You can buy an "AI chatbot" today for £20 a month. The marketing pages are gorgeous. The trial is easy. The pitch is irresistible: cutting-edge AI, no developer needed, live on your site by Friday.
And then six months later you've quietly turned it off, because it was either embarrassing your brand, frustrating your customers, or both. This pattern is so common that we've stopped being surprised by it.
Here is what's actually going on, and when custom is worth it instead.
What you actually get for £20 a month
You get a chat widget on your website. You get an LLM somewhere behind it. You get a small admin panel where you can paste in some FAQ text. The vendor calls this "training" the AI. It isn't. It's giving the AI a small fixed reference document.
What you don't get:
- Real understanding of your products, prices, and policies in their full nuance.
- Access to live data (orders, stock, appointments). It cannot tell a customer where their parcel is.
- Genuine ability to take actions (book a meeting, raise a ticket, process a return).
- A clean handover to a real human when the customer needs one.
- Any meaningful improvement over time without you constantly editing the FAQ document.
The result is a tool that handles the simplest 20% of your enquiries badly, and the rest not at all.
The hidden costs that show up later
The £20 a month is not the real number. The real number includes everything else that gets worse because of it.
Damaged brand
Customers tell each other when they have bad chatbot experiences. They share screenshots. They make jokes. A bad bot on your homepage is doing PR for your competitors.
Lost conversions
If your bot frustrates a customer enough, they leave. Some of those would have bought. You'll never know how many, which makes the cost easy to ignore and easy to underestimate.
Increased support load
Bots that can't answer often dump customers into a support queue, badly framed. Your team now has to deal with confused customers, often after the customer has already had a frustrating experience. Net work goes up, not down.
Time spent maintaining a tool that doesn't deliver
Most teams quietly spend hours a month editing the chatbot's FAQ document, trying to fix the latest gap. This is hidden work that nobody schedules and nobody bills.
A £20-a-month chatbot is not 100x cheaper than a custom build. It's 100x cheaper to start, and often more expensive to keep running once you count the things it makes worse.
Why custom is different
A custom-built AI customer support agent costs more to build (typically low-to-mid four figures for a focused first project). It costs slightly more to run. And it does fundamentally different things.
- It is trained on your real content, in depth: every product description, every policy, every awkward edge case your team is sick of explaining.
- It connects to your real systems: order data, inventory, calendars, ticketing.
- It can take actions, not just talk.
- It hands over to humans cleanly, with full context, when it should.
- It improves systematically, with weekly review of where it got stuck.
The result is a tool that handles 70-80% of enquiries well, escalates the rest properly, and earns its build cost back inside three to six months.
When the £20 chatbot is actually fine
To be fair, there are cases where off-the-shelf is genuinely the right answer.
- You have a tiny website with five FAQs and very low traffic.
- You want to capture an email address for follow-up, nothing more.
- You're testing whether anyone actually uses chat before committing to anything serious.
If that's you, by all means start with a SaaS chatbot. Just be honest with yourself about what it is. It is a contact form with a chat skin, not an AI customer support agent.
How to tell what you actually need
Three quick questions:
- Does the agent need to know things that change weekly (stock, prices, schedules)?
- Does it need to take actions on the customer's behalf?
- Will more than 30% of your enquiries need anything beyond a static FAQ answer?
If you said yes to any of those, off-the-shelf will not work for long. You'll either turn it off, or worse, leave it on and let it cost you customers.
Where to start instead
Don't start with "what AI tool should we buy". Start with "what enquiries are eating our time and money". Once that's clear, the right shape of solution becomes obvious. We wrote a longer guide to the customer support side specifically: AI customer support: what works for UK businesses.
If you'd like a hand thinking through whether custom or off-the-shelf is right for your situation, that's exactly the kind of conversation we have on the first call. Tell us what's happening on your support channels and we'll be honest about which way to go.