AI Chatbot Cost in the UK: What You Should Actually Budget

AI chatbot costs in the UK fall into three bands: off-the-shelf tools from £0 to £500 to set up, platform builds from £1,500 to £8,000, and custom development from £8,000 upwards. Setup and monthly running costs are separate budgets. Volume and integrations move the price more than anything else.

Most businesses asking about chatbot cost get the same non-answer: it depends. It does depend, but the ranges are knowable, and you can size a budget before you speak to a single supplier. An AI chatbot in the UK costs anywhere from nothing at all to £30,000 and beyond, because one word covers three very different products sold to three very different buyers.

The useful question is which of the three tiers matches the job you need done, and what that tier costs to set up against what it costs to keep running. Every figure below is a range with its basis stated, and all prices exclude VAT. For the scoping detail behind these numbers, ProfileTree’s AI chatbot services page sets out what a build involves in practice.

AI chatbot cost by tier: what each band actually buys

The table below shows the three tiers as UK buyers encounter them. Ranges are drawn from published vendor list pricing and published UK agency project pricing at the time of writing, not from private quotes.

An off-the-shelf tool is a subscription product you configure yourself. You paste in a list of questions and answers, drop a widget on the site, and you’re live the same afternoon. The setup cost is mostly your own time. It works when your enquiries genuinely repeat, and the answers don’t change much.

A platform build is where most SME money goes. An agency takes an existing chatbot platform, trains it on your actual content (service pages, policy documents, pricing sheets, past email replies), writes the prompt behind it, tests the failure cases, and connects it to one or two systems. You’re paying for the training work and the judgement, not the software licence. Implementing AI chatbots for SMEs covers what that process looks like from the inside.

Custom development means code you own, running against a model API, usually with a retrieval layer over your own documents. It’s the right answer when the assistant has to write to a system rather than only read from one, when your data can’t sit on a third-party platform, or when conversation volume makes per-message platform fees more expensive than hosting your own.

Four things that move the price of a chatbot

Two businesses can ask for “a chatbot on the website” and get quotes £15,000 apart. The gap almost always comes down to four variables, and you can estimate your own position on each before anyone quotes.

Integrations

This is the single biggest lever. A bot that answers from a document set is cheap. A bot that checks live stock, looks up an order, books a slot in a calendar or writes a lead into a CRM has to authenticate against that system, handle its errors, and behave sensibly when it’s down.

Read-only lookups are the cheaper half. Read-write access, where the assistant changes something in a live system, carries testing and rollback work that typically doubles the integration line on a quote. Count your integrations honestly. Three is not one plus two more; each one adds its own failure modes.

Training content

Chatbots are trained on what you’ve already written. If your policies, pricing, and process are documented in a reasonable state, training is fast. If the knowledge lives in three people’s heads and a decade of email, someone has to write it down first, and that’s a content project sitting inside your chatbot project.

Buyers are often surprised that the largest line on a quote is content preparation rather than engineering. It’s also the line with the best return, because documented answers help your website, your staff onboarding, and your search visibility at the same time.

Languages

Each additional language adds translation of the source material, separate testing, and separate quality review, because a model that reads well in English can produce stilted or wrong output in Irish or Polish without anyone noticing for weeks. Budget roughly 30% to 50% of the original build cost per additional language on a platform build, based on the translation and retesting effort involved. Businesses serving cross-border customers should read AI chatbots for Irish customer service before committing to a multilingual scope.

Volume

Volume is the variable that turns a cheap chatbot expensive after launch. Model providers charge per token, so cost scales with conversations. Published API pricing puts per-message costs in the region of £0.001 to £0.01 depending on model and message length, which is trivial at 500 conversations a month and material at 50,000. Messaging channels add their own fees; WhatsApp Business conversations carry a per-conversation charge in the region of £0.03 to £0.05 in the UK on published rate cards.

Setup cost and running cost are two different budgets

The most common budgeting mistake is treating a chatbot as a one-off purchase. Finance approves a build cost, the project ships, and nobody owns the monthly line. Twelve months later, the assistant is quoting last year’s prices and referring people to a service that no longer exists.

Running cost has four components, and only the first appears on most quotes. There’s the platform or hosting fee, which is predictable. There’s model API usage, which scales with how well the thing works. There are channel fees if you’re on WhatsApp or SMS. And there’s the human time to review transcripts, correct wrong answers, and update the source content when your business changes.

That last one is the one nobody budgets. A chatbot trained on documents is only as current as the documents. Plan on a few hours a month for review at platform level, more if your pricing or stock changes often. If a supplier quotes you a build cost with no ongoing figure at all, ask directly what happens in month four when an answer goes stale.

Data protection is the other running cost that hides outside the quote. A chatbot handling names, email addresses and conversation content is processing personal data, and the ICO’s guidance on AI and data protection sets out what that obliges you to do, including when a data protection impact assessment is required. Some suppliers include this work. Many leave it with you, and it costs either money or your own staff’s time.

When a chatbot does not pay for itself

Plenty of businesses should not buy one, and a supplier who won’t say so is a supplier to avoid. The maths is simple enough to do on the back of an envelope: divide the monthly running cost by your loaded hourly staff cost, and that’s how many hours of work the assistant has to remove every month before it breaks even. Then ask whether your enquiry pattern can plausibly deliver those hours.

Four situations where the answer is usually no.

Low repeatable volume. If you handle a handful of enquiries a week and each one is different, there’s nothing for the bot to absorb. The staff time saved won’t clear the monthly fee, and you’ve added a system to maintain.

High-judgement conversations. Where the first enquiry is genuinely a discovery conversation (professional services, bespoke manufacturing, anything where the price depends on a survey), automating the front door can cost you the qualification signals your sales process runs on.

No written source material. If nothing is documented, a chatbot project becomes a documentation project with a software bill attached. Write the answers down first, publish them, and see how much of the problem that alone solves.

The real problem is the website. A large share of repeat enquiries are questions the site should already answer above the fold: opening hours, service areas, price indication, lead times. Fixing the page is cheaper than building a robot to read the page aloud.

“The honest scoping conversation starts with whether the enquiries repeat. If they don’t, a chatbot is an expensive way to handle a problem that isn’t there, and we’d rather tell a client that at the first meeting than at the invoice.” Ciaran Connolly, founder, ProfileTree.

How ProfileTree scopes a chatbot project

ProfileTree works through a fixed order on chatbot enquiries, and that order exists because the price genuinely cannot be known until the last step is done.

The first step is a source content audit: what is already written down, how current it is, and how much of it needs rewriting before a model can use it. The second is an integration list, with each connection marked read-only or read-write, because that distinction decides most of the build cost. The third is a volume estimate drawn from existing enquiry data, contact form submissions and support inbox counts, which sets the running cost band. Only then does a written scope with a fixed price follow.

ProfileTree is a Belfast-based digital agency working with SMEs across Northern Ireland, Ireland and the UK, and chatbot work runs alongside its wider AI implementation and digital training services. The broader approach is documented in implementing AI chatbots, which covers the build stages in more detail than a pricing page needs to.

Three ways chatbot work gets billed

Two suppliers can quote the same total and structure it completely differently, which changes what you own at the end and what you pay over three years. The billing model is a separate decision from the tier, and it deserves its own five minutes.

Fixed-price build plus a support fee. You pay a defined sum for the build, then a smaller monthly amount for hosting and maintenance. This is the clearest model to budget against and the easiest to compare across suppliers. The risk is change requests: anything outside the written scope becomes a variation, so the scope document matters more than the headline number.

Managed subscription with no upfront cost. The build is folded into a monthly fee, which suits cash flow and gets you live quickly. The trade-off is duration. A £200 monthly managed fee passes the total cost of a £5,000 fixed-price build at around month 25, and you typically don’t own the trained content or the configuration when you leave. That can still be the right call if you want someone else carrying the maintenance burden, but price it over 36 months, not over one.

Outcome-based or per-resolution billing. Some platforms charge for conversations the assistant handles without a human. It aligns cost with value in principle. In practice, get the definition of a resolved conversation in writing before you sign, because a bot that answers unhelpfully and gets abandoned can still be billed as a success.

None of the three is inherently better. Fixed price suits a defined project, subscription suits a business that wants no capital outlay, and outcome billing suits high, predictable volume.

What a chatbot quote should itemise

A quote that gives you one number and a paragraph of description isn’t a quote; it’s a starting position. Ask for these lines separately, and the comparison between suppliers becomes straightforward.

• Source content preparation. Stated as who writes the material that doesn’t exist yet, and how many documents or pages are in scope.
• Each integration, named. Listed individually and marked as lookup only or able to change records in the connected system, since the second kind carries the testing cost.
• Model API usage. Confirmed as included in the monthly fee or passed through on a separate invoice. Both are reasonable, but you need to know which.
• Hosting. Some quoted build prices exclude the server and vector database entirely.
• Data protection deliverables. Impact assessment, data processing agreements with any third party, and updates to your privacy notice, marked as included or excluded.
• Ownership and handover. Who holds the prompt, the trained knowledge base, and any code at the end of the contract, and what it costs to export it.
• Support terms. Response times when the assistant starts giving wrong answers, and whether corrections are included or billed.
• VAT. Most UK agency quotes are stated excluding it, which is a fifth on top of every figure above.

If a supplier declines to break the price down this way, that tells you something about how the project will be run.

Working out your own number before you call anyone

Take three figures into any supplier conversation: your monthly enquiry count, your list of integrations split into read-only and read-write, and an honest assessment of how much of your knowledge is actually written down. With those three, you can place yourself in one of the bands above within about £2,000, and you’ll know immediately whether a quote is priced for the job you described or for a different one.

If a supplier can’t quote from those three inputs, or insists on weeks of paid discovery before naming any range, that’s information too.

FAQs