Somewhere in your company there is a spreadsheet nobody planned. It started years ago as one person’s workaround for something the software would not do. Today it prices your jobs, or schedules your crews, or is the thing month-end cannot close without. Nobody decided it would run the business. It just never stopped.
The research on spreadsheet quality is old, consistent, and ignored. Audits of operational spreadsheets — real ones, in use at real companies — find at least one error in roughly 86% to 94% of them. The rate has held steady across decades of studies because its cause is steady: people make mistakes in about 2% to 5% of complex steps no matter how careful they are, and a spreadsheet, unlike a system, has no mechanism for catching them. The question was never whether yours contains an error. It is which cell, and what it costs.
The examples keep coming. In 2023, Norway’s sovereign wealth fund — the world’s largest, managing about $1.5 trillion — lost roughly $92 million because a date was typed wrong in a spreadsheet: December 1 instead of November 1, in the calculation of the benchmark the entire fund is measured against. It was the largest operational error in the fund’s history, at an institution with resources most businesses cannot imagine. If a workaround can quietly become load-bearing there, it can anywhere. A European research group keeps a running catalog of incidents like this; it has been growing for twenty years and shows no sign of slowing.
The errors, though, are the visible risk, and probably not the largest one.
The spreadsheet became infrastructure without a decision. Real systems get backups, a record of who changed what, and some way of noticing when they break. The spreadsheet carries the same responsibility and gets none of that. It breaks silently, and the break is discovered downstream, later, by its consequences.
It also lives in one person’s head. The formulas are in the file; the reasons for them are in the memory of whoever built it. When that person is on vacation, the company operates on faith. When they leave, a piece of how the business actually works leaves with them.
Here is the more useful way to look at it. The spreadsheet is not the problem — it is the evidence. It marks, precisely, the place where the software you bought does not fit the business you run. Every workaround is a small map of that gap, and every number someone re-types between two systems is a boundary where the tools failed to meet. That also makes it the best possible starting point for improvement: the process it encodes is already defined, already running, and already proven to matter. It is simply being executed by hand.
The test worth applying is not “do we use spreadsheets” — everyone does, and for scratch work they are the right tool. The test is: if this particular one were wrong on Tuesday, when would we find out, and what would it have cost by then? For at least one spreadsheet in your company, the honest answer is unsettling. That is the one.