute rate is only part of the story.

Cost per resolved call is the number to watch

A per-minute price tells you what the agent costs to run. It does not tell you what it achieves. The figure that matters to the business is the cost of a call that actually ended well.

cost per resolved call = monthly cost ÷ calls resolved without a person

This formula is illustrative. Using the Voicecake example from the table, £220 a month across 600 calls:

• If the agent resolves 60 per cent of calls (an assumed figure, not a benchmark), that is 360 calls and about 61p each.
• If it resolves 40 per cent, that is 240 calls and about 92p each.

Nothing about the supplier changed between those two lines. Only the quality of the conversations did. That is why the first month of a pilot is better spent measuring outcomes than haggling over a rate. Our own view is that a modest agent that resolves the call types it was designed for is worth more than a clever one that tries everything.

Build, buy or assemble?

There is no universal answer, but there is a sensible way to decide.

1. Buy a packaged plan when your needs are standard: answer calls, take messages, book appointments into a mainstream calendar. You get a working service quickly and the supplier carries the maintenance.
2. Commission a bespoke build when the agent has to work with your own systems, follow call flows specific to your business, or handle data in a way a package cannot. The monthly fee then pays for the platform and the build fee pays for the design, integration and testing.
3. Assemble from model APIs when you have in-house engineers who can own it, or a requirement such as data residency or custom telephony that platforms cannot meet. The saving per minute is real but it buys you a system you must run.

Softcery's own analysis reaches the same broad conclusion: self-assembly saves on the per-minute fee but only pays for itself once the saving exceeds the cost of the people maintaining it. For most small businesses, that point is a long way off.

Questions to put to any supplier

Take this list into every conversation and ask for answers in writing.

1. What exactly counts as a minute: all connected time, or only talk time on answered calls?
2. What is included in the monthly fee, and what is billed separately?
3. Is there a setup or build fee, and what does it cover?
4. What happens when we go over the included minutes? Is there a cap or a surprise?
5. Are phone numbers, call recordings and transcripts included, and how long are they kept?
6. How do you tell callers they are speaking to an AI?
7. Which of our systems will it connect to, and who maintains those connections?
8. What happens on a call the agent cannot handle?
9. What is the notice period, and can we export our data if we leave?
10. Do the prices include VAT?

Where SystemNexa fits

SystemNexa builds AI voice agents as scoped projects that connect to your own systems, alongside broader AI automation and API integrations. We do not publish a price list, so this article deliberately does not quote one. If you are weighing a package against a custom build, we are happy to look at your call log and tell you honestly which side of the line you are on.

Conclusion

An AI voice agent is cheap to start and easy to under-budget. Published UK plans make the entry price look small, raw model rates make the usage cost look tiny, and neither shows the integration, review and compliance work that decides whether the thing is useful. Price your own call volume, ask what each quote includes, and measure the cost of a resolved call rather than a minute. If it would help to talk it through, you can contact us about a voice agent.