The weekly report nobody reads, and the one nobody trusts

Somebody in your business spends Friday afternoon building a report. It goes out at 5pm to eight people. On Monday, two of them open it. One of those two scrolls straight to the figure they care about, does not quite believe it, and checks it against their own spreadsheet.

Nobody has ever said the report should stop. It was asked for once, by someone senior, for a reason that made sense at the time. The reason has gone. The Friday afternoon has not.

Plenty of firms past twenty people have several of these. Each costs a few hours a week of someone's time, and the cost is invisible because it looks like diligence.

How AI changes this

AI tools make reports cheap to produce. A summary that took two hours now takes ten minutes. That sounds like good news, and for a report people use, it is.

For a report nobody uses, cheaper production means more of it. Once weekly summaries are generated with AI, it is easy to end up producing more of them, for more audiences, at greater length, because the effort is no longer a brake. The pile grows. Attention does not.

There is a trust problem underneath as well. A hand-built report has a person who knows where every number came from. An AI-written summary pulled from three exports has a paragraph that sounds sure of itself and a reader with no quick way to check it. When the reader cannot trace a figure, they fall back on their own spreadsheet, and the report has become one more opinion.

A report register

The fix we recommend is short and a bit unforgiving. Write a one-page register of every recurring report. For each one, four lines:

• The decision it informs. Not "keeps leadership informed." A decision: whether to hire, whether to chase a debtor, whether to reorder stock, whether to move a crew.
• Who makes that decision. One person.
• Where each figure comes from. The source system and the definition used.
• When it was last used to make a decision. A date, or "cannot remember."

Any report that cannot fill in the first line is a candidate to stop. Any report whose last line says "cannot remember" goes on probation for a month. If nobody notices it missing, it was not needed.

The register is a piece of company memory. It holds the purpose of each report and the reason it exists, so that when someone new asks for "a weekly summary of everything," there is a place to check whether one already exists and what it is for.

Where AI fits safely

Once a report has a purpose and named sources, AI can help produce it without making the trust problem worse. The rule is that it reads the register first, uses only the named sources, and labels every figure with where it came from. A reader who doubts a number can follow the label back to the system in a few seconds.

A person signs off any report that goes to a lender, an investor or a board before it is sent. Those readers make decisions with real money on the strength of the figures, and a fluent paragraph with one wrong number in it does more damage than a late report.

The catch

It will not make bad source data good. If jobs are closed late or payments are posted in batches, the report inherits the delay, however clearly it is labelled.

It will also meet resistance. Stopping a report can feel like a demotion to the person who built it, or a loss of control to the person who asked for it. Expect a conversation or two. Say plainly that the time is going to something more useful, and name what that is.

For this week

List every report that goes out on a schedule in your business. For each one, ask its main recipient one question: what did you last decide because of this? Write down the answers word for word. The reports with no answer are your first cuts, and the hours they free up are the first return on this exercise.

Book 20 minutes at pgco.world/book.