Ask AI Where Your Business Is Losing Money

You already know your business is bleeding money somewhere. You just do not know where.

Your business has a leak. You just have not found it yet.

It is Monday morning at Third Coast Physio, a six-person clinic Priya Anand runs out of a converted shopfront in Melbourne. Before she sees her first patient, she spends an hour doing the same thing she does every Monday: opening Cliniko to check the week's bookings, opening Xero to see what actually landed in the account, opening the insurer portal to chase claims, and scrolling a staff roster spreadsheet nobody has renamed since 2024.

Four systems. Zero of them talking to each other.

Priya is not disorganised. She is running the business the only way small business owners have ever been able to: by holding the whole picture in her head, because no single tool shows it to her. That worked when the clinic had two staff. It does not work at six, and it will not work at ten.

Here is the part that should bother you: the money her clinic is losing is not hidden. It is sitting in her own systems right now, in plain sight, waiting for someone, or something, to ask the right question.

You are the reconciliation, and that is the actual problem

Walk into almost any small business and you will find the same pattern under a different name. A plumber checking Checkatrade leads against a paper invoice book. A bakery owner cross-referencing a POS export against a supplier statement by eye. A UK home-care agency, in a real case a consultant named Manny Amoah worked on this month, running finance in Sage, staff in a separate HR tool, and operations in a notes app.

In every case, the owner is the integration layer. They are the one manually carrying a number from one screen into another, and manually noticing, or more often not noticing, when two numbers that should agree do not.

That is not a discipline problem. It is a systems problem wearing a person as a workaround.

And the moment you frame it that way, the fix stops looking like "try harder at bookkeeping" and starts looking like something else entirely: hand the reconciliation to something that does not get tired on a Monday.

A chatbot answers a question. An agent finds the ones you did not ask

Most business owners have already used AI for finances in the smallest possible way. You paste in a spreadsheet and ask, "how much did we spend on software last month?" You get an answer. That is a chatbot: useful, one-shot, forgotten by tomorrow.

What changed this year is the other mode. On 27 July, Corey Haines, a well-known marketing and AI builder, posted a Claude skill called /personal-cfo: give it your take-home pay and drop in a bank statement, and it categorises every line, hunts down subscriptions you forgot you were paying for, and hands back an honest monthly budget with the money it found circled.

That was built for a personal bank account. The same idea, aimed at a business's systems instead of one person's card statement, is where this gets genuinely useful for you.

How can AI find the money my small business is losing?

You give an AI agent like Claude a plain description of your business and a list of the systems that touch money or time in it, then ask it to map where the numbers stop agreeing with each other. It reads your bank export, your invoicing tool, your booking system, and your staff roster the way an outside auditor would: looking for late payments nobody flagged, duplicate or forgotten subscriptions, hours burned on manual admin, and risk signals buried in operational data. You are not asking it a question. You are asking it to go looking.

Manny Amoah, a UK finance consultant who spent 15 years across Visa Europe, the Cabinet Office and the NHS before starting his own AI consultancy, ran exactly this discovery for a home-care agency client. One structured prompt into Claude, business profile and systems list included, surfaced three things nobody at the company was tracking.

None of that money was hidden. It had nowhere to go. The receivables were sitting inside the accounting software the whole time. The four hours of manual reporting happened every single week in front of the CEO. The compliance gaps were visible in the care-monitoring data, if anyone had looked at a month of it at once instead of one day at a time. An AI agent did not invent the £86,400. It just read what was already there and said it out loud.

Isn't this just an expense spreadsheet with extra steps?

Fair question, and the honest answer is: a spreadsheet shows you numbers. It does not show you relationships between numbers, and relationships are where the money actually hides.

A spreadsheet will not tell you that your revenue is up while your margins are quietly falling. It will not connect "three staff called in sick this week" to "deliveries are about to slip." It will not notice that the same client has been on a 90-day payment cycle for four straight invoices while your own terms say 30. Those are cross-system patterns, and cross-system patterns are exactly what a language model reading multiple exports at once is built to catch that a person skimming one tab at a time is not.

That is the actual shift. Not "AI does maths." AI reads across your whole business in one sitting and tells you what does not add up.

Where the money actually leaks, by workflow

Run your own money-discovery prompt this week

You do not need a consultant to start this. You need one hour and the exports you already have.

1. List every system that touches money or time. Booking or job software, invoicing, payroll, your bank, and anywhere compliance or quality gets logged. Most small businesses have four to six of these, no more.
2. Pull the last two or three months of exports. A CSV from your accounting tool, a PDF bank statement, your booking software's report view. You are not building anything permanent yet, just gathering the raw material.
3. Write a discovery prompt, not a question. Tell Claude your business profile, list the systems by name, paste in the exports, and ask it explicitly to find: receivables sitting past your own payment terms, duplicate or forgotten subscriptions, recurring manual work that could be automated, and any risk signal that only shows up when you look across a full month.
4. Use a mid-tier model for this pass. Claude Sonnet handles a discovery prompt like this well and burns far fewer tokens than Claude Opus. Save Opus for when you are building the follow-up system, not for the first scan.
5. Turn the findings into a one-page weekly view. A Claude Artifact, a shared doc, or a simple dashboard. The value is not the one-off answer, it is having somewhere the same findings surface every week without you re-running the prompt from memory.
6. Assign an owner to each leak. A discovery is worthless if nobody chases the £60,000. Put a name and a date against every item the AI surfaces.

Once you have run the discovery once, there are three honest ways to make it permanent, in roughly ascending order of effort and power. Bring in someone who has already built a version of this, and you walk into that conversation already knowing exactly what you need instead of paying them to figure out your business first. Wire it together yourself with a no-code tool like n8n or Make, which is free to start and realistic for a weekend if you are comfortable learning something new. Or go technical with Claude Code and build the connections directly, which takes more setup but scales further once your systems talk to each other automatically instead of on a schedule you have to remember.

What changes once you actually look

This week, you run one prompt and get a list instead of a gut feeling. A specific number next to a specific invoice, not a vague sense that cash flow feels tight.

This month, that list turns into a habit. The weekly report shows up whether or not you remembered to ask for it, and the leaks that used to hide for a full quarter get caught inside thirty days instead.

Six months in, the business runs on evidence instead of memory. New hires get shown the discovery process on day one, because it is now just how the business checks itself.

Same clinic, same care agency, same plumbing business. A completely different relationship with its own numbers.