Everybody has one: the Monday update, the weekly numbers, the status note nobody thanks you for and everybody notices when it is late. It is the same shape every week, it eats most of a morning, and most of that morning is assembly rather than thinking.
It is also the right first thing to automate, for an unglamorous reason: it repeats, you already know what a correct version looks like, and you will notice immediately when the output is wrong. Those three properties are what make an automation safe to trust — and their absence is why so many other candidates should be left alone.
Do one week by hand first
Before touching any tool, produce one week's report the slow way and pay attention to the seams. Which parts are pure assembly — pulling figures, reformatting, writing the same three sentences with different numbers in them? Which parts are judgement — deciding what matters, naming the thing that is going wrong, the sentence that changes what somebody does on Monday?
Automate the assembly. Keep the judgement. A report that has had the thinking automated out of it is a report people quietly stop reading, and you will not find out for a month.
A prompt you can run every week
Write this week's update for [audience], following the structure of last week's, pasted below.
Rules
- Use only numbers that appear in THIS WEEK'S DATA. If something is missing, write MISSING — never estimate it and never carry a figure over.
- Flag any line that moved more than [X]% against last week.
- Keep it under [N] words, in the same tone as the example.
- End with a short list of anything you could not determine from the data.
LAST WEEK'S UPDATE:
[paste]
THIS WEEK'S DATA:
[paste]Two lines there are doing most of the work. "Use only numbers that appear in the data" is what stops a plausible figure appearing where a real one was missing. "End with anything you could not determine" turns the model's gaps into a checklist for you, instead of a silent hole in a document you are about to send.