Straight answer: the most useful thing I've done with AI this year had nothing to do with content, captions or emails. I fed twelve months of business transactions into Claude and asked it to find what I was quietly overpaying for. It found £3,200 a year in waste in under an hour, and it took me longer to redact the bank details than it did to run the whole thing.
Everyone's pointing AI at the wrong job
Open any small business AI guide and it's the same five suggestions on repeat: write your social posts, draft your emails, build a chatbot, summarise your meetings, generate blog ideas. All fine. All fairly saturated by now too, because your competitor read the same article you did.
I've been rebuilding my own business in public for the last five years, and the AI habit that's moved a number for me isn't a marketing one. It's financial. I use it to go through my own spending the way I'd never sit down and do manually, because manually doing it is boring, and boring things don't get done in a small business run by one exhausted owner and maybe a part time VA.
Here's the bit nobody wants to say out loud: most solo consultants and small business owners have no real idea what they're spending money on month to month. Not because they're careless. Because there's no time, and reconciling a year of Xero exports against actual usage is not why anyone started their business.
What I did, step by step
This took me about 45 minutes total, spread over one evening. Here's the exact process.
- Exported 14 months of transactions from Xero as a CSV. QuickBooks and FreeAgent do the same thing, it's usually under Reports or Transactions.
- Manually blanked out account numbers, sort codes and card numbers in a spreadsheet before uploading anything. This matters. Never paste raw bank data with full account details into any AI tool.
- Pasted the cleaned data into Claude with this prompt, close to word for word: "Here are 14 months of business transactions. Group them by vendor, flag any subscription or recurring charge that appears more than once, flag any vendor whose monthly charge increased by more than 5% without an obvious reason, and list anything that looks like a duplicate service doing the same job as another line item."
- Cross-checked the flagged list against what I use, which took maybe fifteen minutes with a coffee.
- Cancelled, downgraded or queried each one directly with the supplier.
That's it. No fancy tool, no dashboard, no automation build. Just a spreadsheet and a specific prompt.
What it found
Three unused subscriptions I'd forgotten I had, totalling £187 a month between them. A transcription tool I'd trialled once for a podcast I never made, a project management tool I'd migrated away from eight months earlier but never cancelled, and a stock photo subscription I hadn't logged into since spring.
A supplier who'd quietly raised their monthly fee by 9% with a one-line note buried in an invoice PDF I'd never opened, because who reads every invoice PDF.
A part time contractor whose logged hours on paper didn't match the pattern of actual deliverables coming back to me, which led to an honest, unemotional conversation rather than an accusation, and a revised arrangement that worked better for both of us.
Total: roughly £3,200 a year, found in under an hour, sitting in data I already owned.
The uncomfortable part of the AI content boom
Here's the thing most people writing about AI for small business won't tell you, mostly because it's bad for engagement. Every brand pushing out AI-written blog posts, AI-written LinkedIn updates and AI-written email sequences is starting to sound the same. Same rhythm, same three bullet points, same "here's what nobody tells you" hook, ironically enough. When everyone has the same tool doing the same job, the content stops being a differentiator. It becomes table stakes, like having a website.
The brands that still stand out are the ones treating AI as a tool inside a distinct system, not a replacement for one. Look at how Ahrefs built its marketing strategy around original data and product-led proof rather than volume content, or how Away built a brand people trust on consistent customer experience long before AI existed to write about it. The content angle was never the moat. The operations behind it were.
So if you're going to spend your limited AI hours on one thing this month, spend it on the boring stuff behind the curtain, not the eleventh version of a caption nobody will remember by Thursday.
Setting this up in your own business this week
You don't need a system, a subscription or a consultant to start. Here's the version anyone can do this week:
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
- Block 30 minutes, no more. This is a one-sitting job, not a project.
- Export the last 12 months of transactions, not just the last three. Twelve months catches annual renewals you'd otherwise miss entirely.
- Redact anything identifying before it goes anywhere near an AI tool. Account numbers, sort codes, full card numbers, none of it needs to be there for the categorisation to work.
- Ask the tool to flag three specific things: duplicates, unexplained increases, and anything you can't immediately name what it's for.
- Set a recurring calendar reminder to do this every quarter, not once a year. Subscriptions creep back in.
If you'd rather have someone build a repeatable version of this into how your business runs, along with the dozen other places AI can quietly save time rather than just produce content, that's the kind of practical work an AI implementation coach should be doing with you, not just workshopping prompts for Instagram captions.
What AI won't catch, and why that matters
It won't catch context. It flagged a "duplicate" subscription for me once that was two different tools doing two different jobs that happened to have similar names. It won't know that the £400 you spent on a conference in March was a deliberate investment, not waste, unless you tell it. It has no idea which relationships matter more than the invoice suggests, and it will never tell you to keep paying a supplier slightly over market rate because they've bailed you out twice in a crisis. That judgment still sits with you.
There's a discipline element here too, one Napoleon Hill would have recognised long before any of us had AI to help with it. Definiteness of purpose and accurate self-assessment were core to how he framed business lessons that still hold up nearly a century later, and going through your own numbers honestly, even with a tool doing the heavy lifting, is still an act of discipline nobody can outsource. Howard Schultz talked about the same thing from a different angle, controlling cost without cutting the things that mattered to the customer experience, a balance he had to relearn more than once as Starbucks scaled, as covered in the business lessons from Howard Schultz. The tool changes. The judgment about what to keep and what to cut doesn't.
Where this fits with the rest of your AI use
I'm not saying stop using AI for content. I use it for drafts constantly. But if you're a small business owner with three hours a month for AI experiments, don't spend all three on content. Spend one on this. It's the least glamorous use of AI I've found and it's also the one that's put actual money back in my account rather than just saved me typing time.
If you want a more visible, customer-facing use of the same "let the tool do the boring counting" idea, building something interactive on your own site, like a cost calculator that does this kind of audit logic for your visitors, is a good next step, and there's a decent walkthrough on building interactive calculators that convert that applies the same underlying thinking to your own website rather than just your own books.
Frequently asked questions
Is it safe to upload bank statements to ChatGPT or Claude?
Not in raw form. Always redact account numbers, sort codes, and full card numbers before uploading anything. Category, amount, date and vendor name is all the tool needs to do the job, and none of that is sensitive on its own.
How much can a small business realistically save doing this?
It depends on how long it's been since anyone looked, but £100 to £300 a month in forgotten subscriptions, duplicate tools and unnoticed price increases is a common range for a business that hasn't audited spending in over a year.
Which AI tool is best for this, ChatGPT or Claude?
Both handle it well. Claude tends to be slightly better at holding a large spreadsheet in context without losing track of earlier rows, which matters when you're pasting in a full year of transactions rather than three months.
How often should a small business do this kind of audit?
Quarterly, not annually. Subscriptions and small price increases creep back in faster than you'd expect, and a quarterly 30 minute check catches them before they've cost you three months of waste rather than twelve.
Related reading: Shadow AI: What Your Team Is Doing With ChatGPT Behind Your Back and How to Build an AI-Powered Calculator That Brings You Leads.
