The intelligence is already there. The knowledge is lacking.

Most businesses that tell me AI did not work for them never had an AI problem. They had a process nobody had ever written down.

I sat down with Clint on the AI Unboxed podcast a few months back to talk about exactly this, and the episode is out now. It covers where small businesses are losing time, what AI genuinely should not be doing, and why being small is an advantage rather than a handicap. The full conversation is below, and the rest of this piece goes deeper on the part I think matters most.

Starting at the tool is starting in the wrong place

The most common pattern I see is a business that has bought a tool before it has agreed what problem the tool is solving. Someone demos a chatbot, a few seats get paid for, and six months later nobody can point at a single hour saved.

Clint called it the cart before the horse, which is about right.

It happens because of the order AI arrived in. It came to us personally first. People used it to tidy up a photo or answer a random question, got comfortable with it, and then walked into work thinking a company is just a bigger version of that. Businesses were slow off the mark and a lot of small ones are still standing at the start line, not sure what the first step actually is.

The real work is not the tool. It is understanding the business. What the processes are, what the pain points are, what actually takes the time. Do that first and the tool choice more or less makes itself.

Why small is the advantage, not the handicap

The businesses I work with sit roughly between three and twenty people. Above that you start getting IT departments, more decision makers, and a longer road to any change.

That band has something larger organisations cannot buy. A small business can decide on a Monday to change how it works and be running differently by Friday. A larger one has policies and processes that have to be ticked before anything moves, and the honest timescale is months or years rather than days.

I spent years running IT departments, so I understand exactly why that friction exists and I would not remove it. When you are head of IT your job is security, compliance, and keeping the systems up. Anything that interferes with those three is a problem, and AI arrives carrying real questions about where data goes and who controls it. Being cautious there is the correct instinct.

There is a second constraint too. Larger organisations are usually already committed to a stack, typically Microsoft and Copilot, so their choices are narrowed before the conversation starts. A five person business has no such commitment. It can pick what actually fits.

So the advantage is genuine. The question is what you spend it on.

Where the time is actually being wasted

Here is the honest answer to "where are companies wasting time with AI".

They are wasting it on single answers. One person asks a question, gets a decent reply, and moves on. Someone else in the same business asks a similar question in a slightly different way and gets a different reply. Nothing is integrated. Nothing accumulates. Every interaction starts from zero.

At an individual level that is still mildly useful. Someone drafting an email faster is a small win. At an organisational level it is close to worthless, because none of it compounds.

What is missing is a layer underneath. A knowledge base that sits between the business and the model, holding:

• Your processes, written down properly. How a job actually moves through the business, not how the org chart says it does.
• Your policies. What is allowed, what needs sign off, what never goes out without a human reading it.
• Your voice and your brand. How you write, what you sound like, what you would never say.
• Access to your systems, so it is working from live facts rather than whatever it was told in the prompt.

Get that in place and the model stops guessing and starts making the decisions you would make. The models are already clever enough. That was never the constraint. The intelligence is already there. The knowledge is lacking. Supply the knowledge and the same AI is worth about ten times what it was the day before.

The uncomfortable part is not technical

People assume the hard bit is the technology. It usually is not.

The hard bit is that a lot of small businesses have never sat down and asked what they actually do. Not for years, sometimes not ever. What are all these tasks? Why do we do it that way? Where does it hurt? Which of these steps exists only because someone left in 2019 and nobody revisited it?

That exercise is unglamorous and slightly exposing, and it is the whole game. Once that information is written down you have something you can hand to an AI directly. Skip it and you are back to buying tools and hoping.

Can a business do this itself? If you are not especially technical but you do not mind getting your hands dirty, there is a real chance of it, because most of the work is not technical at all. It is interviewing your own people and being honest about the answers.

Ask what eighty percent looks like

The mindset shift that does the most work is a small one. For every task you do regularly, the question is not "could AI do this". It is "could AI do eighty percent of this".

Almost nothing goes to a hundred, and chasing a hundred is how projects stall. Eighty is available on a lot more tasks than people expect, and eighty on a job you do every day is a significant amount of your week back.

Two rules sit alongside it, and both are about where the human stays.

First, AI should not have the final say. It is very good at giving you ten ideas quickly and drafting the first version of almost anything. It should not be making the call. Anything that goes out with your name on it gets a human tick first, because that is your brand and your reputation, whatever the courts eventually decide about liability.

Second, this is a business owner decision, but it cannot be a business owner secret. If someone turns up talking about automating eighty percent of a job, the person doing that job has every reason to be worried. So we interview the team, find out what they genuinely do and where it hurts, and aim the work at the mundane parts. In a decent company with decent people the result is that four people do the work of eight rather than four becoming three. Most small businesses want to grow, which means the capacity has somewhere to go.

What you can take from this

If you want a starting point that costs nothing:

1. Pick one process that annoys everybody. Write down every step, including the daft ones.
2. Note what each step actually needs to know to happen. That is the beginning of your knowledge layer.
3. Ask which eighty percent of it could run without a person, and mark where the human tick has to stay.
4. Only then go looking at tools.

One last thing worth knowing before you build anything. Whatever you put in has to be effortless to use, because people will not change how they work unless it is completely frictionless. A better process that takes one extra click will quietly be abandoned inside a fortnight. That constraint should shape the design from the first day, not get discovered at the end.

Being small is the advantage. Spend it on understanding your own business, and the AI part gets a lot easier than it currently looks.