Why AI Will Not Replace Excel in Finance

On September 14, 2026, Microsoft retired the COPILOT function in Excel. It let a user type an AI prompt straight into a cell, the way you would write a SUM. Results already calculated survive as cached values. A cell that recalculates now returns #NAME?.

It was only ever a preview, so the practical loss is small. But I think the episode shows how finance will split the work between AI and the spreadsheet. The AI moves in beside the formulas. The formulas stay in charge.

I have a stake in that view. I led Excel4apps through its sale to insightsoftware in 2018. Its products brought Oracle and SAP ledger data into finance teams’ own workbooks, so finance did not have to leave Excel to get at its numbers. I am making the same bet now with Finatical.

My argument is short. AI will not replace the spreadsheet in finance. It will make it the place where finance checks and owns what AI produces.

Finance stays in Excel for control, not comfort

I made the case for Excel in a short post in September 2025. I argued from flexibility, familiarity and live links to the books, and wrote that AI-driven analytics tools “complement rather than replace Excel.” I still think so. But familiarity is not why finance stays. Control is, and it comes in two kinds:

• A path you can follow. Every figure is either typed in or calculated, and the calculation is one click away. A reviewer can trace a number to its inputs without asking anyone how it was built.
• A model you can change. The person who owns the model can change an assumption the night before the board meeting, without opening a ticket, and watch every dependent number move.

The flexibility I wrote about in 2025 is really the second kind. In my view, most systems built to replace the spreadsheet give up one of the two. They hand you a result you cannot trace, or one you cannot change on your own deadline. That, not habit, is why I think finance keeps turning them down.

AI moved into the workbook

The AI companies seem to agree. Microsoft announced on September 16, 2024 that Copilot in Excel was generally available. On October 27, 2025, Anthropic released Claude for Excel in beta as a research preview: a sidebar in which Claude can read, modify and create workbooks.

Both brought the assistant to where finance already reviews its work. Neither asked finance to carry that work out of the workbook. Excel4apps took ERP data to the same place years earlier.

Note what Anthropic chose to stress. Claude tracks and explains its changes, and a user can jump straight to the cells it cites. The AI’s work arrives in a form someone can check and edit. Finance has always asked that of a spreadsheet.

This matters because I think review in finance is rarely a yes-or-no call. It is an adjustment. The accrual is close, but the controller knows a contract was amended last week. The forecast holds, except that one region always books its orders late. A sealed answer in a chat leaves two options: accept it or redo it. A workbook with live formulas and visible inputs lets the reviewer fix the one thing that is wrong and keep the rest. Who then puts their name to the result is a separate question, and a harder one.

A finance formula has to give the same answer twice

The COPILOT function arrived in preview in August 2025, and Microsoft’s own support page drew the line itself. For “any task requiring accuracy or reproducibility,” it pointed users to native Excel formulas. It said to avoid AI-generated output in financial reporting. And it warned that the function could return different results on recalculation. Thirteen months later the function was gone. Microsoft gives no reason.

Whatever the reason, the lesson for finance holds. Run the same calculation twice and you get the same number. A cell whose answer can shift on recalculation, or vanish when a vendor withdraws a feature, is not something a controller can rely on at quarter end.

So AI belongs beside the calculation, not inside it. It can build the model, write the formula, explain the variance and flag the balance that does not tie. The arithmetic stays with the formula, which does it the same way every time, in plain view. AI writes the logic. The grid holds it still.

AI makes more copies than finance can control

None of this makes spreadsheets safe. Versions multiply. Links break when a folder moves. Someone pastes a value over a formula and nobody notices for a quarter. Those are the first things I would expect an auditor to ask about: which version, what changed, whether the formulas are intact.

AI makes a workbook fast to build. It does not make one any easier to control. A team that can build a forecast in minutes can have six versions of it by lunch, each slightly different, and no clear answer to which one is real.

I have spent years in software that connects workbooks to the ledger, so I am not neutral here. But the point is wider than any product. The hard question in a finance workbook was never how fast it was built. It was which extract of the ledger it rests on. An assistant working from the wrong copy reaches the wrong conclusion with full confidence, and faster than any analyst. A team needs one workbook it treats as the book of record. AI work that lands anywhere else is a draft until someone brings it there.

The test is whether finance can open the file

For people building finance software, I think the implication is direct. Trust is earned where the reviewer already works. That is the difference between AI built for finance and AI bolted onto it: one lands its work where people can inspect and change it, the other hands down conclusions. A tool that routes around the spreadsheet asks finance to give up both kinds of control. I expect many teams to decline quietly, by pasting the AI’s answer into a workbook of their own. That is one more copy, cut off from its source.

For finance leaders, the test of an AI tool is just as plain. Can the team open what it produced, change it, and explain it to someone who was not in the room? If not, the tool has not replaced a spreadsheet. It has added one, with no formula behind the number.

Chris Meyer is a chartered accountant (SAICA, CIMA) and a partner on Born Capital’s Investment Committee. He joined Excel4apps in 2009 to run its Americas business and became its CEO in 2016.