Dashboards create demand for dashboards
Nobody planned an estate of a hundred workbooks. It accreted, one perfectly reasonable request at a time.
I led a migration of an enterprise reporting estate off an end of life server: over a hundred workbooks, more than one hundred and fifty sources, four business functions that did not talk to each other much. It went across without an unplanned outage, which is the part I was hired for.
The part I keep thinking about is what the inventory looked like when we counted it. Nobody had ever decided to have that many. There was no year in which somebody approved a hundred workbooks. Each one had been a small, sensible request that a competent person said yes to, and the total was an accident that took years to have.
The mechanism is not mysterious
A view answers a question. Answering it raises two more, because that is what answers do.
Somebody sees a figure fall and immediately wants it split by region. The split by region shows one region is the whole story, so now they want that region over time. The trend has a step in it, so now somebody wants the same view filtered to the segment that changed. Every one of those requests is legitimate. Refusing any of them individually would be indefensible.
So the demand for reporting is produced by the supply of reporting. Not primarily by the business getting more complicated. The estate is largely a record of curiosity that other parts of the estate created, and it grows at whatever rate the team can build, which means a faster team gets a bigger estate rather than a more finished one.
Nobody owns the growth, because the growth is not a thing anybody does. It is the sum of a hundred people each doing something reasonable, and there is no meeting at which the sum is on the agenda.
Migration is when you find out
An estate like that is unauditable in place. Nobody will read a list of a hundred workbooks and tell you which ones matter, because reading the list costs more than ignoring it, and ignoring it has no consequence today.
A migration has a consequence today. Every workbook that comes across costs work, so for the first time somebody has a reason to ask whether it should. That is the only forcing event I have seen actually work. Not a policy, not a review cadence, not a spring clean that gets scheduled and then moved. A hard deadline where carrying something has a price.
The uncomfortable finding, when you do ask, is not that half of it is unused. It is that nobody can tell you whether it is used, and the question surprises people. Reporting gets built and then falls out of anybody’s field of view, because a working view generates no events. Nothing about it ever asks for attention again.
What I thought would fix it
I used to believe the answer was up front governance. A request process, a justification, a standard. I have watched that mostly slow down the reasonable requests while doing nothing at all to the growth, because the growth is downstream of a real need and a real need routes around a form.
I was also wrong in the other direction, and more expensively. My instinct during that migration was that low usage meant low value, and it does not. A view that two people open once a quarter can be the one that catches something before it becomes a restatement. Usage measures frequency. It does not measure what would have happened.
So I have no clean answer, and I have noticed one thing. In years of doing this nobody has ever asked me to build something that removes a question. Every request adds one. The estate is not really a set of reports. It is a record of everything anyone ever wondered, kept forever, and nobody wants to be the person who decides that a question has finished being interesting.