How I Use Claude For PowerPoint to Build a Board Deck in One Prompt
The month-end board deck is the most predictable three hours of my month. Same template, same charts, same commentary I already wrote in an email two days earlier. I open last month’s file, save it under a new name, and start swapping numbers. By the time it looks right, it’s late and I’ve added zero insight to the business. I just moved data from one rectangle to another.
I watched a controller on a team I worked with lose three hours to this every single month. Smart person, doing mechanical work, because the deck had to get built and nobody had given her a faster way.
This year I started handing that job to Claude. Not the “give me an outline I’ll rebuild in PowerPoint” version. The real thing. Claude builds actual .pptx files, with charts and formatting, right inside PowerPoint, from my finance data, and it matches our company template. This is the workflow I use now, the exact prompts, and the parts where I still keep my hands on the wheel.
What Claude For PowerPoint Can Build
There’s a category of AI slide tools that take a text prompt and generate pretty marketing slides. Big stock photos, three-word headlines, a gradient background. Useless for finance, because they don’t read your numbers and they don’t look like your company.
Claude For PowerPoint is a different animal. It runs as an add-in inside PowerPoint itself. You attach an Excel export, describe the deck you want, and it builds genuine slides you can open, edit, and present. It reads the data, builds the charts from your actual values, writes the commentary, and lays the slides out the way a finance deck should look. You watch it work in a sidebar, step by step, the same way you’d watch a regular Claude chat think.
The difference that matters: one tool decorates a text prompt, the other reads your general ledger and builds a board deck from it. Only one of those saves you the three hours.
Why “real .pptx” changes the math
A slide generator that gives you an image or a web page means you still rebuild everything in PowerPoint for the board. That’s the whole job, just with extra steps. When Claude builds editable slides that already match your template, the deck is most of the way done before you’ve touched it. You spend your time checking numbers and sharpening the story, which is where your judgment actually adds value.
What you need before you start
You need three things in place before the first prompt.
- PowerPoint with the Claude add-in installed. In PowerPoint, go to the Add-ins button, search for Claude, and install the one from Anthropic (it shows as “Claude in Microsoft Office, made by Anthropic”). It opens a sidebar that works like a normal Claude chat: chat history, file attachments, connectors, skills, model picker.
- A model and an edit mode picked. I run Opus 4.8, the current best model, and I turn on “accept all edits” so it can work without stopping to ask. Be careful with that setting and keep a saved copy of any existing work, since it will make changes directly to your file.
- Your data exports and a branded sample deck. A P&L export, whatever transaction or operational data you want, and one existing deck in your company’s format for the brand step below.
Step 1: Get Claude on your company template
This is a one-time setup at the start of each chat. You attach an existing branded deck and tell Claude for PowerPoint to match it.
The prompt that sets the brand
Act as a financial analyst with over a decade of experience building financial PowerPoints. I want you to look at the attached PowerPoint slides I've provided and then set up this PowerPoint presentation to match the colors, branding, and formatting of our existing presentation.Attach a real deck. In my run I gave it our F9 Finance Coffee Shop deck, which is deliberately tricky: dark forest green backgrounds, light mint text, bright green and gold accents, all in Calibri.
What Claude actually does with it
Claude inspects the uploaded deck to understand the theme colors, fonts, masters, and layout. When it noticed the original didn’t use slide masters, it scanned every individual slide for the styling instead. It read the whole brand correctly, pulled in our coffee logo, set up the theme and masters, and confirmed it back before building anything.
Why that confirmation step matters
Claude restating the colors, fonts, and motif it picked up lets you catch a wrong read before it’s baked into a dozen slides. And once it’s set, it remembers. Every slide it builds after this point inherits the branding automatically, with no extra prompting. That’s the whole game.
Step 2: Build a single slide from your GL export
Once the brand is set, I start small. One slide, so the mechanics are obvious before I scale up.
Prepping the export
You don’t need to clean the file first, and that’s part of the point. My P&L export is the same flat shape your ERP gives you. Claude reads that structure fine. Attach the raw file.
The one-slide prompt
Continue to act as a skilled financial analyst. I want you to add a single slide to this presentation, evaluating year-to-date actual performance versus budget based on the financials that I'm going to attach to this chat. In addition to a chart showing the numbers, I also want you to include some commentary based on the numbers.What came back was a single slide with four KPI tiles, a chart, and written variance commentary, already in our colors. Claude aggregated the data and read it the way an analyst would: revenue beat budget by 5%, cost of goods sold ran a little beyond that and compressed margins, marketing came in favorable and offset some of the cost of sales, netting a favorable EBITDA. It even added a footnote defining gross profit and EBITDA and citing where the data came from. That’s better sourcing than a lot of decks I get handed today.
Step 3: Editing by talking instead of rebuilding
A first draft is never final, and this is where a lot of AI tools fall down. They’ll generate something once, but ask for a change and you’re starting over.
Claude edits the slide you’ve already got. You talk to it the way you’d talk to a junior analyst who just handed you a slide. The first draft was honestly strong, but I wanted to show what reworking it looks like.
Great job on the slide. Want to make a few adjustments. Can you put variance commentary under each KPI, explaining each KPI and what the driver was? So one for revenue, one for gross profit, and so on. Then below that, spanning across the page, do the actual versus budget chart, but I would like to see all of the accounts, not just revenue and profit.It expanded each KPI card with its own commentary and widened the chart to show all six accounts: revenue, cost of sales, labor, marketing, rent, and utilities. It kept what was working and changed only what I asked for. It even flagged that revenue dwarfs rent and utilities on the chart, which makes the smaller accounts look tiny, a fair call and exactly the kind of note a good analyst would make.
The deck that flagged its own blind spots
The Situation: For the full build I gave Claude the P&L, point of sale transaction data, and some of our individual manager commentary, then asked for a complete CFO month-end review deck and told it to use its own discretion on the number of slides.
What Changed: Instead of dumping slides, Claude built a plan first: it read all the data, decided on a nine-slide structure, prototyped one location page before building the rest, then verified its own work at the end. Along the way it called out its own limitations without being asked. The point of sale file had no pricing, so it noted the operational metrics are transaction volumes and counts rather than revenue. It flagged that the budget is seasonally phased, not static. On the single slide earlier, it noted the one judgment call it made, that the comparison was against budget only with no prior-year view.
The Result: A deck that tells you where it might be wrong is worth more than one that hides it. Those caveats are exactly the questions a CFO would ask, answered before the meeting. I’d trust a junior analyst more for raising them, and I trust the deck more for the same reason.
Step 4: The full month-end deck in one prompt
Once the brand is set and the P&L is in the chat, the whole deck is a single instruction plus the extra data.
The master prompt
Continue to act as a financial analyst. In addition to the profit and loss data I've already given you, I'm going to attach point of sale transaction data. I want you to build out the entire CFO month-end executive review deck for the F9 Finance Coffee Shop. This should include looks at individual location performance, looks at operational performance, and it should also incorporate some of the individual manager commentary that I'm going to give you. Use your discretion on the number of slides and the content for each, but make sure it serves the CFO executive-level audience.What you get back
Claude planned and built nine slides:
- A month-end review title slide
- An executive summary
- A consolidated P&L
- A location scorecard
- A page each for Astoria, Hell’s Kitchen, and Lower Manhattan
- An operational performance slide (the transaction data showed the biggest volume in the pre-work morning hours, no surprise for a coffee shop)
- An outlook, risks, and recommendations slide
Nine slides, board-ready, in your brand, from one prompt. A build this size takes a while, and that’s fine. It’s writing an entire deck, checking its own work as it goes. The deck that used to eat my evening is a reviewed first draft by the time I get a coffee.
Where Claude still needs a human
I’m not going to pretend this runs unattended. The deck is a strong first draft, and you are still the analyst. Here’s the honest split of what it handles well and what you still own.
| Claude handles well | You still own |
|---|---|
| Reading the raw export and building charts | Verifying every number against the source |
| Matching your template and styling | Final judgment on what story the deck tells |
| Writing commentary in your saved style | Deciding what’s sensitive and how to frame it |
| Flagging its own caveats and data gaps | Anything that goes to the board as fact |
Check the numbers, every time
Claude reads the data well, and you still tie the headline figures back to the source before you present. This is true of any analyst’s work, including a human one. The difference is you’re reviewing a finished draft instead of building from a blank slide.
The judgment calls stay yours
Which layout to ship, how to frame a soft location, what the executive summary should emphasize. When I reworked that chart to show all six accounts, Claude was right that it wasn’t how I’d actually send it. Those calls are about your business and your audience. Claude gives you a clean starting point. The framing is your job.
How to run this on your next close
You can try this on your next month-end without changing a thing about your data. Here’s the order.
- Install the Claude add-in in PowerPoint, pick Opus 4.8, and attach an existing branded deck. Use the brand prompt above and let it set your theme.
- Drop in your P&L export and build one slide. Check the numbers against the file so you trust the read.
- Edit that slide by talking to it. Get a feel for how it revises without starting over.
- Add your transaction and commentary data and run the master prompt for the full deck.
- Review every figure, sharpen the story, and present.
Do it once on a real close and you’ll feel where the three hours went. The first session takes a little longer while you learn the prompts. Every one after that is a build and a review.