"Can AI work with my spreadsheets?" Yes, for a good share of the work. Writing and explaining formulas, cleaning and reshaping data, making sense of a file someone else built and drafting a first layout from a plain description are all things a tool does quickly, in Excel, Google Sheets or a plain export. What it cannot tell you is whether the logic is right, whether the totals match the books, or whether this is the file the business should trust. This article is for the owner of a small or mid-sized company who runs pricing, job costing, inventory, payroll prep or the cash forecast on a few files that one person built years ago.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them, so I have opened a lot of inherited spreadsheets and met the person who understood each one. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs.
What a tool does well
Writing and explaining formulas. Describe what you want in plain English and a tool proposes the formula, then tells you in words what it does. Ask it for the reason as well as the formula, because the reason is what you check.
Cleaning and reshaping data. Dates in three formats, duplicate customers, names spelled five ways, two exports that need to be merged into one list. This is slow, dull work for a person and a good fit for a tool.
Explaining a sheet you inherited. Give it a copy and ask it to walk through the file tab by tab: what each tab holds, where the numbers come from and which cells feed which. It is a fast way to learn a file nobody documented.
Building a first version. From a description of what you track, a tool can draft the columns, the layout and the starting formulas. Treat it as a first draft, not a finished model.
Summarizing a table. A short read of what the data shows, with the odd rows pointed out for you to look at.
Turning a monthly job into steps. The month-end routine that lives in one person's head becomes a written list of steps, which is also the first step to handing it to someone else.
What a person must keep
The logic and the assumptions. Which costs are in the margin, how a discount is applied, what counts as a late payment. A tool follows the rule it is given. Someone has to decide whether the rule is the right one.
The totals, tied to the books or the bank. A sheet that adds up is not the same as a sheet that is right. The totals should agree with a source outside the file, and a person does that.
The decision the sheet feeds. A price, a hire, an order, a loan request. Can AI forecast cash flow applies the same discipline to the most important sheet an owner has.
Knowing which file is the real one. Copies named "final", "final v2" and "use this one" multiply fast. One file is the source, and one person says which.
Who is allowed to change it. Edit rights on the file that sets your prices should be a short list.
The risks that are specific to spreadsheets
A formula that looks right and is wrong in one row. A tool can copy a formula down a column and miss one row, or apply it to a row where the layout changes. The sheet looks fine until someone checks that row.
A tool that rewrites the real file. Some tools edit in place. If you let one change the original, you may not be able to see what it changed. Work on a copy.
Ranges that miss new rows. A total that covers rows 2 to 200 quietly leaves out row 201. This is an old spreadsheet problem, and a tool does not remove it unless you ask it to check.
Invented columns or "fixed" numbers. A tool asked to repair a sheet may fill a gap with a plausible value instead of saying the value is missing. What is hallucination explains why this happens. Ask it to list what it changed and why.
Dates and units mixed. Day and month swapped, pounds read as kilograms, dollars mixed with another currency. Check a few rows by eye.
Confidential data in the wrong tool. Payroll, customer names and margins in a sheet you paste into a consumer tool is your data leaving the building. What happens to your data explains what to ask a provider. Use an approved tool on a company account, and remove what the task does not need.
The one-person file. If only one employee understands the pricing sheet, the business has a key-person risk. A tool can explain the file, but it does not make a second person responsible for it.
Records and retention. Some spreadsheets are business records, and some must be kept for a set time in a form an auditor or regulator can read. Overwriting or reorganizing them can break that. Ask your accountant and lawyer which files have rules. Not legal advice.
Starting safely
Work on a copy, never the original. Keep the real file untouched until you have checked the result.
Start with one sheet. Pick a file that matters but is not the one your payroll or your bank depends on.
Ask for the formula and a plain-English reason. If you cannot follow the reason, do not use the formula.
Check the totals against a source by hand. Tie them to the books, a bank statement or a report you already trust. Can AI analyze my financial statements covers that kind of check.
Test with five rows you already know the answer to. If the tool gets those wrong, stop there.
Keep a version log and a named owner. Date, what changed, who checked it. One person owns each important file.
Add a "checked by / date" cell. A simple line at the top of the sheet shows when a person last verified it.
Keep the tool away from payroll and bank data until the rules are written. Write down what may be shared, with which tool, on which account, and then widen it. The same thinking applies to any routine task; can AI do my admin work covers the wider picture.
Picture it going wrong
A distribution business asks an assistant tool to tidy and extend the price and job-costing sheet the operations manager built years ago. The tool cleans the data, adds a column for freight and copies a formula down. One range stops a row short, and a unit in one supplier's list is read as cases when it was single items. The sheet looks cleaner than it ever has. It drives a quote to a large customer and, a few weeks later, a hiring decision. Nobody reconciled it to the books. Nobody was careless. The tool did the task it was given, the sheet looked right, and no person owned the check.
Mistakes that repeat
Letting a tool edit the only copy of a file. Trusting a total because it looks neat. Pasting payroll or customer data into a personal account. Skipping the five-row test. Never naming who owns the file. Treating an explanation of the sheet as proof that the sheet is correct.
Frequently asked questions
Can AI build a spreadsheet for me?
It can draft one from a plain description of what you track: the columns, the layout and starting formulas. Treat it as a first version. A person checks the logic, tests it on rows with known answers and decides what the sheet is used for.
Can AI fix my Excel or Google Sheets formulas?
Often, yes. Paste in the formula or describe the problem, and ask for the fix and the reason. Test the result on a copy and on rows where you know the answer before it goes into the real file.
Can AI clean up and combine my data?
Yes, this is one of the better uses: dates, duplicates, spelling variants and merging two exports. Ask the tool to list what it changed, and check a sample of rows against the original.
Is it safe to upload my spreadsheet to an AI tool?
It depends on the data and the tool. Payroll, customer details and margins deserve extra care. Use an approved tool on a company account, read the provider's terms on how it handles your data, and remove columns the task does not need. Not legal advice; ask your lawyer about any privacy or records rules that apply to you.
Can AI replace the person who built our sheet?
It can explain the file and help someone new learn it. It does not replace the person's knowledge of why the sheet works the way it does. Use the tool to get that knowledge written down while the person is still here.
Should AI build my budget or forecast?
It can build a first version and update it. A person owns the assumptions, ties the numbers to the books and signs off on the result, because the budget or forecast drives real decisions.
Where should I start?
Pick one sheet that matters but is not tied to payroll or the bank. Work on a copy, ask for formulas with reasons, test five rows you know, tie the totals to a source, and name one owner for the file.
Not sure where you stand? Take the 3-minute AI readiness scorecard.