AI Chatbots for Business: Are They Worth It? An Honest UK Answer

It is ten past eleven on a Sunday night and someone is on your website asking whether you deliver to Truro before Friday. Nobody answers. By Monday morning they have bought from the firm down the road, whose little chat bubble told them yes in four seconds.

That is the scene every chatbot vendor wants you to picture, and it is real. What they leave out is the other scene, where the bot cheerfully promises a refund your policy does not allow and you are the one who has to honour it. I am Armen Andonian, and I help UK owners automate the work that eats their week. Most of the pages ranking for this question are written by companies selling chatbots, so here is the version from someone who does not care which one you buy.

Is an AI chatbot for business worth it?

An AI chatbot is worth it for a business that receives a steady flow of repetitive questions whose answers already exist in writing. Think delivery times, opening hours, returns, booking changes, “do you do X”. If that is a big slice of your inbox, a chatbot can answer it in seconds at any hour and give your team real time back.

It is not worth it if each enquiry is a judgement call, or if the first conversation is where you win the sale. A solicitor or a high end kitchen fitter usually loses more from a robotic first impression than they save in admin.

What does an AI chatbot actually do well, and what does it do badly?

A modern AI chatbot is good at reading your own written material and answering questions from it in natural language. Give it your returns policy and your FAQ page, and it will handle “can I send back a jumper I wore once” far better than the old button menus ever did.

It is bad at anything you have not written down. Ask it about a price you never published and it may simply invent one, politely and with total confidence.

It also struggles with emotion. An angry customer whose order arrived broken does not want a paragraph about the returns window, they want a person to say sorry. The best setups treat the bot as a fast front desk that knows when to fetch someone.

How do you work out whether a chatbot will pay for itself?

Count before you buy. Export a fortnight of emails, chat messages and contact form submissions, then mark each one as either “answer is already written somewhere” or “needed a human”. That single exercise tells you more than any vendor demo.

Here is an illustration with round numbers, not a promise. Say you get 60 repetitive questions a week and each takes someone four minutes to find and answer. That is four hours a week. If the bot genuinely resolves half of them, you have saved two hours, and on Intercom’s pricing those 30 outcomes would cost about $30 plus a $29 seat. Worth it if your time is worth more than roughly £25 an hour. Marginal if the bot resolves fewer than you hoped.

Be careful with resolution rates. Intercom publishes an average of 76% for its Fin agent, which is its own figure across customers with well kept help centres. Tidio guarantees 50% on its Premium plan. Plan on the lower number until your own data says otherwise.

If counting by hand sounds tedious, my free AI Opportunity Scorecard takes two minutes and estimates the hours repetitive work is costing you, which is a sensible first check before you pay for any tool.

What does an AI chatbot cost a UK business?

An AI chatbot costs anything from pennies per conversation to several hundred pounds a month, and the pricing model matters more than the headline figure. Most platforms now charge a monthly fee plus a charge for each conversation the AI resolves, so a busy month means a bigger bill.

Watch the small print on what counts as a billable success. In March 2026 Intercom widened its “outcome” to include some handovers to a human, so read the definition before you model the cost. Zendesk includes only a handful of automated resolutions per agent each month before the extras kick in.

The hidden cost is upkeep. Someone has to read the transcripts, spot the questions the bot fumbled and update the source document. Budget an hour a week for the first couple of months.

What happens when a chatbot gets it wrong?

When a chatbot gets it wrong, the business pays, not the software. Three public cases are worth knowing before you switch anything on.

In 2024, Air Canada’s chatbot told a grieving customer they could claim a bereavement discount after flying. That was not the policy. Air Canada argued in front of a Canadian tribunal that the chatbot was effectively responsible for its own words. The tribunal disagreed and ordered the airline to pay up.

Closer to home, DPD switched off part of its UK chatbot in January 2024 after a customer got it to swear and write a poem about how useless the company was. And a Chevrolet dealer in California watched its bot “agree” to sell a new Tahoe for one dollar after a user simply told it to agree with everything.

None of these needed a hacker. They needed a bot with no boundaries. So give the bot hard boundaries before launch. It answers only from the documents you gave it, and when it does not know, it says “I will pass you to the team” rather than guessing. Anything involving money, such as a discount or a refund, goes to a person every time.

Do you have to tell customers they are talking to a bot?

Yes, in practice you do. The ICO expects you to explain in plain language when AI is processing someone’s personal data, and the EU AI Act’s transparency rules require chatbots to identify themselves as AI, which matters if you sell to customers in Europe.

There is also a commercial reason. Gartner surveyed 5,728 customers in 2024 and found 64% would prefer companies did not use AI in customer service at all. Their top worry was not bad answers. It was being unable to reach a person.

So put a line in the first message along the lines of “I am an AI assistant, type ‘person’ at any time to reach the team”, and make sure that actually works. Then sign the vendor’s data processing agreement and check where the conversations are stored. If the answer is the US, ask what transfer safeguards they rely on. And keep the bot well away from card numbers and health details.

Can you build your own AI chatbot?

Yes, and for many small businesses it is the smartest way to find out whether a chatbot is worth it at all. You have two realistic routes.

The fastest is Chatbase. Point it at your website and FAQ pages, test it against the real questions from your fortnight of enquiries, then paste the widget onto your site. That is an afternoon of work and $40 a month if you keep it.

The route I teach is building it with Claude Code. You describe in plain English what you want: a small chat window on your site that answers only from your FAQ document, refuses anything off topic and emails you the conversation when a customer asks for a person. Claude Code writes the code while you check each step. Anthropic’s Haiku 4.5 model charges $1 per million tokens of input and $5 per million of output, and Anthropic’s own worked example puts around 10,000 support conversations at roughly $37. For most small businesses, that is pocket change. For a non technical owner, expect a weekend for the first version.

The real advantage is not the price. It is that you own the thing. You are not paying a per resolution fee that grows with your success, and the skill stays in your business for the next automation too. If you would rather build that first version with someone beside you, that is what my AI Opportunity Audit is for: a fixed £500, 60 minute session where we find the highest impact place in your processes for AI and build a real automation live in Claude Code.

How would I decide, in order?

1. Count a fortnight of enquiries and mark which ones are already answered in writing somewhere.
2. If fewer than a third are repetitive, stop. Fix your FAQ page instead and revisit in six months.
3. Write or tidy the single document the bot will answer from: prices, delivery, returns, opening hours, what you do not do.
4. Trial a cheap bot on that document for a month, with a clear “talk to a person” option.
5. Read every transcript in the first fortnight and fix the document wherever the bot stumbled.
6. Only then compare the real resolution rate against what you are paying, and decide whether it has earned its place.

Step three is where most of the value hides. Plenty of owners discover that writing the document fixes half their enquiries before any bot exists, because the answers finally live on the website.

Where I would start

If I were running a small UK business tomorrow, I would not sign up for any chatbot this month. I would spend an evening counting my enquiries and another writing the answers down properly, because that document is the asset. The bot is just a way of serving it.

Then I would test the cheapest version against real questions and let the transcripts tell me whether it is worth paying more. If you want to talk through whether a chatbot is the right first automation for your business or whether something else would save more hours, you can get in touch and I will give you a straight answer.