The first finance job most owners hand to AI is the one it does worst: add up a column and promise the total is right.
A small business can use AI in finance and admin, and most should. Use it for the words around the money: chasing late invoices, explaining why costs rose, summarizing a supplier contract, suggesting categories for a pile of receipts, drafting the note to your accountant. Do not let it set a number that goes in your books, move money, or change who gets paid until a person has checked it against the original document. Hold that boundary and the rest of the job is safe.
Why is finance and admin different from other AI jobs?
Finance and admin is different because a mistake in the numbers does not look like a mistake. A clumsy marketing email reads badly and someone fixes it before lunch. A miscategorized expense or a transposed digit in a supplier's bank details reads perfectly well, and it sits in the ledger until a quarter end, a tax return or an empty bank account brings it to light.
Owners already sense this. Intuit's 2026 AI Impact Report, published on 12 May 2026 and built on survey responses from more than 34,000 small and midsize business owners in the US, Canada, the UK and Australia, found that "adoption is highest in marketing, admin, and customer service" and "lowest where human judgment matters most." It also found that "the biggest barrier isn't cost": the top barriers "cluster around privacy and security, fear of errors, and uncertainty about what AI can do."
Fear of errors is the right instinct aimed at the wrong target. Kept away from the numbers of record, AI is one of the safest helpers in the office. Pointed at them, it produces confident arithmetic that nobody checks. So keep it in the finance office, and be exact about where it stops.
The Money Line
The Money Line is a single boundary that tells everyone in the business what AI may do on its own and what needs a person's signature.
The Money Line is how I suggest a small business decides where AI belongs in finance and admin: AI may work freely on the words around the money, but anything that changes a number of record, moves money, or changes who gets paid must be checked by a person against the source document before it counts.
Above the line sits the work that is made of words:
- Drafting polite, firm reminders for overdue invoices, in your tone, for a person to send.
- Turning a bank export or a shoebox of receipts into suggested categories for someone to approve.
- Explaining in plain English why this month's costs are higher than last month's, from figures that have already been checked.
- Summarizing a supplier contract's payment terms, notice periods and price-review clauses.
- Writing the cover note to your bookkeeper or accountant, with the questions you need answered.
- Turning a long page of tax guidance into a checklist for your situation, which you then confirm against the source.
Below the line sit three things, and only three:
- Numbers of record. Any figure that goes into your accounts, a VAT (value added tax) or sales tax return, a payroll run or a tax filing.
- Money moving. Any payment, refund or transfer.
- Who gets paid. Any new supplier, any change of bank details, any change to who is on the payroll.
AI can still help below the line. It can prepare the payment run, suggest the category, pull together the quarter's figures. What changes is the sign-off: a named person checks the work against the invoice, the bank statement or the contract, and nothing counts until they have. If you are thinking about AI agents that act on your behalf, the same logic is why paying suppliers fails the Errand Test: by the time anyone notices the mistake, the money has left.
The line is drawn around consequences, not tools. The chatbot that drafts a gentle reminder above the line becomes a risk the moment someone pastes its total into a VAT return.
What does the Money Line look like in a real week?
In practice the Money Line turns into a handful of habits, most of which take seconds. Take a hypothetical 15-person heating and plumbing firm: an owner, an office manager who keeps the books three days a week, and an outside accountant.
On Monday the office manager exports the list of overdue invoices and asks AI to draft reminders, gentler for the recent ones and firmer for the old. She reads each draft, removes two where the customer has a dispute open, and sends the rest. Above the line, and an hour's chore becomes a short one.
On Tuesday the engineers' fuel and parts receipts arrive as phone photos. AI suggests a category for each. She approves them one by one in the accounting software, and the approval is where the line sits.
On Wednesday a regular supplier emails to say its bank details have changed. This is the most important moment of the week. She rings the supplier on the number already on file, never the number in the email, before anything is updated.
On Thursday the owner asks why fuel costs jumped. AI explains the movement from the checked figures and suggests three questions to ask the engineers. The owner decides what to do.
On Friday she prepares for the quarterly tax update. AI drafts the checklist and the note to the accountant. The figures themselves come from the accounting software, checked against the bank.
Nothing in that week needed a data team or a new system. It needed one person who knows where the line is.
Why does the "who gets paid" line matter more now?
AI has made a convincing fake invoice cheap to write, so the old tells of a scam, the clumsy wording and the odd layout, are disappearing. The FBI's 2025 Internet Crime Report, released in April 2026, recorded 1,008,597 complaints and $20.877 billion in reported losses. Business email compromise, the scam in which a criminal poses as a supplier or a boss to redirect a payment, accounted for 24,768 complaints and about $3.05 billion in losses. The report counted more than 22,000 complaints that included AI-related information, with adjusted losses above $893 million, and warns that AI "enables the creation of convincing synthetic content."
My reading of those figures: this is an argument for a phone call, not for more software. A perfectly written fake email fails the same test as a clumsy one. Ring the number you already had. In the business teams I work with, the safeguard that holds is rarely a clever filter. It is a plain rule that nobody, including the owner, can change a payee from an email.
The Mistake I See Most Often
The mistake I see most often is a capable person pasting a bank statement or a spreadsheet into a chatbot, asking for the totals, and copying the answer straight into something that matters. The output looks tidy. It is formatted, confident and usually close. Close is the problem, because nobody checks a number that looks finished.
Across the business teams I train, the finance question that comes up first is almost always some version of "can it just do the books?" I have trained more than 150,000 people in over 30 countries, and worked with organizations that range from Google and Grammarly to Welsh Water and Hyve Group, and the answer I give a five-person firm is the same one I would give a large one. Let AI do the reading, sorting and drafting. Keep the counting of record in the software built for it, and keep a person's name on every figure that leaves the building.
There is a quieter version of the same mistake: pasting client or payroll data into a personal AI account because it is faster. Decide which tools are approved for financial data, write it on one page, and say it out loud. If staff are already using tools nobody approved, bring them into the daylight before you set any other rule, and fold the finance rule into your AI policy.
What changes for UK businesses under Making Tax Digital?
Making Tax Digital makes digital records the rule, which makes AI more useful above the line and the Money Line more important below it. According to guidance from HM Revenue and Customs (HMRC), updated on 26 March 2026, sole traders and landlords with qualifying income over £50,000 "should've started using Making Tax Digital for Income Tax from 6 April 2026." Those over £30,000 "will need to use it from 6 April 2027", and those over £20,000 from 6 April 2028. In practice that means digital records and quarterly updates sent through compatible software.
AI does not file anything here. The compatible software does. What AI does well is the work around it: turning the guidance into a checklist, chasing missing receipts every month instead of every January, and drafting the questions for your accountant. The figures that reach HMRC are numbers of record, so they sit below the line. US owners face the same split with quarterly estimated taxes and sales tax returns: let AI prepare the paperwork, and keep a person's check on every figure.
Who should hold the line?
One named person should sign off everything below the line, and it should never be "whoever is around." In most small businesses that is the owner or the bookkeeper. Two habits matter: that person can explain any figure without the AI open, and they know which tools are approved for financial data.
That takes training, and owners know it. In a survey of 561 US small and medium-sized business decision-makers released by Thryv on 25 June 2026, 70% of owners said they needed more training to use AI effectively, and 57% said their primary source of training was YouTube and social media. Informal learning is fine for drafting social posts. Finance is where it gets expensive. If you have a junior on the books, build their role around the checking, not the keying, so the judgment that holds the line keeps growing.
What should you do this week?
You can put the Money Line in place in a week without a consultant or a new tool.
- List the chores. Write down the ten finance and admin tasks that eat the most time each month.
- Draw the line. Mark each one above or below the Money Line. Anything that touches a number of record, a payment or a payee goes below.
- Start above the line. Pick two above-line chores, such as invoice reminders and contract summaries, and use AI on them for a week with a person reading every output before it leaves.
- Write the payee rule. One sentence: any new supplier or changed bank details is confirmed by phone, on a number you already hold, before anything changes. Tell everyone, including yourself.
- Name the signer. Decide who signs off below the line, and who covers when they are away.
- Approve the tools. Say which AI tools may see financial data and which may not.
- Keep an error list. At the end of the week, write down everything the AI got wrong. That list is your training plan.
Where to go from here
If your team is already using AI on the books but nobody has shown them where it stops, that is the work my AI training for business teams is built for: practical sessions on your real jobs, with your own tools and policy. If you would rather see where your business stands first, the free Business AI Readiness Scorecard takes about five minutes and ends in a 90-day plan.
Sources and further reading
- 2026 AI Impact Report: Mapping Adoption, Use, and Impact Across Small to Midsize Businesses, Intuit, 12 May 2026
- Find out if and when you need to use Making Tax Digital for Income Tax, HM Revenue and Customs, GOV.UK, updated 26 March 2026
- 2025 Internet Crime Report, FBI Internet Crime Complaint Center, April 2026
- AI Adoption Continues to Rise, but 70% Say They Need More Training to Use It Effectively, Thryv via Business Wire, 25 June 2026
Dan Fitzpatrick is The AI Educator: a Forbes contributor, international keynote speaker and bestselling author who helps business teams use AI well, safely and every day. More about Dan.
