RAG in plain English: giving AI a brain that knows your prices, diary and rules

You have probably seen an AI chatbot confidently tell a customer the wrong price, quote a service you stopped offering two years ago, or invent an opening time. That is not a bug in the AI. That is the AI doing exactly what it was built to do - answering from general knowledge - when what you needed was for it to answer from YOUR information. RAG is the fix for that.

What RAG actually is

RAG stands for Retrieval-Augmented Generation. Ignore the acronym. The idea is simple: before the AI writes its answer, it looks something up. You give it a set of documents - your price list, your booking rules, your FAQ, your returns policy - and when a question comes in, the system finds the relevant bit and hands it to the AI along with the question. The AI then answers from that, not from whatever it half-remembers about your industry.

Think of it like this. You hire a new member of staff. They are bright, they speak well, they can handle a conversation. But they dont know your rates, your lead times, or that you no longer cover the M25 corridor on Fridays. You hand them a folder. Now they do. RAG is the folder.

Why a standard AI gets your business wrong

A large language model - the engine behind ChatGPT and most AI tools - was trained on text from the internet up to a certain date. It knows a lot about the world in general. It knows nothing specific about your business unless you tell it. It will fill the gap by guessing, and it will sound confident while it does it.

The result: a customer asks your AI assistant what the call-out charge is. The AI has seen a hundred plumber websites and knows the average is somewhere in a range. It picks a number. That number is wrong. The customer turns up expecting one thing; you quote another. The job dies or the argument starts.

Thats not a trust problem with AI in general. Its a plumbing problem: the wrong information was connected to the customer-facing tool.

What you can put in the folder

Anything that is true about your business and changes how a question should be answered. Common ones:

• Your current price list or rate card
• Your service area, coverage rules, and any exceptions
• Booking availability rules (“we need 48 hours notice for a new patient appointment”)
• Your terms, guarantees, and what is and is not included
• Product specs, compatibility notes, care instructions
• Answers to the twenty questions you get asked every week

When any of that changes - a price goes up, a service is dropped, a new rule comes in - you update the document. The AI picks up the change next time it looks. No retraining, no developer call, no waiting.

The three options if you want AI to answer accurately

Here is the honest picture of what you can do.

Option one: prompt engineering alone. You paste your prices and rules directly into the AI’s instructions, sometimes called the system prompt. It works for small, stable sets of information - a handful of services, a short FAQ. It breaks down when the information is long, changes often, or needs to be searched rather than read top to bottom. The AI can also “forget” instructions buried deep in a long prompt. Free to try; limited ceiling.

Option two: an off-the-shelf tool with a knowledge base. Several chatbot platforms - Intercom, Tidio, Crisp and others - let you upload documents or point at a URL and they handle the retrieval themselves. Subscriptions vary; check each platform’s published pricing. This is the right answer for a lot of businesses. You get accurate answers without a custom build. The trade-off is you work within their interface, their limits on document size, and their update cycle. If your needs are standard, start here.

Option three: a purpose-built RAG pipeline. A custom build where documents are processed into a searchable index, a vector store - a database that finds meaning rather than just matching words - retrieves the right chunks at query time, and the AI answers from those chunks. More setup, more control. The right fit when the information is complex, the volume of queries is high, the answers need to trigger actions (book a slot, update a record), or the off-the-shelf tools keep getting it wrong.

The honest recommendation: if you have fewer than a dozen services and your information changes rarely, try option two first. It will cost you an afternoon and a subscription, not a build. If you are running something with real complexity - lots of SKUs, dynamic availability, rules that interact with each other - option three earns its cost back in staff time and customer confidence.

The question worth asking yourself now

How many times last week did a customer get a wrong answer - from your website, your AI tool, a staff member working from an old price list - and either pushed back, dropped off, or cost you time to correct? Multiply that by the value of a job. That is what the information gap costs you. The folder is not a technology project. It is a revenue decision.

If you want to work out whether a RAG build makes sense for your setup - what information you have, where the gaps are, whether an off-the-shelf tool covers it - we can look at that in a single conversation. No commitment, just a clear answer either way. See what we build and then come and ask.