Dan·23 October 2026·6 min readYour AI Is a Chat Box. It Should Be a Colleague.
Almost half of UK small business owners now say they use AI tools. Simply Business put the figure at 47 per cent in its 2026 SME Insights Report, more than double the year before. That sounds like a revolution until you look at how it is being used. Another survey of micro and small businesses found only 21 per cent use AI regularly, and just 6 per cent have it embedded in daily work across the business.
That gap is where the value is. Most people open a chat window, type a question, copy the answer somewhere else and close the tab. Every conversation starts from zero. The AI knows nothing about the business, can touch none of its tools and remembers nothing from yesterday.
A chat box answers questions. A properly set up AI does work. The difference is not the model, it is the setup around it.
I run my own studio with AI wired into almost everything: email, the website, video production, analytics, client work. It took a lot of trial and error to get there. This is what actually made the difference.
The five things that turn a chat box into a colleague
Your tone of voice, your prices, your rules, the words you never use, who your customers are. Put it in a document the AI reads every time. I have a short file of standing rules (no dashes in anything I publish, who "email me" means, what never gets hosted without asking) and it saves me correcting the same thing fifty times.
Email, calendar, website, CRM, analytics. An AI that can read your actual data gives answers about your business, not about businesses in general. The key is narrow permissions: give it exactly what the job needs and nothing more.
The best setups keep notes between sessions: what was decided, what went wrong, what you prefer. When I correct something, it gets written down, so tomorrow's session starts where today's finished.
Nothing customer facing goes out without a human seeing the exact version first. The AI drafts the email, the post, the change. I say go. That one rule has saved me more than once.
Prices, documents, diagrams, anything that has to be identical every time should come from a program the AI writes, not from the AI improvising. I wrote about why in Let AI do the hard work, use determinism for consistency.
What the difference looks like
- Starts from zero every conversation
- Knows nothing about your business
- You copy and paste answers into other tools
- Different tone every time
- Nobody knows what it saved, if anything
- Reads your rules, tone and prices every time
- Works inside your email, website and data
- Drafts the finished thing for you to approve
- Remembers corrections and preferences
- Saves measurable hours on named jobs
Where it went wrong for me
I would be lying if I said this was smooth. Every one of these lessons cost me something first, and I have written most of them up honestly.
The pattern behind every one of those
None of these were the AI being stupid. They were the AI doing exactly what it was asked, with more access or less oversight than it should have had. The fix was never "use AI less". It was better rules, narrower permissions and a human checkpoint before anything leaves the building.
Measure what it saves
If you cannot say what your AI setup saves, you cannot tell whether it is working. Pick three jobs you do every week, time them before, and time them after. Writing quotes, answering the same customer questions, preparing social posts, chasing invoices. Real numbers on real jobs beat a vague feeling of being busier.
For me the biggest wins were not where I expected. It was not writing. It was the boring glue: pulling analytics into a report, turning one piece of content into five formats, checking a site after every change.
What I would tell you to do this week
Tone, prices, rules, customers. Make the AI read it every time.
Start with something low risk, like read only analytics or a draft only email account.
Nothing goes to a customer without you seeing the final version.
That number tells you whether to go further.
If you take one thing from this
The businesses getting real value from AI are not using a cleverer model than you. They have spent a few hours setting it up so it knows their business, can reach their tools and cannot do anything daft without asking first.
If you want help getting there, that is a big part of what we do in AI integration.