A pre-flight checklist on a stand beside a small plane on an airport apron at night

Automation

Ask the Chatbot for the Same Monthly Report Twice

11 November 2026 · Yener Adal

It's the third working day of the month. Someone in finance exports last month's debtors from the practice system, opens a chatbot in a browser tab, pastes in 400 rows and types: "Group these by partner, flag anything over 90 days, and write me a summary for Thursday's partners' meeting."

11 seconds later there's a tidy table and three neat paragraphs. It goes into the board pack by lunch.

That's a great 11 seconds. I'd be pleased with it too.

And when the budget's tight, "just use the AI for that" is the cheapest-sounding sentence in the room (and it's a sentence I'd happily say about plenty of jobs). The chatbot is a monthly subscription the firm already pays. Automating the report means a quote, and nobody wants to sign a quote during a spending freeze.

There isn't a client meeting I have these days where someone doesn't say, "Oh, I just put that into Claude to get the output." The first thing that goes through my head is: what happens when you need to create that output again?

I've heard it from lawyers, administration staff and finance experts. I've even heard it from my own team. Hell, I've even said it myself.

For a one-off job, I think the chatbot is the right call. For a report that has to come out the same way every month, it hasn't automated anything. Someone still runs every step, every month. And one of those steps now gives a slightly different answer depending on the day.

Two firms, one monthly report

These two are made up. Same size, 25 people, an accounting practice each, both in Melbourne. Both produce a monthly debtors and work-in-progress report for the partners. Both got the same message from the partners in October: hold off on anything discretionary until the new year.

The first firm gives the report to the chatbot, the way the scene above describes. It costs nothing extra.

The second firm spends a few days having the report built once. A connection pulls the export, a set of formulas does the grouping and the ageing, and the same table comes out the same way every month. (Some of those formulas got written with a chatbot's help, which is a perfectly good use for one.) It costs a small invoice, approved after an argument.

In month one, the reports look identical. The first firm looks smarter. It paid nothing and got the same table.

Month three: someone's on leave

The prompt in the first firm lives in one person's chat history. She's away for two weeks.

Her colleague rewrites the request from memory. The chatbot groups by client manager this time instead of by partner, because that's what "group by manager" sounds like to it. The totals are right, but the table is different. Nobody notices until the partner who chases debtors asks why his name has disappeared from his own report.

Pilots read the same printed checklist before every flight, including the 10,000th one, because memory is the thing that changes on a bad day. A prompt sitting in one person's chat history is a checklist kept in one pilot's head.

The second firm's report ran while she was away. Nobody had to remember anything.

Month five: the export changes

The practice software has an update. A column gets renamed and a new one appears next to it.

In the second firm, the automated report breaks on the first run. Loudly. Someone gets an error, logs a ticket, and it takes an hour to fix. That hour costs money. Automation needs this kind of maintenance every time the systems underneath it change.

In the first firm, the chatbot copes. It looks at the new columns and makes a sensible guess about which one means "days outstanding". Most months it guesses right.

A loud failure gets fixed the same morning, but a quiet guess goes into the board pack.

Month six: a partner asks how the number was made

The debtors total in the report is $14,200 higher than the ledger. A partner wants to know why before the meeting.

The second firm opens the steps. Export, formulas, table. It finds the problem in 10 minutes: an invoice that was credited and reissued on the last day of the month.

The first firm pastes the same export in again, in a new chat, and gets a total that matches the ledger this time. The wording is different too. Now there are two totals and nobody can show their working, because the working happened somewhere nobody can see.

You can test this yourself this week. Paste the same file into two new chats with the same request and compare what comes back, line by line. The Australian privacy regulator's own guidance on commercially available AI products puts it flatly: "AI systems are known to produce inaccurate or false results."

So the 11 seconds come with a second job, and it isn't a small one. Someone has to reconcile the chatbot's table against the ledger every month, which is checking the automated version needed once, when it was built and tested. The National AI Centre's most recent survey of Australian SMEs, covering December 2025 to February 2026, found that only around half of businesses using AI check its outputs before they affect customers.

That's only half. And in this case the customer is your own partners' meeting.

The survey asks about outputs that reach customers, so it doesn't measure internal reports, and it's self-reported. I'd still be surprised if internal reports got checked more often than the ones your clients see. Would you bet the partners' meeting on it?

And every month, the client list goes into a browser tab

A debtors report is a list of client names next to how much each one owes you. That's the document the first firm pastes into a chatbot 12 times a year.

The OAIC's guidance says that as a matter of best practice, organisations shouldn't enter personal information, "and particularly sensitive information," into publicly available generative AI tools. A business plan with training switched off lowers the risk. We wrote in our piece on whether ChatGPT is safe that this setting is off by default on the Team and Enterprise plans, and that on its own it won't stop a leak.

For accountants there's a second layer on top of privacy law. The Tax Practitioners Board's Code has an item on not disclosing a client's affairs to a third party without permission, and we walked through what the TPB has said about AI and that Code item last month.

The second firm's numbers never leave its own systems. The report reads the export where it sits.

Beware: "We only paste in the numbers" usually means client names, invoice numbers and amounts. Check what's in the export before you decide it's harmless.

Where report automation is worth paying for

Report automation isn't free, and not every report earns it. You'll hear plenty of advice this year that AI makes automation projects unnecessary. For recurring reports I think it's the other way round: the chatbot is at its best helping you build the automation, once, and then getting out of the way. A quarterly summary that takes half an hour is a chatbot job, or a person-with-a-spreadsheet job, and spending money to automate it would be silly.

The maths that decides it is simple: how often it runs, how long it takes and how many people touch it. A monthly report that takes one person three hours is 36 hours a year, before anyone checks it. A weekly one is well over a hundred.

That's what our Automation ROI Calculator asks. Six questions about one repeated task, no salaries entered, and the result shows on screen without an email. If the task comes in under about 25 hours a year, it tells you not to bother with a project. I'd rather it said that than sold you one. It's our calculator, after all, so you're entitled to check its working too.

Tip: Before anyone automates anything, write the current steps on one page, prompt included. If nobody can write them down, the report already depends on one person's memory.

We've been building small automations for ourselves for years, starting with a chat bot that books our desks and relays monitoring alerts. The lesson from those hasn't changed. The ones that last are boring and written down, and each one has a name next to it.

And if the answer this year is "not during the freeze", that's a decision your business is entitled to make. Plenty of good projects spend a year on the back burner. Write it down with the risk next to it, and put a date on when you'll look again. A board can live with a decision it can see, but it struggles with a report that quietly changes shape every month.

If you'd like a second opinion on which side of that line your monthly report sits, send us how long it takes and how often it runs. We'll tell you, in writing, whether it's a chatbot job or an automation job, before anyone quotes anything.

Yener

Where these numbers come from

About the author

Yener Adal runs Intuitive IT, a Melbourne managed IT and cybersecurity provider working with accounting practices, wealth and fund managers, law firms and recruitment agencies.

Start with a 30-minute AI risk conversation

We’ll ask what your staff are using, what your policy says, and what’s worrying the partners. You’ll leave with a clear view of where your biggest AI risks sit, even if we don’t end up working together.

Book a time

Get a confidential consultation with Intuitive IT

Book a confidential conversation about your firm’s AI risk.

Get in touch

Acknowledgement to Indigenous Australia

The Aboriginal flag and the Torres Strait Islander flag

Intuitive IT acknowledges the Traditional Custodians of this land, the Wurundjeri people and recognises their ongoing connection to land, waters and community. We pay our respect to the Elders past, present and emerging.