The short answer: a single-purpose AI agent — one that answers customer emails, qualifies leads or produces quotes — is quoted per project, because the work depends almost entirely on what it has to connect to and what it's allowed to do. Larger platforms where agents work across several channels start at £20,000. On top of the build, you pay running costs: the AI provider's usage fees plus hosting.
That's an honest answer, but not a very useful one on its own. So here's what actually drives the price, what running costs look like, when an off-the-shelf chatbot is the better buy, and what to ask any supplier before you commit.
First: what you're paying for
A chatbot answers questions. An agent can also take actions — look up an order, update your CRM, send a quote, book a call — using the tools and permissions you give it. That's where the time savings come from, and it's also where the cost is. The language model itself is the cheap part; the work is in wiring it safely into your real systems.
What drives the cost of an AI agent
- What it has to connect to. An agent that only reads your website and FAQs is simple. One that reads your inbox, checks your order system and writes to your CRM needs an integration for each of those — and each one has to be tested.
- What it's allowed to do on its own. Answering questions is low-risk. Sending quotes, issuing refunds or changing records needs guardrails: limited permissions, spending limits and human approval for anything risky. Guardrails are worth paying for.
- How many channels. Website chat, email, WhatsApp and Telegram each add work. A multi-channel agent platform is a different size of project from a single inbox assistant.
- How it hands over to a person. Good agents know when to stop and pass the conversation to a human — with the full history attached. That handover needs building, not just a "contact us" link.
- Logging and visibility. Every action should be logged in a way a non-engineer can read, so you can see exactly what the agent did and why. If something goes wrong, that log is how you find out.
- Volume and reliability. Ten conversations a day and ten thousand need different queues, retries and monitoring.
Running costs
Once it's live, an agent costs the AI provider's usage fees plus hosting. For most business agents that's small next to the hours it saves — but it depends on volume, so it should never be a surprise. We estimate running costs up front and set usage limits, so there are no surprise bills.
If a supplier can't tell you roughly what it will cost to run, or won't put a cap on it, treat that as a red flag.
Off-the-shelf chatbot or custom agent?
Off-the-shelf chatbot tools are cheaper to start and fine for answering common questions from your website content. The trade-offs: they rarely take real actions in your systems, you rent rather than own, and the monthly fees keep coming.
A custom agent costs more up front, but it works inside your actual processes — your inbox, your CRM, your pricing rules — and you own it.
A simple test: if the job is "answer questions", start with a tool. If the job is "do the work" — quote, update, chase, book — you're looking at an agent.
A worked example (hypothetical)
Say someone on your team spends ten hours a week copying enquiry details from email into your system and sending standard quotes. An agent that drafts those quotes for a person to approve gives most of that time back every week, permanently. Whether that's worth the build cost is easy to check once you have a written number — which is why we quote per project rather than guess.
What to ask any supplier
- Which AI models do you use, and does my data train them? We mostly use Anthropic's Claude and OpenAI's GPT models, through business APIs that don't use your data to train their models by default.
- What can the agent do without a person approving it? You should decide this, not the supplier.
- Can I read a log of everything it did? If not, you can't trust it.
- What happens when it doesn't know the answer? It should hand over, not guess.
- What will it cost to run, and is there a cap?
- Who owns the code and the accounts?
Red flags
- Promises of a fixed percentage saving before anyone has looked at your process.
- "Fully autonomous" with no approvals, limits or logs.
- No answer on running costs.
- A platform you can't leave without losing everything you've built.
How we price AI agents
- Single-purpose agents: quoted per project, with a fixed written estimate within 48 hours of your brief.
- Multi-channel agent platforms: from £20,000, with milestone-based payments agreed up front.
- Built on Claude and GPT, connected to your tools via APIs and MCP, with human sign-off where it matters and readable logs.
You can see what's included on our AI agents page and how our bands compare on the pricing page. If the work is mostly moving data between systems rather than holding conversations, business automation may be the cheaper fit.
Want to see one working? The chat assistant on this website answers questions, hands over to a person when asked, and can book a call — it's the same kind of agent.
FAQ
Is it safe to let AI act on our systems? Only with guardrails: limited permissions, spending and rate limits, human approval for anything risky, and a full log of every action. You decide what it's allowed to do on its own.
Will an agent replace staff? It takes the repetitive volume — triage, data entry, standard replies — so people can spend their time on decisions and customers. People stay in charge.
How long does it take to build? It depends on the integrations. Your written estimate includes milestones and a timeline before you commit.
What's the difference between an AI agent and a chatbot? A chatbot answers questions. An agent can also take actions in your tools — and that's where the time savings come from.
Ready for a real number? Send us a brief and get a written estimate within 48 hours, or book a free 30-minute call to talk it through.