ChatGPT in Excel and Sheets: an NZ ops guide

For most NZ small businesses, the real operating system is not the accounting package or the CRM. It is a spreadsheet. The quoting calculator someone built three years ago. The cash-flow tab the owner checks every Friday. The job-tracker that lives or dies on one person remembering to update it. Spreadsheets are where the business actually runs, and they are usually held together by one or two people who know where the formulas are buried.

So when OpenAI put ChatGPT directly inside Excel and Google Sheets, it landed somewhere that touches almost every reader — not a niche developer tool, but the thing your ops manager opens forty times a day. This is a hands-on look at what it does, two tasks worth trying this week, where it falls over, and the one rule that keeps you out of trouble.

What shipped, and who actually gets it

ChatGPT for Excel and Google Sheets came out of beta and went generally available on 5 May 2026, powered by GPT‑5.5. Instead of copying data out to chat.com and pasting answers back, you talk to ChatGPT inside the spreadsheet: describe what you want in plain English and it builds the sheet, writes the formulas, cleans the data, or explains what an existing formula is doing — without you leaving the file.

Who gets it, based on OpenAI's announcement:

• Free and Go — included with limited usage.
• Plus and Pro — included, subject to your plan's agentic usage limit.
• Business, Enterprise, Edu and K‑12 — a free preview that runs through 2 June 2026, after which usage follows each plan's credits and terms.

It works in Excel (desktop and web) and in Google Sheets. The headline is real: building a working sheet from a sentence is genuinely faster than starting from a blank grid. The detail that matters for an NZ business is everything after the headline — whether you can trust the output, and what you are allowed to put into it.

Task 1: turn a quoting spreadsheet into a working sheet

Start with the spreadsheet that costs you money when it is wrong: the quote calculator.

Open a blank sheet and describe it the way you would brief a new staff member. Something like: "Build a quoting calculator. Inputs: labour hours, hourly rate, materials cost, and a margin percentage. Calculate a subtotal, add 15% GST, and show a total. Put the inputs in a coloured block at the top and lock the formula cells." ChatGPT will lay out the labels, wire the arithmetic, and add the GST line — the kind of build that used to take twenty minutes of fiddling with cell references.

The same approach extends to a cash-flow view: "Add a tab that lists expected income and expenses by week for the next eight weeks and shows a running balance." It will scaffold the structure and the running-total formula for you.

Here is the part to internalise before you rely on it: the model builds the structure, you own the numbers. It does not know your real margin, your GST registration status, or your actual rates — so check the GST line computes 15% on the right base, confirm the margin is applied where you meant it, and put two known jobs through it and see if the totals match what you would have quoted by hand. The build is the time-saver. The verification is non-negotiable, exactly as it is with ChatGPT used anywhere else in the business.

Task 2: clean and summarise a messy export

The second task is the one almost every ops person has waiting in a downloads folder: a messy CSV export. Product names typed four different ways, dates as text, addresses with stray spaces, a column that is half numbers and half "N/A".

Paste it in — or point ChatGPT at the sheet — and ask in plain language: "Standardise the product names to title case, convert the date column to a proper date, strip leading and trailing spaces, and flag any row missing a price." Then: "Summarise total sales by product and by month." For a few hundred rows headed into MYOB, a CRM, or a board pack, this is faster than writing the formulas yourself and cheaper than handing it to a developer.

Two cautions that apply specifically to NZ data, both worth checking on every run:

• Dates. New Zealand writes dates day/month/year. Tools that default to the US month/day/year order can silently read 4/5/2026 as 4 May or 5 April. After any date conversion, spot-check a row where the day is 13 or higher — if it converted cleanly, the order was read correctly; if it errored or flipped, it did not.
• De-duplication and totals. When you ask for a summary, confirm it counted every row. Ask it to show the row count it worked from and compare that to the source. A summary that quietly dropped 12 rows looks just as tidy as a correct one.

Where it falls over

OpenAI's launch announcement does not publish a list of limitations, so treat the failure modes below as known behaviour of large language models writing formulas — things to verify in your own copy, not problems someone else has already caught for you.

Complex or chained formulas. Asking for a single GST line or a SUMIF is well within range. Asking it to build a multi-sheet model with nested lookups, array formulas, and conditional logic that all reference each other is where the risk climbs. The output will look complete and confident whether or not the logic is right — and a wrong formula in a model is harder to spot than a wrong number on a page. The more interdependent the formulas, the more you need to test the edges, not just glance at the result.

Plausible-but-wrong, with no ground truth. This is the same failure that shows up everywhere ChatGPT is used: it will produce something that reads correctly even when it is not, because it has no access to the real answer. In a spreadsheet that means a total that is formatted perfectly and computed on the wrong range. The fix is the discipline from Task 1 — feed it inputs where you already know the output.

NZ formats it can get wrong. Dates (covered above) and currency. If your raw data mixes $1,200, 1200, and 1,200.00, check the cleaned column is genuinely numeric and not text that merely looks like a number — text columns break every downstream SUM. Confirm GST is calculated at 15% and applied to the figures you intended, not bundled into a rate it assumed.

None of this makes the feature a gimmick. It makes it a fast first-drafter that needs a checker — which, for an ops manager who already checks the load-bearing spreadsheet by habit, is a fair trade.

The data rule: what's safe to paste, by plan

This is the part most teams skip, and the part that carries real exposure. The moment you paste a spreadsheet into ChatGPT, you have sent it to a US-hosted service. Under the Privacy Act 2020, if that sheet contains personal information — customer names, emails, addresses, anything that identifies a person — you have made a cross-border disclosure, and you remain accountable for it. Your plan changes how the data is handled; it does not change the fact that you are the one accountable.

The plan line that matters:

• Free, Go, Plus, Pro (consumer plans). By OpenAI's stated policy, conversations on these plans can be used to improve the models unless you turn that off in Data Controls. So before any business data goes in, that setting should be off — and even then, treat customer personal information as off-limits.
• Business, Enterprise, Edu. OpenAI states it does not train on business data by default. That is a better footing for internal data, but it is still offshore processing, so the Privacy Act cross-border obligation still applies.

A workable one-page rule for the team, the same shape I recommend for ChatGPT generally:

• Fine to paste: structural and anonymous data — a quoting template with no customer attached, product lists, made-up sample rows, a CSV with names and emails stripped out.
• Not without a deliberate decision: anything with a real person in it — customer lists, contact details, anything you would not email to a stranger.

The fastest way to get this wrong is to let everyone work it out individually. The fastest way to get it right is one short policy and a team that has actually been walked through it — which is the core of how I run AI team training for NZ businesses.

Your ChatGPT plan changes how your data is handled. It does not change the fact that, under the Privacy Act, you are the one accountable for it.

When this beats automation, and when it's a band-aid

ChatGPT in your spreadsheet is excellent for the variable, one-off, judgement-heavy work: building a new calculator, cleaning a particular export, restructuring a tab before a board meeting. Different shape every time, a human reading the result. That is exactly where it earns its place.

It is the wrong tool the moment the work repeats. If someone is pasting the same prompt to clean the same weekly export, or rebuilding the same report every Monday, that is not a job for a person at a chat box — it is a job for an automation that runs on a schedule with a human approval step at the end. Dressing up a recurring process as "we use AI for it now" is a band-aid; the spreadsheet is still load-bearing and still depends on one person remembering to run it. That is the line I draw with clients between AI automation and a chat tool.

And there is a ceiling. When the spreadsheet has quietly become the system — the quoting engine the whole business runs through, the job-tracker three people depend on, the cash-flow model the bank wants to see — no amount of in-cell AI fixes the fact that critical operations are living in a file with no audit trail, no permissions, and one keeper. That is the point to move it off the spreadsheet entirely and into a custom web app built for the job.

So: use ChatGPT in Excel and Sheets for what it is genuinely good at — fast, checked, one-off spreadsheet work — and treat it as a signal. The tasks you keep reaching for it to repeat are the ones worth automating properly. For more on what to tackle first, the rest of the insights library works through it, and if you want a second opinion on where your spreadsheets stop being a tool and start being a risk, that is the kind of thing I help NZ businesses sort out.