How to build a McKinsey-style PowerPoint with AI (8 rules for consulting-grade slides)
Consultants don't win on pretty templates. The 8-rule AI workflow for McKinsey-style slides: answer first, action titles, a MECE structure, one idea per slide.
A McKinsey deck does not look consulting-grade because of the template. It reads that way because of a structure most people never see: every slide states a single conclusion in its title, the slides stack into one argument you can follow by reading only the titles, and nothing on the page exists that does not prove the point. That system predates PowerPoint — Barbara Minto, the first female MBA McKinsey hired, formalized it in the 1970s and published it in 1985 as "The Pyramid Principle." Get the structure right and a plain slide looks sharp; get it wrong and the fanciest template still reads like a class project.
AI changes the economics of building one, but not the rules. A model can draft action titles, pressure-test whether your buckets are MECE, and turn a paragraph into a clean slide layout in seconds — but it will happily generate a gorgeous deck that says nothing if you skip the thinking. This guide gives you the eight rules that separate a consulting slide from a corporate one, each paired with the exact AI prompt or step that does the work, and it ends by showing how to get the finished argument out of a .pptx and onto the platforms where your audience actually reads it.
The steps
Rule 1 — Start with the answer, then let AI build the storyline. Consulting decks lead with the recommendation and support it downward — the Pyramid Principle. Before any slide, write the one-sentence answer to the question the deck exists to settle. Then prompt the model with the SCQA frame: give it the Situation (what everyone agrees on), the Complication (what changed), the Question it raises, and your Answer, and ask it to return a numbered storyline of 5–10 slide-level claims that prove that answer. That numbered list is your deck outline before you touch a single layout.
Rule 2 — Make every title an action title, not a label. A McKinsey title is a full sentence stating the slide's conclusion — "Churn is concentrated in the first 30 days" — not a topic label like "Churn analysis." The test partners use: read only the titles and the whole argument should hold together. Paste your storyline into the model and prompt: "Rewrite each of these as a complete-sentence action title stating the insight, under 12 words, no colons." Then read the titles in order and check they tell the story on their own.
Rule 3 — One idea per slide. Each slide proves exactly one claim — the one in its action title. If a slide needs two verbs to describe it, it is two slides. When you draft body content, feed the model one action title at a time and ask for only the evidence that supports that single title, so it cannot pile a second argument onto the page. This is the discipline that keeps a 40-slide deck skimmable in two minutes.
Rule 4 — Structure the supporting points to be MECE. Under each claim, the sub-points must be Mutually Exclusive and Collectively Exhaustive — no overlaps, no gaps — the second principle Minto gave McKinsey. This is where AI is genuinely useful as a critic: give it your buckets ("we split the market into enterprise, mid-market, and cheap") and ask, "Are these MECE? Name any overlaps or missing categories." It will surface the "what about self-serve?" gaps you would defend in a live meeting.
Rule 5 — Build the ghost deck before you design anything. Consultants sketch a "ghost deck" — every slide as just its action title plus a rough note of the one chart or visual that proves it — and get that storyline approved before spending a minute on formatting. Have the model output a table: slide number, action title, chart type, and the single data point or takeaway. Review that table as the whole argument. Designing before the storyline is locked is how people waste a day polishing slides that get cut.
Rule 6 — Cut everything that does not prove the point. Signal-to-noise is a design rule with teeth: remove chartjunk, gridlines, drop shadows, logos on every slide, and any number that is not load-bearing. Prompt the model to "list every element on this slide that does not directly support the title, ranked by how safe it is to delete." Then delete from the top. A consulting slide is defined as much by what is absent as by what it shows.
Rule 7 — Let the message choose the chart. The chart follows the claim, not your data dump. Comparison across categories is a bar chart; change over time is a line; parts of a whole is a stacked bar (rarely a pie); correlation is a scatter. Tell the model the exact sentence the chart must prove and ask which chart type says it most directly and what the one annotation or callout should be. Then label the takeaway on the chart itself — the reader should not have to decode the axes to get your point.
Rule 8 — Keep the visual system restrained and consistent. Consulting-grade means one typeface, a two- or three-color palette (usually a neutral plus one accent for emphasis), a fixed grid, and identical title placement on every slide. Give an AI deck tool — Gamma, ChatGPT Work, or PowerPoint's Copilot — your locked storyline and action titles plus that style constraint, and let it lay out the slides against the grid. The AI handles alignment and spacing; you supply the structure and the restraint, because a tool left to its own taste adds decoration you just spent Rule 6 removing.
Common gotchas
AI writes fluent, confident nonsense. Every stat, market size, and "studies show" a model puts on a slide has to be checked against a real source before it goes in front of a client — a fabricated number on a decision deck is a career problem, not a typo.
A beautiful template with no storyline is the most common failure. If your titles do not tell the argument on their own, no amount of formatting saves the deck — fix Rules 1–2 before touching design.
Do not let the model generate all 40 slides in one shot. It will produce topic labels, two ideas per slide, and overlapping buckets. Drive it one rule at a time: storyline, then titles, then evidence, then layout.
MECE is easy to claim and hard to hold. Buckets that sound clean ("small, medium, large customers") often overlap on the edges or miss a segment; make the model argue against your structure, not just approve it.
Pie charts and 3-D effects read as amateur in a consulting context. When a tool defaults to them, override it — a labeled bar or line almost always states the point more honestly.
The deck is not the deliverable your audience remembers. A 40-slide .pptx gets skimmed once in a room; the ideas only travel if you also pull them into formats people actually read on their own time.
Where Kompozy fits
A consulting deck is a persuasion machine that runs exactly once. You spend a day building the storyline, the action titles, the MECE argument — and then it lives in a .pptx that gets skimmed in a single meeting and dropped in a shared drive. The thinking is the valuable part; the slide file is the least durable place to keep it. Be honest about the tool boundary first: Kompozy does not build the .pptx. For the actual slide layout, a deck tool like [Gamma](/alternatives/gamma) or [ChatGPT Work](/news/openai-chatgpt-work-launch) is the right instrument, and this guide points you there in Rule 8.
Where Kompozy comes in is the step after the deck is done — turning that locked argument into published content, which is a generation job, not a design one. Your storyline is already structured content: each action title is a claim, each slide has one takeaway, the whole thing is MECE. Feed that structure into Kompozy and it generates net-new assets from it. The action titles become a brand-exact Carousel rendered through HyperFrames, one insight per card, so the deck's spine becomes a scrollable post. The executive summary becomes a [Blog Article and a set of Text Posts](/glossary/output-buckets) written under your [Persona Brief](/glossary/persona-brief), so the recommendation reads in your firm's register instead of the model's default voice. A single chart becomes an Infographic Photo. And the core recommendation becomes a [Persona Short](/glossary/persona-shorts) — a face-locked talking-head video that delivers the answer in 30 seconds for the people who will never open the deck.
Then it publishes: the same argument fans across the eight social platforms plus blog and email, on a schedule, behind a per-post review gate so nothing ships until you approve it. One deck, one afternoon of consulting-grade thinking, becomes a week of content that carries the ideas past the room. Honest framing: if you only ever need the slides for one presentation, skip this and just build the deck. Kompozy earns its place when the argument is worth repeating — Starter ($99/mo, 5,500 credits) for a solo consultant or founder turning each deck into a content set, Pro ($299/mo, 18,000 credits) for an agency or team running many decks into multi-format output, Enterprise custom.
Frequently asked questions
Can AI build a McKinsey-style PowerPoint on its own?
It can build the artifact but not the argument. AI tools like Gamma, ChatGPT Work, and PowerPoint Copilot generate clean slides fast, but the consulting quality comes from the structure — answer-first storyline, action titles, MECE buckets, one idea per slide — which you have to direct. Give the model that structure and it produces a strong deck; ask it for "a McKinsey deck about X" cold and you get a pretty deck that says nothing.
What actually makes a slide look "McKinsey style"?
Not the template — the discipline. Full-sentence action titles that state the conclusion, a Pyramid-Principle storyline you can follow by reading only the titles, MECE supporting points, exactly one idea per slide, ruthless signal-to-noise, and a restrained visual system (one font, two or three colors, a fixed grid). Barbara Minto formalized the structural half of that at McKinsey in the 1970s.
What is the Pyramid Principle?
A structuring method Barbara Minto developed at McKinsey and published in 1985: lead with your answer or recommendation, support it with a small set of arguments, and back each argument with data. On slides it means the deck opens with the conclusion and every slide's title is a claim that supports it, so an executive can grasp the whole argument from the titles alone.
Which AI tool is best for consulting-style slides?
For the slide file itself, dedicated deck tools — Gamma, ChatGPT Work, or PowerPoint's Copilot — lay out clean, on-grid slides from your storyline. For the thinking (storyline, action titles, MECE check), any strong LLM like Claude or GPT works. Use the LLM to lock the structure first, then hand a deck tool the finished titles and outline to design against.
How do I get the deck's ideas in front of more people than the room?
Repurpose the storyline. Each action title is a carousel slide or a LinkedIn post, the executive summary is a blog article, the key charts are infographic posts, and the recommendation is a short talking-head video. An engine like Kompozy turns that finished argument into all of those and publishes them across platforms, so the deck's thinking travels past the one meeting it was built for.