How to Automate Your Business: A Seven Step Plan You Can Run Yourself

I get asked how to automate a business by owners who have already read the lists. They know automation can chase invoices and send booking reminders. What they are really asking is something narrower and more practical: where do I actually start on Monday morning, and how do I avoid building something that quietly breaks in March while I am on holiday?

That second worry is a good instinct. Most guides stop at “pick a tool”. I want to take you further than that, through seven steps I use myself, with one real process carried from the first switch to the moment you stop thinking about it.

The running example is invoice chasing, because it is dull and nearly every business has it. Swap in your own process as you read.

How do you automate your business, in short?

You automate a business one process at a time: find the process, write it down, choose the lightest tool that can run it, test it alongside the manual version, then give it an owner and an alarm. Repeat. The order matters more than the tool.

The reason to bother is not abstract. The American Express SME Barometer published in July 2026 found UK small business owners spend around 11 hours a week on admin and finance, roughly twice the time they give to sales. That is the pool of hours this plan is fishing in.

Step 1: What can you switch on today without building anything?

Check the software you already pay for before you buy anything new. A surprising amount of business automation is a setting nobody has turned on.

Xero has automatic invoice reminders built in, with editable templates you can schedule for days before and after the due date. QuickBooks offers recurring invoices and reminders on most UK plans. Calendly can send confirmations, a reminder the day before and a follow up email afterwards. Outlook and Gmail both have rules that file supplier invoices into a folder the moment they land.

For the invoice chasing example, this step alone might solve most of the problem. Switch on Xero’s reminders, edit the wording so it sounds like you rather than a debt collector, and watch what happens for a month.

Only build what is left over.

Step 2: Which process should you automate first?

Automate the process that swallows the most minutes across a week, as long as a bad day would only be embarrassing rather than expensive. You find it with a stopwatch, not a brainstorm.

For one normal week, jot down every repetitive task and how long it took. Use a note on your phone. Be honest about the small stuff, because five minutes of copying a Stripe payout into a spreadsheet, done daily, adds up faster than one heroic afternoon of quoting.

At the end of the week, give each task a rough score. Minutes per week, multiplied by how often it happens, then halved if a mistake would reach a customer or move money. The top of that list is your first automation. It is rarely the one you would have guessed.

If you would rather get a quick estimate of where your hours are going before you start timing, the free AI Opportunity Scorecard takes about two minutes.

Step 3: How do you map a process before automating it?

Write it on one page as a trigger, then the steps in order, then the exceptions. If you cannot fit it on one page, it is two processes.

For invoice chasing it might read like this. Trigger: an invoice passes its due date unpaid. Steps: check whether the client has paid by bank transfer that has not been matched yet, send a polite reminder, wait seven days, send a firmer one, then flag it to you. Exceptions: the client on a payment plan, the one who always pays on the 28th, the invoice that is disputed, the council whose finance team will not pay without a purchase order number on it.

Those exceptions are where automations fail. Write every one you can think of.

This page is also where most people discover the process itself is a mess. Automating a messy process gives you a faster mess, so tidy it here, on paper, where changes are free.

Step 4: Which tool should run it?

Use the lightest layer that can do the whole job. There are three, and each one costs more effort than the one before.

Wiring is the right answer when data moves in the same shape every time. A Tally form submission becomes an Airtable row and a Slack message. Nothing to interpret.

Building is the right answer when something has to be read and understood. Say your clients reply to reminders with “paid Friday, ref INV 2041” or “can we split this over two months?” A wiring tool cannot sort those replies. With Claude Code, you describe the job in plain English, it writes the script, runs it against last month’s inbox, and shows you what it would have done. You are the reviewer, not the programmer.

This is the route I would push most owners towards once the settings run out. You end up owning the thing, and you learn a skill that carries to the next ten processes instead of renting it from an agency.

Step 5: How do you test an automation before trusting it?

Run it in shadow for two weeks. The automation does its work but stops one step short: it drafts the reminder instead of sending it, or writes to a test spreadsheet instead of the real one. You carry on as normal and compare.

This is the least exciting step and the one I would never skip. The first fortnight surfaces the exceptions your one page missed. Maybe a client with two contact emails gets chased twice. Maybe a credit note sends the balance negative and the automation panics.

Fix each one, then let it send for real.

Step 6: Who keeps the automation running afterwards?

A named person does, with an alarm and a written note. Automations break for boring reasons, usually a changed password or a supplier who redesigns their invoice PDF. The damage comes from not noticing for three weeks.

So give every automation three small things before it goes live. An owner, meaning one person whose job it is to fix it. A failure alert, which Zapier, Make and Power Automate can all send by email, and which a Claude Code script can send to your phone. And a short note, stored somewhere obvious, saying what it does, what triggers it and where the logins live.

That note is what separates owning an automation from depending on whoever built it. It is also the first thing I would ask any agency for before signing.

If you are putting customer data through any of this, treat it like any other processing under UK GDPR. Send only the fields the automation needs, and prefer business plans of AI tools that do not train on your data.

Step 7: When should you automate the next process?

When the first one has run for a month without you touching it. Not before.

Then go back to your stopwatch list and take the next item. The second automation takes roughly half the time of the first, because the one page map and the shadow fortnight are now habits rather than homework. By the fourth, most owners stop asking how to automate their business and start noticing processes that should not exist at all.

Where I would start if this were your business

Honestly, with the settings. Before any tools or any build, spend an hour inside your accounts software and your email rules, switching on what you have already paid for. It is the fastest win in this whole guide and it costs nothing.

Then pick one process and take it through all seven steps properly, including the dull ones. If you would like help doing that, my AI Opportunity Audit is a fixed fee 60 minute session where we review your processes, find the highest impact place to start, build a real automation live in Claude Code, and you leave with a prioritised roadmap. You can read how I work and why first if you prefer. Either way, the goal is that you end up able to do the next one yourself.