The four-day close nobody puts on a roadmap

Ask an operations lead what their team's biggest recurring cost is and they will name a vendor. Ask them what the month-end reporting pack costs and you usually get a pause, then an estimate, then a worse number than the vendor.

It is rarely one person for four days. It is a day from the analyst who pulls the exports, half a day from whoever owns the mapping tab, two hours of someone in finance disagreeing with a figure, and a long afternoon at the end where the charts get rebuilt because the ranges moved. Spread across four people it never looks like four days, which is precisely why it never becomes a project.

Why it stays invisible

A cost gets fixed when someone owns it. This one has three properties that stop that happening:

1. It is distributed. No single person loses a week, so no single person escalates.
2. It always gets done. The pack ships. Nothing burns down, so nothing forces a decision.
3. The fix looks expensive. "Get this into the warehouse properly" is a quarter of engineering time nobody has, so the comparison is always four days a month against a quarter of work, and four days wins every time.

That third one is the trap. The comparison is wrong because it assumes the only alternative is the full modelling project.

What actually has to change

The recurring cost is not producing the numbers. It is re-producing the structure — re-joining the same files, re-applying the same exclusions, re-pointing the same charts — every single month, by hand, from scratch.

So the thing worth fixing is not the report. It is the fact that last month's work does not carry forward.

In Dashful that carry-forward is the dataset. You prepare it once: the agent profiles your files, works out how they relate, and proposes a structure; you approve it, which is where your judgement about returns and test orders and the month the warehouse moved gets encoded. From then on the dataset is a thing that continues. Next month's export lands on it, the rows append, the quality checks re-run, and every dashboard reading it refreshes.

The four days become an upload.

The part people underestimate

The saving is real, but it is not the biggest one. The bigger one is that the pack stops being a person.

When the structure lives in a workbook, it lives in the head of whoever built the workbook. They go on holiday and the close slips. They leave and someone spends a fortnight reverse-engineering a lookup tab written in 2024. When the structure is an approved dataset with named measures and stated assumptions, it is legible to the next person without an archaeology project.

That is worth more than the four days, and it is even harder to get onto a roadmap.