Do you need help & advice with AI & Automation?
Two questions, and most businesses only ask the second one. What should we automate, and how do we automate it.
Which jobs should we automate first?
The ones that take a lot of time and add very little.
Not the most important work. Not the most interesting. The jobs that quietly eat somebody’s week and would not be missed if they happened by themselves.
The quickest way to find them is to go department by department and ask what everybody does that they would hand over tomorrow.
| Where | The kind of job worth looking at |
|---|---|
| Estimating and bids | Pulling standard sections together, chasing subcontractor prices, collating tender documents |
| Site | Turning site reports into something readable, summarising meetings, writing up variations |
| Accounts | Processing invoices, reconciling, chasing applications and payments |
| Admin and compliance | Standard paperwork, prequalification questionnaires, the same forms every month |
| Client and supplier email | The replies that say the same thing every time |
You are looking for repetition, not importance. If a job is repetitive and low value, that is the front of the queue.
One exception, and it matters. Do not automate a job that irritates everybody. If a process annoys people, it is usually because it is a bad process. Automating it gives you the same bad process, faster and harder to change. Fix it or stop doing it.
How do we actually automate it?
Build the process by hand first. The worst thing you can do is automate from the outset.
That sounds like a delay. It saves months.
Run the job manually, deliberately, and write down what you are doing as you go. Who does what, in what order, and what has to be recorded at each point. If you are ringing round subcontractors, that means noting when you will speak to them again and where you got to. Five or six steps, written down.
Do that and three things fall out of it:
| What you get | Why it matters later |
|---|---|
| The steps | You cannot automate a process you cannot describe |
| The language | Everyone calls the same thing the same name, which is most of the work |
| The handoffs | You find out where it actually gets stuck, which is almost never where people assume |
The reason this matters is simple. Nobody has done the work of asking what it looks like when a person does the job properly. So the automation copies whatever was happening before, including the parts that were not working.
What about buying a big tool instead?
A big tool will expect you to work its way, and you need to know your own way to judge whether that is a fair trade.
Sometimes it is. Established software has usually learned something about how the job should be done, and adopting its way is a shortcut. Sometimes it is not, and you end up reshaping a process that worked fine in order to fit software you bought in a hurry.
You can only tell the difference if you mapped your own process first. Otherwise you are comparing a sales demo to a vague memory.
Expect some rework either way. That is normal and it is not a sign you chose wrong.
[VIDEO_PLACEHOLDER]
So, what now?
Pick one job this week. Repetitive, low value, and irritating to whoever does it. Then instead of looking for software, spend an hour writing down how it actually happens now.
That hour is the whole difference between automation that sticks and automation that gets quietly abandoned.
If you would rather have someone work through this with you, that is part of what an AI Manager does. If you are earlier than that, start with getting started with AI in your business.
Frequently asked questions
What should a small business automate first?
A job that is repetitive, takes real time, and does not need judgment.
Standard document sections, chasing, routine replies, and turning notes into something readable are the usual first wins. Leave anything involving a price, a contract term or a person’s name until you have the basics working.
Should I map the process before automating?
Yes, and by hand. It is the step everybody skips and it is the one that decides whether the automation survives.
You need the steps, the language and the handoffs written down. An hour of that beats a week of rebuilding something that automated the wrong version of the job.
Is it worth automating something people find annoying?
Usually not. If a process annoys everyone, the process is the problem.
Automating it gives you the same bad process running faster and harder to change. Work out why it annoys people first. Quite often the answer is to stop doing it.
Do I need AI agents to automate a process?
No, and for most first projects you should not use one.
Plenty of useful automation is ordinary rules-based work that does the same thing every time. Agents make their own decisions, which is powerful and needs somebody accountable for what they decide. Get one simple thing working first.
How long should a first automation take?
Weeks, not months, and if it is heading past that the job was too big.
Pick something small enough that you can tell whether it worked. A first project that takes two quarters teaches you nothing except that automation is hard.
Last reviewed: 17 August 2026.
Related
- Every Microsoft Copilot explained
Seventeen products share the name. Which one you already have.
