Which tasks are worth automating?

The tasks worth automating are the ones that happen often, follow a rule you could explain in a sentence, start out on a computer rather than on paper, do little harm if something goes wrong, and free up time that's genuinely valuable. Automating simply means getting software to do a routine job for you, the way a standing order pays a bill without you remembering to.

Choosing the right task matters more than choosing the right tool. UK surveys put the share of businesses using AI anywhere between 16% and 54%, and as we explain in how many businesses actually use AI, most of that gap is between businesses that have AI somewhere and businesses that have done anything deliberate with it. Picking the right job is usually what gets you from one to the other.

Here are five tests. It's worth running the last two weeks of your work through them.

Does the task happen often?

How often a task happens matters more than how much you dislike it. Something you do forty times a month without really thinking about it is usually a better candidate than something you dread once a quarter.

The reason is simple arithmetic. Setting up any automation takes time to get right, and that time is spent once, whether the task happens weekly or yearly. A quarterly job has four chances a year to pay that back. A daily one has around two hundred and fifty.

This test alone rules out a lot of the ideas that come to mind first. The task you hate most is often not the task you do most.

Can you explain the rule in a sentence?

Try saying out loud what decides the outcome. "If the invoice is more than fourteen days overdue and we haven't already chased twice, send the second reminder." That's a rule. It can be written down, and software can follow it.

Now try: "I decide what to quote based on how the job looked when I walked round it." That isn't a rule. It's judgement built up over years of experience, and most of it has never been put into words.

That difference matters more than any choice of software. Work that's difficult because of the process can be handed over. Work that's difficult because of judgement is much harder to hand over, and attempts to do so tend to produce answers that sound confident and plausible but are wrong, which can be worse than no answer, because someone has to catch them.

A useful check: could you write a page that lets a capable new starter do this without asking you anything? If yes, you have a rule. If they'd need to ask you about half the cases, it's really judgement that looks like a process.

If you've found a task with a clear rule and plenty of repetition, we're happy to set up an automation that follows it for you.

Does the task start on a computer?

Something that starts as an email, a form on your website, a calendar entry, a line in your accounts software, or a message in a shared inbox can usually be automated today. Something that starts as a photo of a handwritten note, or a conversation nobody wrote down, has a gap at the start that has to be closed first.

That gap can be closed, and closing it is often the single most useful thing to do, but it's a separate job, and it's easy to skip. If the first step of a process is "someone remembers to tell me", nothing after that can be fully reliable.

If paper notes or phone messages are where things get lost, we're happy to look at automating the way that information gets captured in the first place.

What happens if it gets something wrong?

Ask two things: how would you find out, and how much would it have cost by then?

Some mistakes are obvious straight away. A draft that reads badly gets rewritten before it goes anywhere, and costs you thirty seconds. Others are quiet and build up, like a figure put in the wrong category every week for eight months and only spotted at year end. A few can't be undone, because they've already reached a customer.

It's best to start where a mistake is only a mild embarrassment. That isn't just a precaution for the early days. It's a lasting rule about which work suits which approach. Anything that goes out without a person checking it should be work where the worst realistic outcome is having to send a correction.

This is the test that's most tempting to skip, because the exciting projects are usually the customer-facing ones. Those are the ones best left until later.

Whose time does it free up, and is that time worth anything?

This is the test that decides whether the whole thing was worth doing.

If you free up two hours a week of an owner whose time can be charged to customers, that time has real value. It can be billed, spent winning new work, or simply taken back. If you free up two hours of a salaried employee who then has slightly less to do, you haven't gained anything you can measure. The cost of the automation is the same in both cases. The return isn't.

Be honest about which situation you're in, and be careful with published figures on time saved. Many are based on people estimating their own savings rather than anyone measuring them. Your own numbers are the ones that count.

What these tests won't tell you

Run your two weeks through those five tests and you'll have a shortlist. That's well worth doing, and it only takes an afternoon with a notepad.

What the shortlist can't tell you is which ideas will actually work with your systems. That's where things get fiddly: whether your accounts software lets other programs connect to it (we've explained how programs connect to each other separately), whether your customer system's export includes the details you need, whether two systems record a customer differently, and what happens if an automation half-fails in the middle of the night. An idea on a shortlist is a starting point. Most of the work is in turning it into something that still runs reliably months later.

That's the part we take on. If you'd like someone to take your shortlist and build the automations that make sense, we're happy to help, and you can see how our pricing works.

Either way, do the shortlist first. Automating something that didn't need doing is a far more expensive mistake than picking the wrong tool.

Sources

• AI adoption research - Department for Science, Innovation and Technology, updated 13 February 2026; 3,500 UK businesses with five or more employees, fieldwork 12 February to 2 May 2025
• Half of SMEs using AI - with limited headcount impact so far - British Chambers of Commerce, March 2026

Checked on 22 September 2026.