Reporting that survives the person who built it

Every company has at least one report that only one person can produce. It works. It has worked for years. And the week that person is away, it does not happen, and everyone agrees that is fine because it is only a week.

Then they leave, and the report either dies or gets rebuilt from scratch by someone who does not know why the numbers were the way they were.

Why reports become people

Not through carelessness. Through accumulation.

A report starts simple. Then a supplier changes their format and a workaround goes in. Then finance asks for inter-company transfers to be excluded and a filter appears. Then there is the month the warehouse moved and the comparison stops being valid, so a note gets added — in an email, not in the file. Each step is reasonable. None of them is written anywhere durable, because a spreadsheet has no place to write down why.

After two years the report is a set of undocumented decisions wearing a chart. The person who made those decisions can explain every one. Nobody else can explain any.

What "durable" actually requires

Three things, and only one of them is technical.

The structure has to be separate from the output. In a workbook they are the same object, which is why you cannot inspect one without disturbing the other. A dataset that produces dashboards can be read on its own: its grain, its columns, its measures.

The assumptions have to have a home. This is the one that matters most and gets skipped most. When the agent proposes a dataset it states what it assumed — including the ones you will disagree with — and approving it is how those assumptions become part of the thing rather than part of somebody's memory. A dataset that excludes transfers says so where the next person will find it.

The definition has to be used, not copied. A measure defined once on the dataset and referenced by every dashboard cannot drift. A formula copied into six workbooks will, and the drift is invisible until two numbers disagree in a meeting.

The test

Here is the one we use, and we would suggest it regardless of what tooling you choose:

Could somebody who joined last month produce this report correctly, without asking anyone, from what is written down?

If the answer is no, you do not have a reporting process. You have a person, and a risk that is currently being carried by their loyalty and their holiday schedule.

Most companies discover the answer at the worst possible moment. It is a cheap thing to check early, and the checking costs nothing but an uncomfortable half hour.