A small business owner used ChatGPT instead of a financial planner

Sara Rojas runs finance for two businesses. Facing a pile of loans and credit cards taken on to open the shop, she listed every balance, every APR and her target, put the lot into ChatGPT, and got back an order of attack: which debts to pay first, which loans to refuse, and where the down payment was too high to be worth it. She reports it did the job she would otherwise have paid a financial planner for.

What she actually asked

The framing came from frustration rather than curiosity. "At one point I'm like, well, what am I doing wrong? Like, how should I change this? Because I feel like I just keep going into debt."

So she broke the whole position down — how much she owed on each line, the APR on each, and the goal she was aiming at — and handed it over in one go. What came back was directional and specific: do not take this loan, the APR is too high; the down payment on that one is too much; pay this first and then work down the line.

She is clear about what the value was. It was not arithmetic she could not do. It was getting a structured plan out of a mess she had been carrying around unstructured. "That really helped me kind of to put my thoughts together and see where I should be putting more in."

It spread to the rest of the finance work

Having worked once, the same pattern took over the other numbers she is responsible for:

• Payroll
• Sales ratios
• How much to spend on food costs
• How to price items in the store

"Instead of, you know, looking for someone to help, I just put it there and within five minutes I had everything, a plan kind of to tackle." The comparison she keeps making is not against doing it herself. It is against the time cost of briefing an outside professional, which for a two-owner business means assembling all the same information anyway and then waiting.

Why this use lands and the proposal use does not

It is worth noticing what these finance tasks have in common with her inbox win and not with her proposal failure. Debt ordering, payroll ratios and menu pricing are all reasoning over information she already holds. She supplies the numbers; the model supplies the structure. Nothing has to be fetched, and nothing has to be assembled into a designed artefact.

Her proposals fail on exactly those two counts — they need live prices from rental suppliers and a laid-out document at the end. Same operator, same tool, same year of accumulated context, opposite outcome. The dividing line is not task difficulty. It is whether the work is thinking or fetching and assembling.

A caveat that belongs on the record

Neither owner presents this as advice, and neither should a reader. What is reported here is what one operator did and what she says it saved her. A payoff order generated from balances and APRs is a calculation, not a professional opinion, and it carries none of the obligations that come with one. Nothing in the interview suggests she checked the output against a second source, and the interview does not establish what the plan cost or saved her in the end.

What it does establish is the shape of the demand. An owner with no finance background, holding real debt and no budget for advice, found that a general-purpose chatbot could turn a shoebox of numbers into a sequence of decisions in five minutes. Whatever else is true, that is the bar any product aimed at her has to clear.

Where this comes from

S2E2: The secret AI strategy this small NYC business uses to save thousands — the full interview with Sara Rojas and Luisa Ramirez. Listen or watch: YouTube, Spotify or Apple Podcasts.