Six tasks a functional head should run through AI before the next monthly review

Six tasks below, all from a functional head's desk, all runnable before your next monthly review. Each one takes work you already do, hands the typing to ChatGPT, and ends with a check you run yourself. The check is where your judgment shows up. Run two of the six and you will learn which outputs you had to correct.

Where do you start with AI if you already run a function?

Start with the variance commentary on last month's numbers, because you already write it and you already know when it is wrong. Paste the variance table, with customer names and employee data stripped out, and ask for the biggest movers, the questions your boss will ask about each, and the actions you would take. The draft lands in minutes and you spend the rest of the hour on the parts that need you. Your check is to recompute two lines yourself against the ledger before the draft goes anywhere. A commentary built on one wrong subtraction reads exactly as confident as a correct one.

Write the competitor summary from three reports instead of one

Three sources beat one, and the contradictions between them are the useful part. Upload three syndicated category reports on the same market and ask for one master table that names where they disagree rather than hiding it. Ask for the table, a short summary, and the source URL in every row. Your check is to open three of those links and confirm the claim in the cell. A row with no source is a row you delete.

Turn the vendor comparison into a table you can attack

A vendor comparison is a ranking, and the criteria are yours. List the handful of things that actually decide it for your function, weight them, then have ChatGPT fill the grid from the quote documents and spec sheets you upload. Blank cells stay blank and you go back to the vendor. Your check is to change the weight on one criterion and see whether the order survives. If the ranking flips on a small change, you have a tie rather than a winner.

Draft the AI use policy for your own function first

Nobody in your company has written down where the data line sits, which is why your team is guessing at it. Draft a one-page rule set for your function: which tools, which data never leaves the building, who signs off on anything customer facing. Ask for the draft, then mark every sentence that names a system or a type of data. Your check is to take those marked sentences to IT and confirm each one against the policy they have actually published. Your name goes on the page, so the line is yours to own.

The meeting pre-read and the board question bank come out of the same pack

Both run off the review pack you are already building, so do them together. For the pre-read, ask for one page: what changed, what it costs, what needs a decision in the room. For the question bank, ask for the hardest questions a board member could ask about that same pack. Your check on the pre-read is to read the page against the pack and strike any line the numbers do not carry. Your check on the bank is to answer every question in it yourself in one line. The ones you cannot answer are your real agenda, and the pre-read gets rewritten around them.

How do you get better with AI once these tasks are running?

You get better by keeping count of where the drafts were wrong, not by running more tasks.

What do I do when the output is right but I cannot use it?

Rewrite the brief, not the output. A correct draft you cannot send is usually a format problem: wrong length, wrong reader, wrong order. Say who reads it, how long it runs, and what it has to decide, then run it again. Two rounds of that beats editing the same draft by hand.

Can I hand a working prompt to someone on my team?

Yes, hand over the prompt and keep the sign-off. Write down what the prompt is for, what it must never do, and the one check that has to pass before anything leaves the function. Your direct report runs it. You still read the check.

What do I say when the board asks for my view on AI?

Describe what you have run and what you corrected. A functional head who can say which two tasks now start as drafts, and where the draft was wrong last month, has a view. One corrected output from your own numbers carries further than any slide about adoption.