When a spreadsheet outgrows the decision it was built for
The spreadsheet that runs your daily replenishment isn't the problem — until ten people share it, it forgets everything overnight, and the same order goes out twice. Here's when a file stops being enough.
Almost every retailer runs a critical daily judgement on a spreadsheet somewhere. Replenishment is the classic one: which stock goes to which store, before it runs out, without burying anyone in inventory they cannot sell. The logic inside that workbook is often excellent — years of accumulated merchandising sense, encoded formula by formula. The problem is rarely the logic. It is the container.
A spreadsheet has three limits that only show up at scale. First, it is single-threaded: it processes one category at a time and one person can really own it at a time. Second, it has no safe way to be shared — when ten planners work from the same static file, two can act on the same line and nobody sees it until double the stock arrives at the loading dock. Third, and most quietly damaging, it has no memory: every morning starts from a fresh export, so a problem that has recurred every day for a fortnight looks identical to one appearing for the first time.
That last point is the real cost. Without memory, you cannot tell a one-off from a chronic issue — and a line that keeps coming back usually is not a daily decision to remake; it is a signal that a setting upstream is wrong. You also cannot see whether the backlog of exceptions is growing or shrinking, which means you cannot tell if the work is getting easier.
The instinct is to add AI. We would argue the opposite comes first: move the same trusted logic somewhere a team can work in it together and it can remember. Keep the recommendation engine deterministic — the rules decide, auditable and repeatable — and let AI do the part it is genuinely good at: explaining a number in plain language and surfacing what has changed since yesterday. Rules for decisions; AI for language and analysis. That division is what makes automation safe to put in front of a team whose trust you cannot afford to lose.
The upgrade path does not mean throwing away what works. It means proving a new system reproduces the old one line-for-line first, then giving it the things a file never could: a shared live view, a guard against duplicate orders, and a memory of what is chronic versus new.
Read the full story
From a spreadsheet everyone shared to a system that remembers