Three AI Checks That Replace Three Expensive Money Guesses

It's the last week of the month and there's a bank export open on your screen. Rows and rows of transactions, no categories, and someone who charges by the hour waiting on it.

Here are three checks you can run this week with ChatGPT or Claude: the categories, the months ahead, and the copy you're about to publish.

Three AI checks can replace three expensive money guesses in a small business: sorting a bank export into categories, forecasting which month will be tight on cash, and flagging an ad claim before it becomes a legal problem. Each one runs in ChatGPT or Claude this week, for free, and the decisions get easier once the numbers are sorted.

Play 1: Sort Your Bookkeeping With ChatGPT or Claude

The problem: The categories are the slow part. You know what every line is; you just have to say so, line after line.

The fix:

1. Export your transactions from your bank or accounting app (.csv or .xlsx).
2. Delete any column with your account or routing number, before you upload.
3. Go to ChatGPT or Claude: both have free tiers.
4. Upload the cleaned file and type: "Sort these transactions into categories and give me a total for each one."

A summary your accountant can read, in about ten minutes.

Settle one thing before the AI ever sees your file: which accounts the export covers. This is the same root cause behind why your store, Meta and your processor each report a different total.

Money moves through more places than most owners pull: a second payment processor, a card used only for software, a bank account that exists to pay one annual bill. A tidy summary of two-thirds of your money is worse than none, because it reads as finished. List your accounts by hand and export every one.

Say refunds out loud in the prompt too. A charge that came in and went straight back out nets to nothing, and left alone it reads as a sale. Add the line: "Treat refunds and chargebacks as negatives and show me the net per category."

Then read the output like a reviewer. The AI guesses on anything ambiguous: a hardware store run could be a repair or an asset, a restaurant charge could be a client meal or your lunch, a transfer between your own accounts is not income at all.

So add: "Put anything you weren't sure about in a separate needs-review list." Rule on those yourself before the file goes anywhere.

Safety note: Delete the account number before you upload. Full stop.

Play 2: Build a Cash-Flow Forecast With AI

The problem: "Can I afford to hire in August?" and "which month will be tight?" are the questions you answer at 11pm with the calculator app and a feeling.

A forecast extends what already happened, so it's blind to anything without a pattern yet. A subscription that started this month, a contractor you just brought on, a usage-based bill that rises with traffic instead of sales: none of that is sitting in three months of history.

My worst cash surprise didn't come from a slow month. In July my Anthropic bill kept climbing, about $45 a day, so I went looking. I blamed the chatbot on my intake page first and hardened it in a day.

The real spender was the AI back office I run this company with, running on a metered API key instead of the subscription I was already paying for. Nothing was down, nothing errored, nobody complained. The only symptom was a number on a bill I read once a month.

A six-month forecast built on my previous three months would have missed it entirely: a forecast only extends what already happened, and this was a brand-new line item. Write down every recurring charge that started in the last 60 days and hand the AI those separately, as knowns.

A forecast is good at the squeeze already sitting in your own numbers. Poor cash-flow management, or a poor understanding of the cash flow you have, is tied to 82% of small business failures, per research by Jessie Hagen, formerly of U.S. Bank, cited by Preferred CFO. Which month runs thin is knowable in advance.

The fix:

1. Put your last three months of income and expenses in a spreadsheet: money in, money out, per month.
2. Delete any account numbers, then upload the file to ChatGPT or Claude.
3. Ask: "Based on these patterns, give me a 6-month cash flow forecast."
4. Then ask what you want to know: "Can I afford to hire in August?"

It'll point at the months your own pattern says will be tight.

Safety note: Remove account numbers first. Same rule as Play 1.

Play 3: Flag Risky Ad Claims Before You Publish

The problem: You wrote something good about your own work, and one word in it is doing more work than your evidence can support. "Best." "Guaranteed." "Proven."

FTC truth-in-advertising rules require every advertising claim to be truthful and backed by evidence before it runs. "Number one," "guaranteed," an implied health result: each needs proof you could produce if asked.

The claim includes whatever you point at as proof. A screenshot, a demo link, a testimonial, a before-and-after photo: all of it is part of the ad. If the screenshot came out of a demo store with sample data in it, the caption says so.

If the demo link sits behind a login, the reader gets a sign-in screen instead of your best work, and a link checker will call that page fine. Open every proof link yourself, in a private window, before the page goes live.

The fix:

1. Paste the copy into ChatGPT or Claude before you publish it.
2. Ask: "Flag anything in this copy that looks like an FTC advertising risk."
3. Read what it flags and swap the vague superlatives for specifics you can back up.
4. Ask a second question: "For each claim you flagged, what evidence would I need to hold to keep it?" That answer is your checklist against your own files.

The specific version is quieter and much easier to defend: "best pet portraits" becomes "hand-finished on 340gsm paper." If the specific version reads weaker to you, that tells you something about your evidence.

AI can't give you legal advice, and it doesn't know what evidence you hold. What it catches is the superlative you stopped seeing on your own page. For a launch or a health claim, get a lawyer.

What to watch for: "Best," "guaranteed," "clinically proven," "number one." Also anything implying you fixed a health or safety problem without proof.

These three run once, this week, in a chat window. For checks you'd rather have running every Monday without you, an AI agent can watch your store instead.

Quick Recap

• Strip the account number, upload the export to ChatGPT or Claude, and ask for a sorted summary.
• Check that the export covers every account your money moves through, including any nobody wrote down.
• Three months of income and expenses is enough for a 6-month forecast, and a brand-new expense won't be in it.
• 82% of small business failures trace back to cash flow. Find your thin month while it's still months away.
• Ask ChatGPT or Claude to flag FTC risks in a big claim before publishing. High-stakes still needs a lawyer.

Start Here

If you know which of the three you'd run first, run it this week. If not, that's worth finding out.

At daisyguti.ai/work-with-me, there's a short intake: about nine questions, a few minutes. Daisy reads every submission herself, assesses whether checks like these are a fit for your business, and replies with next steps. She's a 20+ year engineer who builds these systems for small business owners.

Sources

1. Preferred CFO - Cash flow failure statistics (Jessie Hagen / U.S. Bank research): https://preferredcfo.com/insights/cash-flow-reasons-small-businesses-fail-2026
2. FTC - Advertising FAQ's: A Guide for Small Business: https://www.ftc.gov/business-guidance/resources/advertising-faqs-guide-small-business
3. FTC - Advertising and Marketing (overview): https://www.ftc.gov/business-guidance/advertising-marketing
4. OpenAI - ChatGPT Free Tier FAQ: https://help.openai.com/en/articles/9275245-chatgpt-free-tier-faq
5. Claude plans and pricing (free tier listed at $0): https://claude.com/pricing