The 4:45-on-a-Friday question

It’s 4:45 on a Friday. The CEO messages: “Quick one, why is programme spend up so much on last year? Need it for a call Monday morning.”

Everyone in FP&A knows what happens next. Somebody, usually the analyst who was about to close the laptop, starts the archaeology. Which spreadsheet has that cut? Is it the March forecast or the reforecast? Six tabs deep, cross-checking against the board pack, because the last time someone answered from memory the number turned out to be from the version before the restatement.

The answer already exists. It’s sitting in a workbook the team produced weeks ago. The work isn’t analysis. It’s finding, and it still eats the evening.

It’s not an interruption. It’s a tax.

One Friday question is an anecdote. But count them across a month: the board member’s follow-up, the audit query, the “what did we assume for…” from strategy, the CEO’s quick ones. Each pulls a skilled analyst off real work for an hour or three, to retrieve a number that already exists somewhere in the team’s own output.

Finance teams are permanently stretched for a structural reason: finance is a cost centre. The business case for another FP&A hire loses to a revenue hire every single time, no matter how loudly the team is drowning. So the ad-hoc tax doesn’t get staffed away. It just compounds, quarter after quarter, paid in evenings and in analysis that never gets done.

The part that stings

Here’s the uncomfortable arithmetic: the team’s most expensive skill is judgment, knowing what a variance means, what to do about it. But a huge slice of the week goes to retrieval, knowing where a number lives. Retrieval is necessary. It is also, increasingly, automatable.

The teams that figure this out first won’t look like they hired anyone. They’ll just answer the Friday question in ten minutes, verified, with the source attached, and go home.

flowchart TB
  subgraph F[The Friday version]
    direction TB
    A1[Question lands at 4:45] --> A2[Hunt for the right workbook] --> A3[Cross-check versions and tabs] --> A4[Answer at 7:30, fingers crossed]
  end
  subgraph G[What good looks like]
    direction TB
    B1[Question lands at 4:45] --> B2[Ask a system that has read everything] --> B3[Verified answer, source attached, by 5:00]
  end

What good looks like

Whatever tool or process you land on, hold it to the standard the Friday question deserves:

1. It has read everything. Not the three files someone remembered to upload.
2. It shows its source. The exact schedule, the exact cell, one click away.
3. It knows which version. Budget, Actuals or the March forecast, and it never blends them.
4. It says “I don’t know” when the documents don’t support an answer. A confident wrong number at 4:45 on a Friday is how Monday’s call goes wrong.

That last one matters most. The goal isn’t answering faster. It’s answering faster and being right about it, at the moment nobody has time to double-check.

This is part of a series explaining AI and the systems around it for finance people, in their own language. I build AI systems for finance teams; the series is what I’ve learned doing it.