Start with the boring bit.

The question we get asked most often, once the word "AI" has been in the room for a while, is the one nobody quite wants to ask out loud. Where do I actually start?

The tempting answer is the shiny one. A plugin that puts Claude inside Excel. A copilot in PowerPoint. An assistant living in a browser tab that turns a three-hour task into a forty-minute one for the person who remembers to open it. None of that is wrong. It's just not the thing.

It makes one person faster on one task. They still have to remember to use it. And — as we covered in a previous post — the moment client data gets involved, it drags you straight into a governance conversation most organisations haven't had yet.

The thing that actually compounds is different. It's the plumbing.

A form replaces an email thread. A weekly report writes itself on Friday morning instead of eating someone's afternoon. An invoice gets extracted and routed to the right place without anyone touching it. Nobody has to remember to use any of it. It just runs. And because it runs in a tenancy you control, it stays inside your governance boundary by default.

The low-hanging fruit lives on that side — the business doing things for itself in its own infrastructure, not individuals reaching for clever tools in a browser.

So the real question isn't "what should I automate?" It's "what am I already doing that's automatable?" — and the answer isn't missing from your business. It's hiding in your week.

Here are the ways we've found to surface it.

The template test. If you've already made a template for something — an email template, a quote template, a new-starter checklist — you've done the thinking half of automation. The template tells you the shape. The automation fills the slots. Start where you've already imposed structure on yourself, because everything hard has already been done.

The Monday test. Anything that happens on a schedule — the weekly report, the end-of-month reconciliation, the Friday timesheet chase. Repeating plus time-bounded is the purest signal there is. If a calendar tells you when to do it, a machine can probably help you do it.

The copy-paste test. Anything that moves data between two systems with a human in the middle. Email attachment into the accounting software. Form submission into a spreadsheet, then into an email. CRM into invoicing. If you are the integration between two apps, that's almost always the strongest tell on the list.

The interruption test. The small asks that break your focus. "Can you send me the latest quote?" "Where are we on invoice 1234?" "Can you approve this?" Each is tiny. In aggregate, they are the day. A self-service answer to a recurring question is almost always a candidate — and usually the one that buys the most back, because it saves two people's time every time it runs, not one.

The new-employee test. The best automation auditor in your business is the newest person on the team. They haven't normalised the weirdness yet. Ask them: what's the bit where you thought, surely this isn't how it works? That's gold. Write it down before they stop noticing.

The spreadsheet-as-middleware test. If there's one spreadsheet that's load-bearing — known in detail by one person, held together by macros and muscle memory, and the business would stutter if it went missing — that isn't a quirk. That's a process that grew beyond its container. Those are always worth looking at, because the fragility is doing hidden work you can now do properly.

None of those tests need a strategy. They need a week of paying attention.

And the instinct to watch out for is the one that dismisses things because they're small. Fifteen minutes, three times a week, is thirty-nine hours a year. Smallness is the point. You're not trying to impress anyone. You're trying to get your Fridays back.

The other instinct to watch out for is the one that wants to start with a list. Picking the ugliest single thing on your candidate list and fixing it end-to-end will teach you more about what your business actually needs than a six-month discovery project ever will. Most automation programmes die because they try to design a strategy before they've automated anything.

A quick note on where AI changes the answer, because this is the bit that tends to get muddled. Most of what comes up from the tests above is plain rules-based automation. When X happens, do Y. Sync this system to that one. Send the reminder on day three. No intelligence required — just reliability. That's most of the work and most of the value — which is the part of the conversation AI keeps drowning out.

AI earns its place when the input is messy and a human has been doing soft judgement on it. Classifying inbound emails. Pulling fields out of invoices that arrive in eighteen different layouts. Drafting a response in the house style. Summarising a meeting. The shape of those problems is different, and the tools are different, but the principle is the same: you don't need to know which kind you need until you've named the problem.

Find the task first. The tool question answers itself once you have.

We've been through this ourselves. We run our own business, and we've had to work out the same question on ourselves first — which one is worth moving on this week, and which one can wait. Same instinct either way. Pick one. Fix it. Go again.

The businesses getting ahead right now aren't the ones making the boldest claims about what AI will let them do. They're the ones who noticed the boring thing in their week and quietly removed it.

That's the whole exercise. Start with the boring bit.