How to turn your weekly management report into a brief ChatGPT can run
By the end of this you have one written brief for your Monday management report, saved once and used again every week. The first pass takes about an hour. After that it is a paste and a file upload. The brief names the report, the fields in it, the questions you want answered, and the shape of the answer you want back.
How do I use ChatGPT for work I already do every week?
Start with a document you already produce, not with a new idea. For most owners that document is the weekly management report: sales by item, dispatches, collections, outstandings, rejections, plant downtime. It repeats, the numbers in it are already checked by your own people, and you read it anyway.
The worked example below is a corrugated box plant. One plant, one sheet, Monday morning. Every step here is done on that sheet. If your business is distribution or logistics, the fields change and the method does not.
Write the brief before you open the chat
Write the brief in a text file first, because a brief you can see is a brief you can correct. My 30-day guide gives the five parts to write: Role, Context, Task, Format, Constraints. Use them in that order.
For the box plant the brief reads like this.
- Role: you are reviewing a weekly operations and sales report for the owner of a corrugated box plant.
- Context: the plant runs two machines, sells to FMCG and pharma packers, and bills on dispatch.
- Task: list the three movements that matter most this week, the questions the owner should ask the plant head, and the decisions that cannot wait.
- Format: one table of movements, then two short lists, no introduction and no summary at the end.
- Constraints: use only the attached file, name the column and the row behind every number you state, and say "not in the file" where the answer is not there.
That last constraint is the one that earns its place. It forces every line of the output back to a cell you can check.
You know the brief is done when it makes sense to a new assistant who has never seen your plant. Read it once as that person. Anything you filled in from your own head goes into Context.
Strip the data before you upload it
Take out what does not belong in a public tool before the file leaves your machine. Customer names become codes, and you keep the code sheet in your own folder. Employee names and salaries come out. Bank details, signed rate contracts and anything under a confidentiality clause come out.
What stays is what the analysis needs: item, quantity, rate band, dispatch date, rejection reason, machine. The box plant's sheet loses the buyer column and gains a buyer code column. Nothing about the movements you want explained is lost.
The check: open the stripped file and read it as an outsider. If you can identify a customer, a price to that customer, or a person, strip again.
Run the brief on last week's numbers
Paste the brief, attach the stripped file, and read the output next to the sheet, not on its own. ChatGPT takes an uploaded spreadsheet, runs code over it, and returns tables and charts from the rows in it. Read the first run for two things: a movements table you can tie to cells, and questions you already knew to ask.
Every brief I have written started here. Fix it in the brief, not in the chat. If the output treated rejections as a cost when you wanted them as a quality signal, that belongs in Context. If it ranked by value when you run the plant by machine hours, that belongs in Task. Rerun on the same file and compare the two outputs line by line.
When a run needs no change to the brief, the brief is the asset. The report is now a standing task with a written specification, and a long-standing assistant can run it at the start of the week while you read the result.
What do I check before this goes to my team?
Two checks, every week, before anyone else sees the output. First, recompute two numbers yourself from the sheet. Pick the largest movement and one number you did not expect. If either does not tie to the cells named, the output is wrong and the run is scrap.
Second, fact-check three statements that are not arithmetic. A cause named for a rejection spike, a buyer described as slowing down, a trend called out across weeks. Ask for the column and the row behind each one. A statement with no cell behind it is a sentence the tool produced, not a finding, and it does not go to your plant head with your name on it.
When a run fails both checks twice, the brief is too wide. Cut the Task to one question and build back up. This is the work I do on a client's own recurring report in A.D.A.P.T. AI Coaching.
What comes up in the first month of running this?
Do I have to upload the file every week?
Yes, each new week's numbers have to be attached to the chat that runs them. What you do not rewrite is the brief. Paste the same brief, attach the current file, and the only thing changing week to week is the data and your reading of it.
What if my report comes as a PDF from my accountant?
Ask your accountant for the same report as a spreadsheet. A table inside a PDF has to be read out before it can be added up, and that step introduces errors you will not see. If only the PDF exists, check every number the output names against the PDF yourself.
Should I keep this in a ChatGPT project instead of a new chat?
A project is the better home once the brief settles. A project holds files, chats and its own instructions in one place, so the brief lives in the project instructions and you stop pasting it. Until the brief stops changing, a plain chat each week is easier to correct.