AI Automation Examples, Priced: Buy It, Wire It, or Build It Yourself

A while back an owner asked me to quote for a system that would chase her overdue invoices. Someone had already told her it would run to a few thousand pounds and she had budgeted for it.

Ten minutes into the meeting I opened her accounting software and switched on the automatic payment reminders sitting in there, turned off, included in the subscription she had been paying for four years. The meeting was booked for an hour. It took twenty five minutes and there was nothing to invoice for.

I lead with that because almost every list of AI automation examples online is published by a company that sells automation, and none of them is ever going to tell you the fix is a setting you already own. They hand you a category called invoice processing with no price attached and no clue about who should build it.

So here is my version. What makes it different is the label on each one: your software already does it, it is an hour of wiring, it is an afternoon with Claude Code, or it is genuinely worth paying a human for. Real prices where I could verify them from the vendor, and I will say so where I could not. If you want the short version of how I work, it is on my about page.

What makes an AI automation example worth copying?

Two questions sort the useful examples from the vanity projects. Does the job happen often enough to matter? And can you tell quickly when it gets something wrong?

Something you do twice a year is not worth automating no matter how tedious it feels in the moment. And if a mistake would sit undetected in your accounts for six weeks, that automation needs a human approving each run. Price that in before you build rather than after.

Worth defining the term while we are here, because it gets stretched. An AI automation is a process that fires on a trigger and has a judgement step in the middle. Ordinary automation moves data that arrives in the same shape every time. The AI part is what lets the same process read a supplier PDF or a call transcript, work out what it is looking at, and act on it.

It is worth knowing the ground you are standing on. The Office for National Statistics found around 35 percent of UK businesses with 10 or more employees were using at least one AI technology by June 2026, up from roughly 12 percent in late 2023. Then look at what they use it for: large language models at 18 percent, visual content creation at 16 percent, machine learning for data processing at 12 percent.

Adoption went wide and stayed shallow. Nearly everyone has used AI to write an email. Very few have wired it into the thing that actually eats their Tuesday.

Which of these does your software already do?

Start here, because the cheapest automation is the one you have already bought.

Bank reconciliation is the clearest case. Xero and QuickBooks both suggest matches for incoming bank transactions and let you set rules that code the recurring ones for you. Owners still sit there matching lines by hand on a Sunday, then get quoted for a smart reconciliation build. Turn the feature on and spend the money on something the software cannot do.

Automatic invoice reminders are the second. Both packages will send a chaser a set number of days after the due date and stop the moment payment lands. That is the entire scope of a lot of the credit control automation being sold at the moment.

Receipt and bill capture is worth checking before you commission anything. Dext and Hubdoc read supplier documents and push them into your ledger, and if your accountant put you on their package you may already have one of them included and never switched it on.

I am opening with the unglamorous tier deliberately. If turning on what you own makes the pain go away, you have saved yourself a build. And you now know the next quote you receive is for something genuinely new.

Which examples are an hour of wiring rather than a project?

The next tier connects two apps you already use and puts a model in the middle to make one decision. Make and Zapier are built for this and neither expects you to write code. Make has a free plan with 1,000 credits a month, then Core at $9 a month. Zapier is free to 100 tasks a month, with Professional from $19.99 a month billed annually.

Here is the one I would wire first, in full.

A customer fills in the enquiry form on your website. The scenario fires and sends the message to Claude with a short instruction telling it what your business does and what your jobs look like. It classifies the enquiry and scores how urgent it is. Then it drafts a reply in your tone. A new job goes to whoever quotes. Anyone chasing an existing order goes to operations instead, and obvious spam goes nowhere at all. The draft lands in the right inbox as a draft, never sent on its own, and a record appears in your CRM.

Setup takes about an hour if your form and CRM are common ones.

The value is not really the draft reply. It is that nothing sits unclaimed in a shared inbox for two days because four people each assumed someone else had picked it up.

A few more that live in this tier:

Google review replies. Only a minority of businesses answer reviews at all, while BrightLocal’s 2026 consumer survey found 89 percent of consumers expect a business owner to respond and 81 percent expect it within a week. A scenario that drafts a reply matched to the star rating, holds it for approval, and pings you immediately on anything under three stars closes that gap without you living inside Google Business Profile.

Client onboarding. One trigger fires the welcome email, the document request, the calendar invite, the folder creation and the internal checklist. No AI judgement needed at all, which is rather the point.

Call notes into the CRM. Fathom or Otter transcribes the call, the model pulls out the actions and the agreed next step, the CRM record updates itself, and a follow up email waits as a draft.

None of that needs a developer. If someone quotes you four figures for one of them, they are charging you for the hour you did not fancy spending in Make. If you are not sure which of your own jobs sit at this level, the free AI Opportunity Scorecard takes about two minutes and gives you an estimate of the hours and money currently sitting in your process.

Which examples are worth building yourself with Claude Code?

The tier above wiring is where the quotes get silly, and it is the tier that has genuinely changed.

These are the jobs that touch your own files and your own rules, the ones that never fit a template someone else drew. They used to need a developer for exactly that reason. UK agency quotes for this kind of build tend to land somewhere around £2,500 to £15,000 with a monthly retainer on top, although every source publishing those ranges also sells builds, so treat them as a ceiling to argue with rather than a benchmark to accept.

Claude Code moved that line. It comes included with Claude Pro at $17 a month billed annually or $20 monthly, and you drive it by describing the job in plain English while it writes and tests the script. If you can write a clear brief for someone covering your desk next week, you can brief it.

Four from this tier that pay for themselves fast.

Supplier invoices into draft bills. A PDF lands in your accounts inbox. The script reads it, pulls out the supplier, the date, the net, the VAT and the line items, then creates a draft bill in Xero waiting for approval. Nothing posts without a person. The running cost is close to nothing: Anthropic’s own worked example puts 10,000 support tickets through Claude Haiku 4.5 for about $37, and an invoice is a smaller job than a support ticket.

The Monday morning summary. A scheduled script reads last week’s sales, the jobs completed, the outstanding invoices and the cash position, then writes a one page plain English note on what changed and what needs your attention. Plenty of owners rebuild that by hand every week out of four different systems, which is the most reliably wasted hour in a small business.

Quotes drafted from your own history. Point it at the folder holding your last twenty quotes and your rate card, and it drafts the new one priced the way you actually price, not the way a generic tool guesses. You still check it before it goes. Nobody sells you this one, because it only works on your files.

The quarterly tidy up before your accountant sees it. Making Tax Digital for Income Tax went live on 6 April 2026 for sole traders and landlords with qualifying income over £50,000, which means four quarterly updates a year rather than one January panic, with the threshold dropping to £30,000 in April 2027 and £20,000 in April 2028. A script that scans the quarter for uncategorised transactions, missing receipts and VAT flags that do not match saves half a day every quarter, and it is only going to matter more.

The full list of AI automation examples, by tier

What should you never automate?

Anything where a wrong answer reaches a customer before you have seen it.

That single rule is worth more than the rest of this article. Every example above that produces an outbound message stops at a draft deliberately. The failure mode with these tools is almost never a dramatic crash you would notice. It is a confident, plausible output that is quietly wrong in a couple of places and gets discovered six weeks later by a client who is now annoyed.

There is a legal edge to it in the UK. If your automation processes personal data you need a lawful basis for it, and the ICO expects most new AI use cases involving personal data to require a Data Protection Impact Assessment before launch. Your privacy notice has to say AI is being used and how. Any decision made solely by automation that carries a legal or similarly significant effect needs a route to human review. None of that is an argument against automating. It is an argument for keeping a person on the approve button for anything customer facing, which you wanted regardless.

Two others I would leave alone. Payroll, because the cost of a quiet error is somebody not being paid properly and finding out on payday. And the judgement your clients actually hire you for. Automating the thinking is not efficiency at that point, it is taking your own product apart.

How do you choose which one to do first?

Score the candidates on four things, and ignore which one would sound most impressive at a dinner party.

1. Count the hours honestly. UK small business owners spend an average of 11 hours a week on admin and finance work, roughly six working days a month, against 3.6 days a month on sales and business development, according to the SME Business Barometer that American Express runs with Small Business Saturday. Find where your own 11 hours actually goes before you build a thing.
2. Check the tier. Run the job down the ladder above. If it turns out to be a setting, you are finished this afternoon and you have spent nothing.
3. Check you can see it fail. If you could not tell within a day that the output was wrong, either add an approval step or choose a different job.
4. Do one. Not a programme. One process, live, proven over a fortnight, before you touch the second.

The ONS found the largest barrier UK businesses report is not cost. It is lack of expertise. That gap closes by finishing one small thing, not by reading another list of examples.

Start with the job you resent most

If you keep one thing from all of this, keep the tiering rather than any particular example. Once you can tell a setting you already own from an hour of wiring, and both of those from a real build, most of the sales pressure in this market simply stops working on you.

And start with the job you resent, not the one that would demo well. The invoice entry you do at nine at night. The report you rebuild every Monday out of four systems that will not talk to each other. Resentment turns out to be a good signal, because it tracks how often the work happens and how little judgement it really needs.

If you would rather have someone map your own processes against this ladder with you, that is what the AI Opportunity Audit is for. It is a fixed fee 60 minute session after a free 15 minute intro call, it finds the highest impact places to apply automation in your business, we build one live in Claude Code together, and you leave with a prioritised roadmap. Either way, pick one this week. Choosing from the list was never the hard part.