Automating a one-person back office: what to automate first

There is no team here. No one answers the emails I don't get to, no one chases the newsletter sign-up that fell out of a spreadsheet, no one double-checks that a form actually reached my inbox. Whatever doesn't run itself, I run myself, at the expense of whatever I was meant to be building that day. That's the entire case for automation when you're one person: it isn't about efficiency in the abstract, it's about which hours of your week belong to the work only you can do, and which get eaten by coordination that a form, a template or a script could handle just as well.

I didn't automate everything at once, and I'd argue you shouldn't either. Doing that first, before you know which jobs actually recur and which just felt urgent once, wastes the one resource a one-person business can't get back: your own attention, spent building the automation instead of the business.

The rule I actually used

I didn't sit down with a grand plan. What I used, without writing it down until now, was three questions about any recurring task:

Does it happen often enough to matter? Does it need the same response every time? And does it tend to land at the worst possible moment?

If the answer to all three is yes, automate it. If a task is rare, or every instance genuinely needs a different answer, or it never seems to clash with anything else you're doing, leave it alone — the automation will cost more attention to build and maintain than the task itself ever would. It's a short filter, but it's done more work for me than any tool I've bought.

Attention is the only resource a one-person business can't buy back. Automation should be spent protecting it, not spent for its own sake.

What went first: the inbox

The first thing that failed the "worst possible moment" test badly enough to fix was the contact form. Early on, every enquiry meant me writing a reply from nothing, at whatever time it happened to land — which, inconveniently, was often the same evening I was trying to finish a Vault guide or set up a takeaway's phone line. The fix wasn't clever: a proper contact form with a clear prompt, wired straight to my inbox with no server of my own to maintain, and an honest note on the page about how fast I actually reply — one working day, Monday to Friday. That single change did two things at once. It meant nothing depended on me being at my desk the second a message arrived, and it set the right expectation up front, so a slower reply on a Saturday isn't a broken promise.

What went in next: delivery

The next job on the list, once the Vaults store opened, was the moment right after someone pays: getting the guide into their hands. Doing that by hand — watching for an order, attaching a file, writing an email — is exactly the kind of task that fails all three of my questions the hardest. It happens on no schedule I control, it needs the same response every single time, and it lands whenever a customer happens to buy, which is not going to be convenient. So it was automated from the first sale: payment clears, the download appears on the order page and the link goes out by email straight away, and if it ever gets lost, a resend request gets it sent again within a working day. I'd rather build that before the first sale than patch it together after the first person emails asking where their file is.

The one I built for someone else first

NeverMissed came from the same logic, just applied to somebody else's back office rather than mine. A takeaway's busiest moment — Friday and Saturday evening — is precisely when a missed call happens, and precisely when a rushed answer goes wrong. The three questions apply just as cleanly there: it happens constantly, the response needed is nearly always the same shape (take the order, confirm it, send it through), and it always lands at the worst possible time. Building that for a takeaway taught me something I then applied back to my own operation: the tasks worth automating are rarely the interesting ones. They're the boring, repetitive, badly-timed ones nobody wants to keep doing by hand.

What I have deliberately kept manual

Not everything should be automated, and I think this list matters as much as the one above.

• Refunds and complaints. These need a person reading the actual situation, not a template. I reply to these myself, every time.
• Setting up a new NeverMissed shop. I go in person, load the real menu, and test a live call before I leave. That's the part of the product that's supposed to feel human, so it stays that way.
• Anything a customer flags as unusual. The same principle that has NeverMissed's AI hand a call to a person applies to me: when something doesn't fit the normal shape, a script shouldn't be the one deciding what happens next.
• Writing the actual work. The guides, this note, the copy on every page — automation clears the coordination around the work. It was never going to do the work itself, and I wouldn't want it to.

If there's a single takeaway, it's that automation in a business this small isn't a project you finish. It's a habit of noticing when a task has started repeating itself in a way that costs you at a bad time, and building it away before it becomes the reason something else doesn't get done. The order doesn't matter as much as the willingness to keep asking the same three questions.