If your business runs on spreadsheets — and most do — you may already have AI tools you haven't touched. Both Microsoft 365 and Google Workspace have been quietly adding AI features to Excel and Google Sheets over the past year or two. No new vendor, no new login, no data migrating to a platform your accountant has never heard of.
That low friction matters. The hardest part of most AI projects is the setup. Here, the setup is "open the file you already have."
But "it's there" and "it's useful" aren't the same thing. These features are genuinely good at four specific tasks — and quietly unreliable at a few others. This piece covers both.
What's Actually Available, and Whether You Have It
Microsoft calls its AI layer Copilot. Inside Excel, it can help you write formulas, clean data, and summarize a table. Google's equivalent lives inside Sheets as Gemini (previously called Duet AI), with similar capabilities.
Whether you have access depends on your plan. Microsoft 365 Business Basic and Standard don't include Copilot by default — it typically requires an add-on or a higher-tier plan. Google's Gemini features are available on some Workspace tiers and not others.
The practical move: open Excel or Sheets, look for an AI or Copilot button in the toolbar or sidebar, and see what happens. If it prompts you to upgrade, you'll know. If it opens, you're in. This isn't a licensing guide — those details change faster than any article can track — but five minutes of clicking will tell you more than a pricing page will.
Both tools also have a workaround: you can paste your formula or your messy data into ChatGPT, Claude, or Gemini (the standalone chat tools) and get the same kind of help, then bring the result back into your spreadsheet. No special plan required. That approach is what most of the examples below describe.
Four Things That Genuinely Work
1. Describing a Calculation in English and Getting the Formula
This is where AI earns its keep in spreadsheets, especially for owners who didn't grow up living in Excel.
Say you want to flag every invoice that's more than 30 days past due and hasn't been marked paid. You know what you want — you just don't know how to write it. Instead of digging through formula documentation, you describe it:
"I have a spreadsheet with invoice dates in column B, due dates in column C, and a paid/unpaid status in column D. Write a formula that shows TRUE if the invoice is unpaid and the due date is more than 30 days ago."
The AI will write you something like =AND(D2="Unpaid", TODAY()-C2>30) and, critically, it will explain what each piece does. That explanation is where the real value is. You're not just getting a formula — you're learning to read it, which means you can spot when it's wrong.
Always ask for the explanation. A formula you don't understand is a formula you can't check.
One important prompt habit: be specific about your column layout and your exact values. If your status column says "Not Paid" instead of "Unpaid," the formula won't work — and the AI has no way to know that unless you tell it.
2. Cleaning Messy Data
Most spreadsheets that have been touched by more than one person have at least one of these problems:
- Names formatted inconsistently — "Smith, John" in some rows, "John Smith" in others
- Dates typed as text — so your sort order is alphabetical instead of chronological
- Freeform fields — a column that was supposed to say "Yes" or "No" but now has "yes," "Y," "NO," "n/a," and "—"
- Duplicate entries that don't look like duplicates because of a trailing space or a different capitalization
AI can help with all of these, though the workflow varies. For structural fixes — splitting a full name into first and last name columns — you can describe what you want and get a formula:
"Column A has full names in the format 'Last, First'. Give me formulas to put the first name in column B and the last name in column C."
For spotting the rows where a date looks like a date but is actually stored as text (which causes sorting and calculation problems), you can ask:
"Write a formula that checks whether the value in column B is a real date or just text that looks like a date."
For auditing an inconsistent dropdown column, paste a sample of the values and ask the AI to list the unique variants — then you'll know exactly what you're standardizing.
3. Explaining a Spreadsheet You Inherited
This is underrated. If you've ever taken over from a bookkeeper, a controller, or a previous owner and found yourself looking at a formula like:
=IFERROR(INDEX($B$2:$B$500,MATCH(1,(A2=$C$2:$C$500)*(TODAY()>$D$2:$D$500),0)),"")
…you know the feeling. It runs. It produces a number. You have no idea if it's right.
Copy that formula, paste it into an AI chat, and ask: "Explain what this formula does, step by step, in plain English."
You'll get a readable breakdown — what each function does, what data it's pulling from, what the IFERROR wrapper is there to prevent. From there, you can ask follow-up questions: "What would happen if the date in column D were blank?" "Is there a simpler way to do this?"
This is one of the most immediately useful things AI can do for a non-technical owner. Your previous bookkeeper's logic is no longer a black box.
This kind of formula-literacy also matters when you're getting your data ready for AI work downstream — understanding what your spreadsheets actually calculate is a prerequisite for trusting what you feed into any AI system.
4. Summarizing or Charting a Table Without Hunting Through Menus
If you have the built-in Copilot or Gemini integration running inside your spreadsheet, you can highlight a table of data and ask questions about it in plain English:
"What were my top five products by revenue last quarter?"
"Show me which months had expenses more than 20% above the average."
"Create a bar chart of monthly sales by region."
For basic summaries and quick charts, this is faster than building a pivot table from scratch — particularly if you don't build pivot tables often enough to remember the steps. The AI handles the menu navigation; you just describe what you want to see.
For owners who manage their own books or review financial reports weekly, this is a genuine time-saver on routine reporting. It pairs naturally with the broader category of AI for back-office tasks like bookkeeping and admin, where the goal is getting routine information out of the way faster so you can focus on the decisions.
Where AI Gets It Wrong — And How to Stay Ahead of It
None of the above means you should trust the output without checking. Here are the failure modes worth knowing:
It will produce a formula that runs and returns the wrong number. This is the most dangerous outcome because it doesn't look like an error. The cell doesn't show #VALUE! or #REF! — it shows a number, confidently. The formula might be logically correct for a slightly different question than the one you asked, or it might handle edge cases (blanks, zeros, dates before a certain year) in a way that quietly skews your totals.
The fix: always test against a handful of rows you can verify by hand. If you're writing a formula to sum unpaid invoices, manually add up three or four of them and make sure the formula agrees.
It doesn't know your business. The AI has no idea that "MISC-2" is your owner draws account, that your fiscal year ends in March, or that a negative value in column F means a return rather than a credit. Every assumption that lives in your head — or in your previous bookkeeper's head — is invisible to the model unless you state it explicitly. If your data has any business-specific conventions, name them in the prompt.
It's unreliable on large, complex sheets. The more tabs, cross-references, and named ranges a workbook has, the more likely the AI is to make an assumption that breaks something. For a simple table, it's excellent. For a 12-tab financial model with circular references, treat it as a starting point, not a solution.
The built-in tools are less capable than the standalone chat tools for complex work. Copilot inside Excel is convenient for quick tasks; for a genuinely tricky formula or a nuanced data-cleaning problem, you'll often get a better result from ChatGPT or Claude, where you can have a back-and-forth conversation, paste in sample data, and iterate. Learning how to write better AI prompts pays off in spreadsheet work just as much as anywhere else.
The Governing Rule
Use AI to draft and explain. Use your own judgment to decide and verify.
That means: let it write the first version of the formula, then check it. Let it explain the inherited logic, then confirm the output against something you can verify independently. Let it generate the chart, then make sure the underlying table is correct before you share the chart with anyone.
The moment to be most careful is when the output looks right and you don't have time to check. That's exactly when you should slow down, because a confident-looking wrong number in a spreadsheet has a way of traveling — into reports, into decisions, into conversations with your accountant or your lender.
A Quick Reference: What to Ask
Here are four prompts you can adapt right now. Adjust the column references to match your actual sheet.
Formula from a description:
"Column A has invoice amounts. Column B has the due date. Column C says 'Paid' or 'Unpaid'. Write a formula for column D that calculates how many days overdue each unpaid invoice is. Show the result as a positive number, and leave column D blank for paid invoices."
Explain an inherited formula:
"Explain this formula step by step in plain English: [paste formula here]. Tell me what it's calculating and what could cause it to produce the wrong result."
Clean inconsistent data:
"Here are the values in my 'Status' column: [paste a sample]. List the unique variants, and give me a formula that standardizes all of them to either 'Active' or 'Inactive'."
Spot text-formatted dates:
"Write a formula that returns TRUE if the value in cell B2 is stored as text rather than as a real date, so I can find and fix those rows."
The AI features in Excel and Google Sheets are the lowest-friction AI available to most small businesses. They don't require a new vendor, a new budget line, or a technical implementation. What they require is knowing which four things to use them for — and knowing when to check the work.
If you'd like help thinking through which AI tools are actually worth your time across your whole operation, let's talk. A short strategy call is often enough to find the two or three places where the effort-to-return ratio actually makes sense for a business like yours.