AI for managers

You don't need another spreadsheet. You need the answers it has been giving while nobody had time to read them. That's where AI earns its keep: it reads volume, spots the pattern, and hands back short text someone can actually read before the meeting starts.

The method barely changes from one case to the next: paste the real data, ask for a specific format, check what came out. The examples below are ready to run — swap in your own numbers.

From raw spreadsheet to executive report

Export your data as CSV and paste it into ChatGPT. Then:

The requested format is what turns an answer into a report. Without "1 page", you get four dense paragraphs nobody opens. With it, you get something that fits on a screen and drops into the deck as is.

When the numbers matter, always paste the real base and never accept a figure you didn't provide: AI reads what exists convincingly and invents what's missing with equal confidence — the subject of ChatGPT making things up.

Three management routines, three prompts

Meeting prep, cash flow reading, and proposal comparison eat up any manager's calendar — and all three have a ready-made prompt:

1. Results meeting: "Based on this data [paste], prepare 5 slides for a results meeting. Each slide: 1 key number, 1 insight, and 1 discussion question."
2. Cash flow: "Here are our inflows and outflows for the last 6 months [paste]. Flag the months with the highest deficit risk, the seasonality patterns, and a projection for the next 3 months."
3. Vendors: "Compare these 3 proposals [paste a summary of each] and build a table with price, timeline, terms, and risk. Recommend one, with justification."

Notice the pattern: what to paste, what format should come out, and what to do with the result. A request missing those three parts comes back generic — and you burn the next round fixing it.

What AI may — and may not — see

Customer data is personal data. Before pasting a name, tax ID, contract, or case number, ask whether there's a lawful basis for it. When in doubt, anonymize: swap the name for "Customer A" and keep only the fragment the analysis needs.

Keep a fixed list of what's cleared for use — aggregated numbers, open budget, product metrics. Anything outside it goes in summarized, not pasted. That's the logic behind the NIST AI Risk Management Framework: decide what the tool may do with the information before you use it.

When the output affects a person, people decide

Promotions, reviews, headcount cuts, hiring. AI helps with structure — criteria, a draft evaluation, a conversation outline. Not with the verdict.

If your team starts using AI to read resumes or performance reviews, define the criteria up front, let AI draft, and have a human confirm the conclusion afterward. The same holds outside the company — see AI for career.

Rule of thumb: the model takes part in the draft, never in the verdict.

Frequently asked questions

Can I paste the company's financial data into ChatGPT?

Yes, if it's internal and already open — budget, metrics, aggregated data. Customer data with names, contracts, or documents needs a lawful basis. Anonymizing takes seconds and barely changes the analysis.

How do I stop AI from inventing numbers?

Paste the source and ask for each number to come with its origin. Then check dates, values, and names against the original file: if it isn't in the file, it shouldn't be in the output.

How long until this becomes routine?

The first report takes minutes: export, paste, ask for the format. The payoff shows up in the next cycle, when the template already exists and you only swap the data.

Should AI decide on hiring or headcount cuts?

It can organize — criteria, drafts, outlines. It shouldn't decide. Any decision that affects a person goes through someone who answers for it.

If you're just getting started, see 10 ways to use AI at work and the Before/After Method.