If you spend any time reading about AI, you will keep bumping into three words used as if they are rival products you must choose between: prompting, a knowledge base, and fine-tuning. Vendors have opinions, usually shaped by what they happen to sell. So here is the honest version for a small business owner, free of any sales angle, because getting this choice right saves you a lot of money and effort, and getting it wrong wastes both.
The short version, which most of this guide just explains and justifies: for almost every small business, the answer is prompting and a knowledge base. You very probably do not need fine-tuning. Let us go through why.
Three different jobs, not three brands of the same thing
The mistake is treating these as three flavours of one decision. They solve different problems, and the trick is knowing which problem you actually have.
Prompting is how you brief the AI. It is the instructions and information you give it in the moment: who it is helping, what you want, and the facts it needs, pasted right into the chat. Brief it well and a general AI does a great deal of your work with nothing else added. Most owners underuse prompting and reach for something complicated before they have got good at the free, instant thing.
A knowledge base is prompting’s facts made permanent and searchable. Instead of pasting your prices and policies every single time, you store them once somewhere the AI can look them up when it needs them. This is the “giving your AI a brain” idea. It fixes the core problem, which is that the AI does not know your business, and it fixes it by making sure your real information is there to be read. The mechanics of how it looks things up are covered plainly in our guide on RAG without the jargon.
Fine-tuning is different in kind. It means taking a model and further training it on a pile of your own examples so its default behaviour shifts. Crucially, fine-tuning teaches a model a style or a skill, a way of responding, not a set of fresh facts. That distinction is the whole thing, and it is where most of the confusion lives.
Why “just fine-tune it on our data” is usually the wrong instinct
The phrase sounds right. Train the AI on our business, and it will know our business. It is a reasonable thing to assume and it is mostly wrong, for a few honest reasons.
First, fine-tuning is poor at facts. It nudges how a model tends to respond, but it is not a reliable way to make it memorise your exact prices or policies, and it certainly will not quote them back with a source you can check. If your real problem is “the AI does not know our specifics”, a knowledge base solves that directly and fine-tuning solves it badly.
Second, it does not update cleanly. Change a price and you cannot just tweak a fine-tuned model, you are looking at retraining. Compare that with a knowledge base, where you edit one document and every future answer is current. For a business whose prices, lead times and services actually change, that difference is decisive.
Third, it costs real money and expertise. Fine-tuning needs a good quantity of quality examples, someone who knows how to do it, and money each time you redo it. That is a serious commitment for a payoff that, for most owners, a well-written knowledge base already delivers more cheaply and more flexibly.
None of this means fine-tuning is useless. It earns its place when you have a narrow, stable need for a particular style or format at high volume, and you have already got everything you can out of prompting and a knowledge base. That is a real but uncommon situation, and it is a long way from most people’s first move.
The order to actually work through
Because these solve different problems, the sensible approach is not to pick one, it is to work through them in order and stop when your problem is solved. Most people stop well before the end.
-
Prompt well first. Get good at briefing the AI clearly and pasting in the facts it needs. Free, instant, and it fixes more than you would expect. If you have not felt the limits of this, you are not ready to spend on anything else. Our guide on building a second brain this week is really prompting with your facts written down.
-
Add a knowledge base when there is too much to paste. The signal is a specific pain: your useful information is now too big to hand-feed every time, or other people need it, or it changes often, or the answer has to be exactly right. That is when you make your context permanent and searchable. This is where the great majority of businesses should live.
-
Only then consider fine-tuning, and only for a narrow need. If, after prompting well and building a solid knowledge base, you still have a specific, repeating, style-shaped need at real volume, look at fine-tuning for that one thing. If you cannot name that need in a sentence, you do not have it yet.
Follow that order and you spend effort where it pays and skip the expensive step almost everyone can skip.
What we actually do
For what it is worth, we build real tools on top of a persistent context system of our own, and we get the results we do from exactly this ladder: strong prompting over a well-kept knowledge base. The honest account of running that, including why maintaining the facts is the real job, is in how we run our own AI brain. Fine-tuning is not what makes the difference for the work small businesses need doing. Knowing your business, kept current, is.
Where to take this next
Our flagship course Build an AI-ready business: give your AI a brain walks the whole ladder in plain English, so you can see exactly where your business sits and what to do next.
To get the terms straight first, the Plain-English AI Glossary on our free resources shelf defines fine-tuning, knowledge base and the rest on one printable page.
And if you want to talk it through with people who build these systems for a living, come to our AI Automation Masterclass in Manchester. Tickets are normally £20. This one’s free, a limited-time offer to launch the series.