What a business automation consultant should tell you not to automate

A business automation consultant looks at how work actually moves through your business, works out which repetitive jobs are worth handing to software, and builds or configures the thing that does them. In a small business, the useful ones spend a good part of their time telling you what to leave alone.

That second half is the bit nobody advertises, so this is mostly about that.

The job I told a client to stop

Two sessions in with a landscape construction firm, I told them to stop trying to automate the thing they had hired me to look at.

They take measurements off design drawings. A PDF, a scale, and someone who knows what they are looking at. The team can do it fairly quickly between them, and the numbers have to be exact or the quote is wrong.

We tried it. The AI threw out so many queries that it was slower than doing it by hand.

So we left it alone.

That is a slightly awkward conversation to have when somebody is paying for your time. It is also the most useful thing that happened in the two days.

What a business automation consultant actually does

Strip out the language and it is four things.

They work out where the time goes. Not where you think it goes. Where it actually goes, which usually surprises people, because the jobs that eat a week are rarely the ones anybody complains about.

They decide what is worth handing over. Some of it is worth it. A fair amount is not, either because the job is not repetitive enough or because getting it right matters more than getting it fast.

They pick the tools, and for most small businesses that means choosing between things you already pay for rather than buying something new.

Then they build it, and they show somebody how to run it after they have gone. That last part is the one people skip, and it is why a lot of automations quietly stop working about four months in.

Where the time actually goes

The Department for Science, Innovation and Technology asked UK businesses what was holding them back on AI. Seventy one per cent said they had not identified a business need. Sixty per cent cited limited skills. Cost came further down the list.

So the common problem is not that this is expensive or frightening. It is that nobody has looked properly.

The looking is not complicated. You take a week and you write down what happens in it. Not the big projects. The recurring, unglamorous jobs. The invoice run. The quote pack. The report somebody builds every Monday by clicking through a system and writing numbers down.

I ask people to put two figures next to each one. Roughly how long, and roughly how often. Twenty minutes weekly is seventeen hours a year. Half a day monthly is six days.

Once those numbers are on a page, the conversation changes. People stop asking what AI can do and start pointing at things.

What to automate first

While we were failing at the takeoff, something else turned up.

The firm's cost catalogue got updated ad hoc. Someone would notice a price had moved, check a couple of recent invoices, change the line, and get on with their day. Every few weeks. When there was a gap.

Nobody had ever decided that was the system. It was what happened when a busy business had no time to do it properly.

Halfway through the second session the quantity surveyor worked out the fix herself, while we were talking about something else. All the supplier invoices already land in one place. Feed the last twenty in, cross check them against the current catalogue, flag anything that has moved. Run it once a month.

I had not suggested it. She found it, because she is the one who does the job.

That is the shape of a good first project. Boring, small, finished inside a fortnight, and it stops the firm quoting off prices that are four months old.

Not the most valuable thing you could do. The most achievable one. Confidence is the actual constraint in a small business, and confidence comes from a finished thing rather than a plan.

What not to automate, and why nobody says it

Three sorts of job are usually worth leaving alone.

Anything where the input is messy and the answer has to be exact. The takeoff failed on both counts. Measurements from a drawing have to be right, and the drawings vary.

Anything that happens rarely. If a job comes round twice a year, the time spent building and maintaining the automation exceeds the time it saves, and you have added a thing that can break.

And anything where the judgement is the job. On the same project, two supplier quotes listed what looked like different plants. African lily on one, agapanthus on the other. They are the same plant. I am a keen gardener and I did not know that. The quantity surveyor did.

No amount of prompting gets you that. It is domain knowledge, and it belongs to the person who has spent fifteen years in the trade.

A consultant who says yes to everything is either not looking properly or is being paid by the project. It costs me money to talk somebody out of work. It is also the reason the same people ring me back.

Which tools, and whether you need anything built

Most small businesses do not need anything built.

The British Chambers of Commerce, with the University of Essex, surveyed 668 UK firms, most of them small. Among those using generic AI tools as bought, three per cent reported a headcount decrease. Among those using bespoke AI, twenty one per cent did.

Their own authors say the data is cross sectional and should not be read as causal, and the bespoke group was small at around sixty five firms, so treat it as a signal rather than proof.

The point is that building is a different kind of decision. It costs more, it changes more, and you own it afterwards. Somebody has to fix it when the thing underneath it changes.

For most firms of one to fifty people, the answer is the tools you already pay for, used properly by somebody who has been shown how. Microsoft 365 includes Power Automate and hardly anybody opens it. Claude or ChatGPT on a paid plan, plus two days of training, will move more than a build will.

Bespoke comes later, when you have something specific enough that nothing off the shelf covers it, and you can describe it in one sentence.

When to hire somebody, and when not to bother

Hire somebody when you have identified the jobs and cannot get them done, or when you have tried and it half worked and nobody trusts it.

Do it yourself when the job is small and you have somebody in the business who enjoys this sort of thing. That person exists in a surprising number of firms and is usually doing something else.

Do neither, for now, if you are in the middle of moving premises, changing your accounting system, or you have just lost two people. This work needs a bit of slack to land in. A business with none will start, get interrupted, and conclude it does not work.

What it costs in the UK

Every guide you will find on this is American and quotes dollars, which is not much use in Trafford.

Here is the honest UK version. Independent consultants doing this work sit somewhere between £250 and £500 a day, and I am at that end of the market rather than the agency end. A diagnostic conversation should be free. Nobody should be charging you to find out whether they can help.

A first project, properly scoped, is usually a few days rather than a few weeks. Somewhere between one and four thousand pounds gets most small firms a working automation and the training to run it.

If the number you are quoted has a comma in it and the work has not been broken into pieces you can buy one at a time, be careful. S&P Global research, reported by CIO Dive, found forty two per cent of organisations abandoned most of their AI initiatives during 2025.

That survey covered enterprise IT, not small businesses, so read it as a warning about how big projects fail rather than a description of your situation. The shape still holds. Nobody abandons the thing that took a fortnight and worked.

Be equally careful of the time savings you get quoted. Research by Opinium for OpenAI found UK small businesses using AI report saving 5.2 hours a week. That was funded by an AI company and it is self reported.

The Federal Reserve Bank of St Louis, measuring rather than asking, put it nearer 2.2 per cent of working hours, which is about an hour. I would plan on the smaller number. It is still fifty hours a year, and it is more likely to be true.

How to choose one

Three questions, and you can ask all of them on a first call.

What is the smallest version of this that would tell us whether it works. If there is no small version, it is too big to start with.

What would you recommend we do not automate. Anyone who says nothing has not looked.

What happens to this when you are not here. A workflow only one person understands has an expiry date, and if that person is the consultant, that is worse rather than better.

The thing that catches people out

The ONS found that the average UK business using AI runs 1.6 AI technologies, and only one in ten describes its use as extensive. The same department found that inside firms which have adopted AI, only around thirty per cent of staff actually use it.

So the licence gets bought and most people carry on as they were.

That is not a software problem and no consultant fixes it by building something better. It gets fixed by the people whose job changes being in the room when it is decided what should change.

What you could do on Monday

Take the last month of supplier invoices, or purchase orders, or whatever lands regularly with numbers on it. Put twenty of them alongside whatever price list or catalogue you quote from.

Ask what has changed.

That is a useful hour and it costs nothing. If it turns up three prices that moved without you noticing, you have found your first automation and you found it yourself, which is the best way round.

The honest limit

None of this fixes a process nobody has written down.

If your catalogue is out of date because only one person knows how it is meant to work, automating the update moves the problem rather than solving it. You will have an accurate list and still be one holiday away from nobody knowing what to do with it.

That is a documentation job before it is an automation job, and it is worth knowing before you spend anything.

The takeoff, by the way, might be fine in a year. I have been told about four things now that were going to change the industry, and two of them were fax machines.