How to Decide What to Automate in Your Small Business

Published 2 August 2026 · 7 min read

You have a list of things quietly eating your week. Invoicing. Chasing replies. The weekly report you are fairly sure nobody reads. Someone told you to automate it, so you lost a Saturday wiring up a tool — and three months on you are still doing the task by hand, except now there is a broken automation to maintain as well.

That is the normal outcome, and it is not because you picked the wrong software.

Almost everything written about how to decide what to automate in your small business is published by companies that sell automation software. They are structurally incapable of telling you the best answer, which is frequently: stop doing this task at all. So you get lists of twenty-five things to automate, and no way of working out which two apply to you.

Here is the order we actually use with clients. Four questions. The order is the whole point — most founders start at the fourth one, which is how you end up automating something that should have been deleted.

Question 1: Does this task need to exist at all?

Before you make something faster, check whether it should be happening.

A surprising share of recurring work exists because it made sense two years ago. A report built for a client who left. An approval step added after one bad week in 2024. A weekly check-in that could be monthly.

The test: if you stopped doing this tomorrow, who would notice, and how long would it take them? If the honest answer is "nobody" or "about a quarter," you have found something better than an automation. You have found a deletion.

Deletions are free, instant, and never break. Automations are none of those things.

Question 2: Do you do it the same way every time?

Automation does not fix a messy process. It runs the mess faster and removes the human judgment that was quietly patching the gaps.

Write down the steps of the task. Then do it by hand for two weeks and mark every time you deviate from what you wrote. If you go off-script more than about one time in seven, the process is not settled enough to automate. Stabilise it first — same steps, same order, same place — and revisit in a month.

This is the step people skip, and it is the reason most small business automations get switched off within a year.

Question 3: Does it need a judgment call?

Most tasks are not one task. They are a sequence, and only part of it needs a person.

"Follow up with quiet leads" sounds like judgment work. Break it apart and it is mostly not: pulling the list of people who have not replied in fourteen days is mechanical. Deciding which three are worth a personal note is judgment. Writing that note is judgment.

So automate the pulling. Keep the deciding. You get the time back without sending the kind of message that makes a client feel processed.

The rule: anything involving apology, negotiation, bad news, or a genuine exception stays human. Automation should open the door, not stand in it.

Question 4: Does the arithmetic work?

Only now does it become a numbers question — and this is where the honest answer is usually no.

Work out two figures.

• What it saves: minutes per run × runs per year, in hours.
• What it costs: hours to build and test, plus hours per year keeping it working.

Then apply what we call the 5× rule: over the first year, an automation should save at least five times what it costs to build and maintain.

Five, not break-even, for three reasons. Your build estimate is optimistic — everyone's is. Automations break when a tool updates, and you will fix them at the worst possible moment. And the task you are automating may not exist in eighteen months.

The 5× buffer absorbs all of that. If the honest number is 2×, do the task by hand and spend the Saturday on something else.

A worked example

A client of ours sent a weekly status email to nine customers. It felt like it took forever, and the obvious move was to automate it. Running the questions in order:

Question 1. Four of the nine were on quarterly retainers and had never once replied. We asked. None of them wanted a weekly email. Cutting those four removed 44% of the work before anyone touched a tool.

Question 2. The remaining five were mostly consistent, except the numbers came from three different places and one was sometimes out of date. Not stable. We put the three figures in one sheet first.

Question 3. Pulling the numbers is mechanical. The two lines of commentary at the top — the part clients actually read — is judgment. Split it.

Question 4. Five emails at 12 minutes each is an hour a week, so 52 hours a year. Automating the number-pulling saves roughly 60% of that, about 31 hours. Building it took 5 hours, with maybe 2 hours a year of upkeep. That is 31 ÷ 7, or 4.4× — just under the line, so it was a genuinely marginal call. We built it, because the sheet was useful on its own.

But look at where the win actually came from. The automation returned 31 hours. Question 1 returned about 40 hours, for free, in a ten-minute conversation.

That is the pattern almost every time, and it is the one thing a company selling automation software will never lead with.

What not to automate

Leave these alone, even when you technically can:

• Anything you do a handful of times a year. The maths never works, no matter how annoying it is.
• A process you have changed twice in the last three months. It is still moving. Automating it now just makes it harder to change again.
• First contact with a new enquiry. An automated reply is fine as a receipt. It is not fine as the answer.
• Anything where being wrong is expensive. Invoicing the wrong amount automatically is much worse than invoicing late by hand.
• Work you are avoiding because you dislike it. Dislike is not the same as inefficiency. Sometimes the right answer is to delegate it, or to accept that it takes twenty minutes and do it.

Questions we get asked

What should a small business automate first?

Whichever task passes all four questions with the highest multiple — not whichever one is easiest to set up. In practice it is often invoice reminders or appointment scheduling, because both are frequent, identical every time, and need no judgment. But run your own numbers rather than copying someone else's list.

Do I need special software?

Often not. A recurring calendar reminder, a saved email template, or one spreadsheet formula solves a surprising number of these problems with no subscription and nothing to maintain. Reach for a dedicated tool when the task has to cross between systems that cannot talk to each other.

How do I know if a process is stable enough?

Write down the steps, work through them by hand for two weeks, and mark every deviation. More than about one in seven and it is not settled. That is question two, and it is worth the fortnight.

Can AI handle the judgment tasks now?

It can draft the judgment. It cannot make it. AI is genuinely good at producing a first version of something that needs a decision — a reply, a summary, a recommendation — and genuinely poor at being accountable for it. Treat it as a way to shorten the work in question three, not a way to skip question three.

The short version

• Does it need to exist? If not, delete it. Free, instant, never breaks.
• Same way every time? If not, stabilise it before you automate it.
• Judgment involved? Split the task. Automate the mechanical half.
• Does it clear 5×? If not, do it by hand and move on.

Most founders start at question four. Starting at question one is usually where the hours actually are.

If you want a hand with this

Working through a task list like this is most of what our AI-powered efficiency work involves — and a fair amount of the time the recommendation is to automate nothing and change the process instead.

If you have a list of things eating your week and you are not sure which two are worth attacking, send us a message. Happy to look at it with you.