Your AI Chatbot Doesn't Need a £200/Month SaaS Wrapper

That £200 Monthly Bill for a Chat Widget

You're scrolling through Intercom's pricing page, or perhaps you've already taken the plunge, and there it is: £199 per month for "Fin," their AI chatbot. Or maybe you're looking at Tidio at £89/month, or Drift pushing £300 for their "conversation intelligence." And you're thinking, surely this thing must be doing something remarkable to justify that price.

Here's the uncomfortable truth: it's probably not. The AI chatbot market has developed a convenient pricing strategy, and it has very little to do with the technology actually being any good at answering your customers' questions.

What's Actually Happening Under the Hood

Let's pull back the curtain. When a customer asks your chatbot "Where do I find my invoice?" or "Can I change my shipping address?", here's what happens:

First, the system grabs whatever context it can, usually your help centre articles or a few past conversations. Second, it packages that context alongside the customer's question into a prompt, which is essentially a long, carefully structured text instruction. Third, it sends this to an LLM (large language model) like GPT-4, Claude, or MiniMax. Fourth, the LLM generates a response, and fifth, this response gets sent back to your customer.

This is a remarkable piece of engineering, yes. But here's what the SaaS platforms are hoping you won't notice: the actual AI processing costs them roughly £0.002 to £0.01 per conversation. That's less than a penny.

You're not paying for AI magic. You're paying for a monthly subscription to avoid using a form, some buttons, and an API key.

The rest of that £199? That's the user interface, the onboarding flow, the "drag-and-drop" builder, the analytics dashboard, and the sales team's commission.

The Simple Architecture That Costs Pennies

You can build exactly the same thing yourself. The architecture looks like this:

• A knowledge base: This is just a collection of your FAQs, product info, and policies. You already have this.
• A context document: A plain text file that tells the AI who it is, how it should respond, and what it can and cannot do.
• An LLM API: The actual brain. Providers like OpenAI, Anthropic, MiniMax, or dozens of others charge only for what you use.
• A chat interface: The bit your customers see. This can be a simple widget, or even just a form that submits to the AI and displays the response.

The Real Numbers: API Costs vs SaaS Subscriptions

Let's look at what you're actually paying for with the leading AI chatbot platforms, and what it would cost to build the same functionality yourself.

The MiniMax M2.5 model is particularly worth noting here. It's a frontier-quality model that costs roughly $0.30 per million input tokens and $1.20 per million output tokens. For context, a typical customer service exchange might use 500-1,000 tokens total.

This chart makes the difference stark. You're paying £199-£300/month for something that costs between £5-£30 to run yourself. That's a markup of 1,000% to 6,000%.

For a small business handling 500-1,000 customer conversations per month, the math is straightforward: even with generous buffer for testing and edge cases, you'd struggle to spend more than £20-£40 on API calls. The rest is pure margin for the SaaS platform.

What About the Overhead?

But wait, there's development time to consider. You're thinking, "I'm not a developer, so I need to pay for the convenience." That's fair, and it's worth putting a realistic number on it.

A basic FAQ chatbot, properly built with good context documents and safety guardrails, takes a competent developer 4-8 hours to create. At £50-£100 per hour for a freelance developer, that's £200-£800 one time. Compare that to paying £200/month for three years: that's £7,200. Even accounting for maintenance and updates, you'd be looking at a one-year payback period, at worst.

Many agencies, including us, can build this for you at a fraction of what you'll pay in annual SaaS subscriptions. And unlike the SaaS platforms, you're not locked in.

Addressing the Legitimate Concerns

Before you go tearing up your Intercom contract, let's address the reasonable objections. You might be thinking: "What about hallucinations? What if the AI makes things up?"

This is a valid concern, and it's where the "context document" part of the architecture becomes critical. A well-designed prompt explicitly tells the AI what it can and cannot answer, instructs it to say "I'm not sure, let me connect you with a human" when it's uncertain, and grounds every response in your actual documentation.

Does this guarantee perfection? No. But neither does paying £199/month to Intercom, whose AI still hallucinates regularly enough that they've built an entire "human takeover" feature to handle when things go wrong.

The other concern is tone. You don't want a chatbot that sounds like a robot, or worse, inappropriately casual. This is again a matter of prompt engineering. You can literally write: "You are a friendly but professional customer service agent for a UK-based business. Use British English spelling. Never use slang. Always be polite but efficient."

The AI will follow these instructions. It's genuinely that flexible.

What a Sensible, Affordable AI Assistant Looks Like

Here's what you actually need for most small business use cases:

1. An FAQ bot that answers the 10-20 questions you repeat daily. This is the lowest-hanging fruit and the biggest time saver.
2. A lead qualifier. Ask the visitor a few questions, qualify whether they're a good fit, and book a call or send an email to your team.
3. A triage assistant. Gather context before handing to a human. "What's your order number? What are you trying to achieve?" This alone can cut support ticket handling time by 30-50%.

None of these require a £200/month subscription. None of them need the full suite of features that Intercom or Drift provides. What they need is good context, a decent LLM, and a clean interface.

You deserve to understand what you're paying for. The AI chatbot market has done an excellent job of making their products seem essential and complex. The reality is far more mundane: they're charging premium prices for a commodity technology that costs almost nothing to run.

You don't need permission to build this yourself. You don't need a developer with AI expertise. You need someone who can write a good context document, connect an API, and put a chat box on your website.

If you're currently paying £100, £200, £300 a month for a chatbot that's supposed to be saving you money, take a closer look at what it's actually doing. Chances are, it's not doing anything you couldn't build for the cost of a nice lunch each month.