Versions break
Email, cloud drive, and desktop each hold a copy. The meeting never uses the latest one, and nobody can see who changed a hidden formula.
Friday night, an assistant is still merging an ERP export, the finance ledger, and store WeChat sheets into “Weekly_final_v7_really_final.xlsx.” Monday the owner opens it and finds a hidden column changed, and returns entered as positives. That is not an attitude problem. The tool was asked to do a job it cannot own.
Illustration: the same KPI is calculated separately on different sheets, copies, and codes. Collapse usually starts while it “still looks usable.”
Definitions break
Returns, free goods, uninvoiced sales—some add, some subtract. There is no document, only “we have always calculated it this way.”
Permissions break
A store manager can edit the HQ total. Finance will not lock the file. When someone leaves, macros and data sources leave with them.
Excel is not the villain. Trial margin math and a one-off campaign-cost column are faster there than in any BI tool. What breaks is treating it as the only operations ledger: no master data, no audit, no drill-down, no subscription. Stores go from 3 to 30, one assistant becomes a rotation, formulas go from readable to untouchable, and the weekly report turns from a tool into a risk. Why you need a dashboard you can challenge, see Why an operations dashboard is not optional.
The surface failure is a wrong number. The root is often master data. The Nanshan store is “Nanshan One” in the business system, “SZ-NS” in finance, and “Nanshan District Store” in the export. One VLOOKUP off by a row and the whole margin column drifts. Without code governance, visualization lies too; see Without master data, visualization lies. Exporting a slice from each system and stitching by hand is even riskier; see When every system calculates its own numbers, what breaks.
Some teams say: we bought BI, we will wash Excel first and pour it in. If what you pour still has no definition and no primary key, BI only paints the collapse more nicely. Decorative BI is the same as doing nothing; see Buying BI as decoration is the same as doing nothing. The right order: lock three to five metrics that must be watched and their source tables, reconcile row counts, then let the reporting platform replace Friday stitching. Do not estimate cost as “a cooler Excel replacement”; see What dashboards or a data platform cost, and how long they take.
On DaXi projects, turning off parallel Excel is part of acceptance. A lit dashboard with weekly reports still circulating privately is two truths. Phase one does not need a data platform. Extract APIs, close permissions, and drill to documents. Then the next chart is faster; see Once APIs are standardized, the next chart is faster. To move the weekly report off the desktop into an auditable operations entry, go to the Data service page and tell us who maintains that v7 today.
Two people editing at once, hidden columns, or a file nobody can rerun after someone leaves—start a dashboard
Judge whether Excel has gone out of bounds on three things: do two people edit at once, are there hidden columns, and can you rerun it after someone leaves. If any one is true, start a dashboard. Do not hire someone who is “better at Excel.”
Scratch sheets can stay, but they must be labeled “not the official definition.” The weekly meeting only trusts KPIs signed on the dashboard. Two number sets in parallel do not remove collapse. They move it.
A quieter failure: macros depend on a path on someone’s laptop. Swap the notebook on Friday and nothing runs. A dashboard puts extraction on a server and APIs, not a personal desktop. The assistant’s job should move to checking exceptions, not being human ETL.