How a function head turns index data into a seven-slide deck
A recurring deck has two halves. The data half and the slide half. Run them as two separate jobs and both get better. The prompt in my library does exactly that: a 30-day closing value comparison table for Nifty 50 and NYSE Composite, then a clean, dark-themed 7-slide deck built from that same data. Here are the steps and the checks.
How do I get better with AI?
Take the deck you rebuild every month and run it end to end, twice. That is the method. Skill does not come from learning more features of the tool. It comes from taking a full piece of your own work through to a finished output, finding where the output was wrong, and rewriting the instruction so the next run lands closer.
For a senior functional leader the obvious candidate is the review deck. Some market or category movement, a comparison against something outside the company, then a handful of slides for a meeting where people ask hard questions. The comparison in my prompt library is two equity indexes over 30 days, Nifty 50 and NYSE Composite. Yours might be two geographies, two plants, two distributor regions, or your category against the broader market. The structure is identical. Two series, one window, one table, one deck.
Run it twice and the second pass starts from a brief you have already corrected. That is what getting better looks like.
What gets decided before anything is generated
Five decisions, written on paper first. Which two series you are comparing. Which window, with an explicit start date and end date. What a value means, which here is the closing value and not an intraday level. Which columns the table carries, such as the date, both closing values, both daily changes, and the gap between them. And where the numbers come from, named, so any row can be traced back.
That last decision carries the weight. Index levels are public, but a table assembled inside a chat window is not evidence until someone can point at the source behind each row. A leader who skips it ends up defending a figure in a review with nothing underneath it.
The window is where most of these tables go wrong. A 30-day comparison means 30 days of data, and two markets may not share one trading calendar. Check whether they do before you fix the window. Decide in advance what happens to a date that exists on one side and not the other. The gap gets filled one way or another, and the choice should be yours.
What the brief said, in two passes
Two prompts, not one. The first asks for the 30-day closing value comparison table for Nifty 50 and NYSE Composite. The second asks for a clean, dark-themed 7-slide PowerPoint deck built from the data the first pass produced.
Splitting the work is deliberate. The checked table becomes the record. Every slide after it traces back to a row somebody verified. When the deck is generated in the same breath as the data, there is no point at which a wrong figure can be caught, and it arrives already formatted and looking confident.
The first brief also carries the constraints: the date range, the column list, the sourcing requirement, and an instruction to leave a marked gap rather than quietly fill a missing date. The second brief carries the slide count, the theme, and one rule worth writing down. No number appears on a slide unless it appears in the table.
What the seven slides carry
A slide plan, written by the operator before anything is generated. Seven slides is a tight budget, so each one needs a job. A plan I would write for this comparison runs in three parts. The question the deck answers, then the method and the window. The table itself. Then the movement of each index, the gap between them read side by side, and what happens next.
Your function will want a different seven. A supply chain head puts service levels where a market movement sits. A marketing head puts category share there. The discipline is the same. Write the plan into the brief, because a deck built without one comes back as seven slides restating the table.
What do I check before the deck leaves my desk?
Three claims against a source outside the chat, every time. Check the first and last closing value of each series against the exchange page or the market data service you already use. Recompute one percentage by hand. Then read every sentence that explains why a market moved.
That third check is the one that matters most. Causal lines are what a reviewer remembers, and they are the least supported thing a closing price table can produce. Strike them or source them. I do the same with any sentence that looks forward. A 30-day table supports description and comparison. It does not support a view on next quarter. The view is yours, and your name is on the slide.
The formatting pass is quick and worth doing with a pen. Slide count as briefed. Every figure matched back to its row. Axes labeled, units stated, the window printed on the slide that carries the table.
This is the shape of work I run with clients inside A.D.A.P.T. AI Coaching.
What else do function heads ask about data-to-deck work?
Can I trust the numbers an AI output puts in a table?
Not until you have checked the endpoints and the sourcing. Verify the first and last closing value of each series against a market data page you already use, then recompute one daily change by hand. If those agree, the middle rows are usually sound. If any disagree, rebuild the table.
What if a slide shows a figure the table does not?
Fix the table first, or strike the figure. A number that appears only on a slide has no source behind it, so treat it as an error in the second brief and run that brief again against the checked table. Never correct it by hand on the slide, because the next run brings it back.
What if my data sits in an internal system?
Then start with public numbers, which is exactly why an index comparison makes a good first piece of work. Before company data goes near a public tool, ask whoever owns the policy for the line in writing: what may leave, in what form, for your function. Build the workflow while you wait.