Ask an owner where AI could help and they will point at whatever annoyed them that morning. Ask their bookkeeper and you get a different answer. Ask the person on the front desk and you get a third. All of them are right about something and none of them is looking at the whole picture, which is why so many small businesses automate the wrong thing first.
The fix is not more enthusiasm. It is a short, honest inventory of the repeatable work in the business, with a number on each line.
Three questions per job
For every task that happens more than a few times a week, we ask three things. How many hours does it consume across the team in a month? What does an hour of that person's time cost, including what they are not doing while they do it? And what happens if it goes wrong: is a mistake embarrassing, expensive, or a regulatory problem?
The first two give you a pounds-per-month figure. The third tells you how carefully the automation needs to be built and how much human oversight it needs. A job worth £1,400 a month that carries low risk is an obvious first candidate. A job worth £1,400 a month where a mistake means a complaint to the FCA is a second-phase candidate, not a first one, and it needs approvals built in.
What usually comes out on top
Across the roadmaps we have run, the same handful of jobs keep landing in the first thirty days, and they are rarely the ones the owner named. Inbound enquiries are almost always first: answering the phone and the inbox, qualifying who is worth a conversation, and booking the ones who are. Quote follow-up is a close second, because most businesses send the quote and then wait, and the money is in the chase. Invoice reminders and payment collection come next. Filling cancelled slots, for anyone who runs a diary, is often the biggest single saving per hour of build.
These share a shape. They are frequent, the rules are simple once someone writes them down, the systems involved (phone, inbox, calendar, accounting) all have decent integrations, and the cost of a mistake is small and recoverable.
What should wait
Anything where the rules live in one person's head and change weekly. Anything that touches money leaving the business without a check. Anything where the volume is low enough that a person does it in twenty minutes on a Friday. And anything that depends on a system with no integration, where the automation would have to click around a screen like a person. Those can all be done. They just should not be done first, because the payback is slower and the build is more fragile.
There is also a category we tell people to leave alone entirely. Payroll, for most small businesses, is one. So is anything with a legal signature. The saving is small and the downside is not.
The 90-day shape
A good roadmap puts two or three jobs in the first thirty days, all low risk and high volume. Days thirty-one to sixty add the jobs that need approvals or a system connection that takes longer. Days sixty-one to ninety are for the things that only make sense once the first agents are producing data, like a weekly insight brief that reads the numbers and says what changed.
Everything else goes on a list with a date to revisit it. Not because it is a bad idea, but because a business that automates three things well in a quarter is in a far better position than one that starts twelve.
The number that matters
When the roadmap is done, add up the pounds-per-month on the first-thirty-days jobs. If it is not comfortably more than the cost of building and running them, do not build them. The point of scoring everything first is that you find that out before you have spent anything.