AI in Excel and Google Sheets: what changes in your office

Most small businesses have one spreadsheet holding half the operation together: the orders sheet, the customer list, the quotes tracker, or the one someone built four years ago that nobody dares to touch. And for a few months now, that spreadsheet has come with an AI assistant built in. Open Excel or Google Sheets and the button is right there.

This piece is for small business owners whose working day runs through Excel or Sheets and who are wondering whether this actually changes anything. By the end you'll know which office tasks get much faster with an assistant inside the sheet, which ones you should keep doing the old way, and what ground rules to set before everyone on your team starts using it however they like.

What's actually inside your spreadsheet now

Leaving the marketing aside, there are two things on the market that matter to you.

Copilot in Excel. A side pane you talk to in plain English. According to Microsoft's FAQ, it has three modes: one that only chats, one that proposes a plan before touching anything, and one that edits the workbook directly. It can sort, filter, highlight, generate and transform data, look things up on the web and search your work files for context. You need a Microsoft 365 subscription that includes it.

The AI function in Google Sheets. You type something like =AI("classify this comment as complaint, question or order", B2) into a cell and Gemini fills it in. Per Google's help page, it can generate text, summarise, classify and pull in information from Google Search. The limits are clear: it returns text only, generates at most 350 cells per selection, and there are daily usage caps. It also depends on your Workspace plan and on what your admin has switched on.

Both vendors say the same thing in the small print: the assistant can get things wrong and you need to check its work, especially on financial or legal matters. Hold on to that line, because everything else follows from it.

The tasks that do change

Where these assistants really earn their keep is text work that used to be done by eye, row by row. Five very ordinary office examples:

1. Sorting free text. The "reason for return" or "customer comments" column nobody analyses because everyone writes whatever they like. Asking it to tag three hundred rows into four categories takes minutes. It used to take an afternoon.
2. Cleaning a contact list. Names in capitals, phone numbers with and without the country code, the same company spelt three ways. The assistant suggests a tidy version and you check it.
3. Summarising so you can decide. "Tell me the three most common complaints this quarter and which products they're about." A first read that nobody used to do for lack of time.
4. Writing the formula you don't know how to write. An XLOOKUP, a pivot table, a conditional format. You describe what you want and it builds it. For anyone who isn't an Excel whizz, this is probably the biggest time saver.
5. Making sense of an inherited sheet. "Explain what the Margins tab calculates and where it gets its data." Invaluable when the person who built it has moved on.

Notice what they have in common: they're tasks where a nearly-right answer already saves you work, and where you can see at a glance if it's gone wrong.

The tasks that don't change (and shouldn't)

This is the part the adverts leave out.

Exact numbers are still formulas. The till reconciliation, the quarter's VAT, commissions or margins aren't questions you ask the assistant to answer. A language model doesn't add up; it writes something that looks like a sum. The right move is to let it help you write the formula, and then let the formula do the calculating, the same way every time.

Anything that has to come out the same twice. If you classify the comments today and ask again tomorrow, a few rows may land in a different bucket. Fine for getting a feel for things. Not fine for a report someone will compare month on month.

A messy sheet stays a messy sheet. If your order process lives in a sheet five people edit, with columns everyone reads differently, the assistant will help you type faster inside the mess. It won't fix it. That's a process problem, not a tooling one.

Decisions are yours. "Which customer should I stop serving?" is a question it can help you prepare for. Not one it should answer.

The lesson of a function that no longer exists

A recent case sums it up nicely. Microsoft trialled an in-cell =COPILOT() function, much like Google's. It didn't last: according to its own support page, as of 14 September 2026 it's no longer available. Workbooks keep the last calculated value, but as soon as the cell recalculates it returns a #NAME? error. Microsoft now points people to the side pane instead.

If someone on your team built a quoting template on top of that function, they woke up one day to broken cells. Not a disaster, but a useful warning: don't build a process your business depends on around an AI feature its vendor still treats as experimental. Use it to produce a result, check the result and paste the values. Let the sheet store the answer, not the question.

Four ground rules before your team dives in

The real risk isn't the tool; it's everyone using it on their own terms. Before that happens, write down four things. They fit on half a page.

• Which sheets stay away from AI. Anything with health data, payroll, or personal details you have no reason to send anywhere. If a sheet doesn't need the assistant, the assistant doesn't touch it.
• Work on a copy or a new column. Never on the original column. If something goes wrong, you delete it and start again with nothing lost.
• Check a sample before trusting the output. Twenty random rows out of the three hundred it classified. If two or three are off, fix the instruction before carrying on.
• Mind shared files. Microsoft spells it out: once saved, the assistant's changes are visible to anyone with access to the file. Whoever runs AI on a shared sheet needs to know that.

And one more, just common sense: decide who actually gets a licence. Paying for the whole team "just in case" adds up fast in a small business. Start with the two people who spend most time in the sheet and measure for a month.

When it isn't worth it

If your biggest sheet has a hundred rows and you open it once a week, the assistant will save you little and cost you a licence. If the underlying problem is that a whole process lives in a sheet that breaks whenever someone's off, this isn't the fix either: that process needs to move out of the spreadsheet. And if nobody on your team will check what the AI does, hold off switching it on.

For everyone else, which is most people, the takeaway is simple: AI in the spreadsheet saves real hours on anything that's text and on anything that's "I don't know how to do this in Excel". For money and exact figures, same as ever: formulas, checked by a person.

---

If your team is already using these tools each in their own way and you'd like to bring some order to it — what gets used, with which data, and who checks — that's exactly what we do in AI for teams.