How to Generate a McKinsey-Style Consulting Deck from Your Notes

How to Generate a McKinsey-Style Consulting Deck from Your Notes

Summary

  • A managing director scans a 50-slide deck in roughly five minutes by reading titles alone, so action titles must carry the entire argument.
  • The workflow is four steps: storyline notes with an AI agent into Marp Markdown, export with npx @marp-team/marp-cli your-deck.md --pptx --theme uncover, apply the layout rules, then lint with a second AI call.
  • Markdown-first drafting turns every revision into a diffable text edit and every template swap into a single --theme flag change.
  • Linting catches structural violations, but numerical claims and chart arithmetic still need a human pass before the deck reaches a client.
  • For regulated teams that cannot send client material to a public service, Jinba AI Consulting scopes and deploys the same workflow inside their own environment.

Building a high-quality consulting deck by hand takes hours. Formatting work alone (aligning text boxes, enforcing consistent typography, reformatting bullets pasted from older slides) consumes time that should go to analysis. The output is also fragile: a single round of stakeholder feedback repeats the same manual process.

The principles behind a McKinsey style presentation are not secret. The pyramid principle, action titles, and the one-idea-per-slide rule are well documented. What is missing for most teams is a repeatable, code-driven workflow that applies those principles consistently and makes revision tractable.

This article walks through a four-step workflow: use an AI agent to storyline raw notes into structured Markdown, convert that Markdown into a PDF or editable PPTX with Marp CLI, apply the core consulting layout rules, and optionally run an automated checker to enforce them.

What makes a consulting slide work

A managing director reviewing a 50-slide deck before a client meeting will spend roughly five minutes on it, reading titles only. The titles must carry the full storyline on their own. This is the titles test: read the slide headings in sequence and the argument should be complete without any supporting content.

That requirement drives every layout rule.

The pyramid principle at slide level. The pyramid principle is applied within each slide as a three-level stack:

  1. The action title: the main conclusion, at the top.
  2. Two to four supporting arguments in the body.
  3. Evidence, data, and charts at the bottom.

Every element on the slide serves a single purpose: to make the reader believe the action title.

The action title. An action title states the slide's single takeaway as a complete sentence. The concept was coined by BCG consultants in the 1990s. The rules for writing one:

  • Start with a strong verb (answer first).
  • State a conclusion, not a topic or a process.
  • Be specific; use numbers where the evidence supports them.
  • Keep it to 15 words or fewer, on no more than two lines.
  • Use the active voice.

The takeaway on a consulting slide lives in the action title itself. The title is the message.

One idea per slide. If the action title contains the word "and," the slide almost certainly requires splitting into two slides. A simplified, well-structured visual improves audience comprehension by an estimated 41% compared to a cluttered one.

The full anatomy. A standard consulting slide carries: the action title, supporting content, a section label (top-left for navigation), a footnote citing data sources, a footer with client and project details, and a page number. This layout is consistent across firms because it serves the same function: a deck that a partner can navigate without a guide.

The four-step workflow

Step 1: Storyline your notes with an AI agent

Storylining is the first stage of the MBB document process. It turns unstructured thinking into a slide map by writing a short paragraph for each future slide, where the first sentence of each paragraph becomes that slide's action title. Performed manually, this is the most time-consuming part of deck production. Performed by an AI agent, it takes minutes.

Use ANTHROPIC_AGENT with Claude Opus (claude-opus-4-7) and the following prompt. Paste your raw notes at the end.

You are a strategy consultant from a top-tier firm. Your task is to take my
raw notes and structure them into a logical storyline for a consulting
presentation.

The output must be a single Markdown file structured for Marp CLI.
Use '---' to separate each slide.

For each slide, follow these rules:
1. Create a verb-first action title as an H2 heading (##). The title must
state the single key takeaway in a complete sentence, be under 15 words,
and use the active voice.
2. Write 2–4 supporting bullet points that directly prove the action title.
3. If a concept is too complex for one slide, split it. Do not use "and"
in action titles.
4. At the bottom of each slide, add a standalone takeaway in italics that
rephrases the action title as a one-line summary.

Here are my notes:
[INSERT YOUR UNSTRUCTURED NOTES HERE]

The output is a plain .md file, structured and ready for the next step.

Step 2: Convert Markdown to PPTX or PDF with Marp CLI

Marp CLI converts Marp-formatted Markdown into HTML, PDF, and editable PowerPoint files from a single source. It runs without a local installation via npx, requiring only Node.js v18+.

# Generate an editable PowerPoint deck
npx @marp-team/marp-cli your-deck.md --pptx --theme uncover

# Generate a PDF
npx @marp-team/marp-cli your-deck.md --pdf --theme gaia

Marp ships three built-in themes: default, gaia, and uncover. Changing the look of the entire deck is a single flag change, with no reformatting, template pasting, or misaligned bullets.

Converting to PDF or PPTX requires a local browser (Google Chrome, Microsoft Edge, or Firefox). For sandboxed environments such as E2B where no local browser is available, Marp provides an official Docker container image and standalone binaries for Linux, macOS, and Windows.

Step 3: Apply the consulting layout rules in Markdown

Before converting, review the .md file against the core rules:

  • Action title check: Every ## heading should start with a verb, state a conclusion, and contain no "and."
  • One-idea check: If a slide's body runs to more than four bullet points, split the slide at the --- separator.
  • Standalone takeaway: Add an italicised summary line immediately before each --- separator to reinforce the title for skimmers.
  • Typography: McKinsey's standard pairing is Georgia for titles and Arial for body text. Custom Marp themes support font configuration for teams that need to match a client template.

Making these edits in a text editor is faster than making them in PowerPoint, and the changes propagate cleanly to the next export.

Step 4 (optional): Lint the deck with an automated checker

Existing automated tools do not enforce consulting slide rules. Readability checkers such as SlideSpeak's AI checker measure Flesch scores and sentence length. PowerPoint's built-in checker covers accessibility. None of them checks for action titles.

A second AI call using ANTHROPIC_INVOKE acts as a linter. Paste the generated Markdown and use this prompt:

You are a consulting Engagement Manager reviewing a slide deck in Markdown
format. For each slide, flag any violations of the following rules:

1. Action title: Does the H2 heading start with a strong verb? Is it a
complete sentence? Is it over 15 words?
2. One idea per slide: Does the action title contain "and" or "or,"
suggesting it should be split?
3. Pyramid structure: Does the body content directly support the action
title, or does it introduce a separate point?

Return your feedback as a numbered list of suggested changes, referencing
each slide by its position in the deck.

[PASTE YOUR MARP MARKDOWN HERE]

This step catches structural violations before the deck reaches a stakeholder and eliminates one complete round of internal revision.

Why Markdown-first solves the revision problem

The revision cycle is the most demanding part of deck production: gathering stakeholder feedback, incorporating edits from multiple reviewers, and keeping formatting consistent across versions. A 100-slide deck built in PowerPoint can absorb more revision hours than its initial construction.

When the deck lives in a .md file, every revision is a plain text edit. Changes are diffable in Git. Reviewers can comment on specific lines rather than annotating a PDF or passing around Final_v3_partner_edits.pptx. The iteration cycle that consultants describe as "ping pong" becomes a structured, asynchronous process with a clear audit trail.

Switching client templates is a one-line change to the --theme flag, with no copy-paste step, no reformatted bullets, and no colour inconsistencies introduced by moving content between files. The content and the presentation layer are fully separated.

AI-generated decks carry one known failure mode: arithmetic and visual errors. A waterfall chart that does not sum its components correctly, or a chart that contradicts the action title above it, undermines the entire deck's credibility. The linting step in Step 4 catches structural violations, but numerical claims in the body content still require a human verification pass before the deck goes to a client.

What to apply first

The workflow above is most effective when the AI storylining step is taken seriously. The quality of the Markdown the agent produces depends directly on the quality of the input notes. Before running the ANTHROPIC_AGENT prompt, consolidate your notes into a single document and mark the conclusion you want the client to leave with. That conclusion becomes the governing action title for the deck's executive summary slide, and the rest of the storyline flows from it.

For teams operating in mckinsey ai banking and other regulated sectors where client data cannot leave a controlled environment, the Marp Docker image and the ANTHROPIC_INVOKE approach both support sandboxed execution. The .md source file never needs to touch a public service.

The four steps in sequence:

  1. Run ANTHROPIC_AGENT with Claude Opus to convert raw notes into slide-by-slide Markdown with verb-first action titles.
  2. Run npx @marp-team/marp-cli your-deck.md --pptx --theme uncover to generate an editable PPTX.
  3. Review the Markdown against the one-idea-per-slide and action-title rules; edit in plain text.
  4. Run ANTHROPIC_INVOKE as a linter to flag structural violations before the deck reaches stakeholders.

The principles behind a strong McKinsey style presentation have not changed. What has changed is the time required to apply them. This workflow shifts the time investment from formatting and initial structuring to analysis and message refinement, where it produces value.

Frequently asked questions

What is a McKinsey style presentation?

A McKinsey style presentation is a consulting deck built on action titles, the pyramid principle, and one idea per slide, so that the titles alone carry the full argument. Each slide stacks a single conclusion at the top, two to four supporting arguments in the body, and the evidence underneath. The layout is standardised across firms (section label top-left, sourced footnote, client footer, page number) because it lets a partner navigate the deck without a guide. The style is less about aesthetics than about making the argument readable in five minutes of title-only scanning.

What is an action title, and how do I write one?

An action title is a complete sentence at the top of a slide that states the single takeaway the reader should believe, rather than naming the topic. To write one, start with a strong verb, state a conclusion instead of a process, add a number where the evidence supports it, and keep it under 15 words on no more than two lines in the active voice. If the title contains "and," the slide almost always needs to be split into two.

How do I convert Markdown into an editable PowerPoint deck?

Run Marp CLI against your Markdown file with the --pptx flag. The command npx @marp-team/marp-cli your-deck.md --pptx --theme uncover produces an editable PowerPoint; swap --pptx for --pdf to export a PDF instead. The conversion requires Node.js v18+ and a local browser (Chrome, Edge, or Firefox), or Marp's official Docker image and standalone binaries in sandboxed environments where no browser is available.

Do I need coding skills to use this workflow?

No. The workflow uses plain-text Markdown and one command-line call, so anyone comfortable writing a document can produce the source file. The AI agent writes the Marp-formatted Markdown from raw notes; the remaining work is editing text and pasting a single command into a terminal rather than writing code.

Is Marp CLI free, and can I use it on client work?

Yes. Marp CLI is open source and free to use, including on commercial and client-facing work, and it needs no licence, account, or paid tier to export PDF or PPTX. It ships with three built-in themes (default, gaia, and uncover) and accepts custom themes, so client branding becomes a reusable theme file rather than a per-deck formatting exercise.

Can I apply these consulting slide rules without Markdown?

Yes. Action titles, one idea per slide, pyramid structure, and sourced footnotes are presentation principles, not software features, and they apply equally in PowerPoint, Google Slides, or Keynote. What Markdown adds is enforcement and revision speed: edits become diffable plain text, and a theme change propagates across the whole deck instead of one slide at a time.

Can I apply a client's own template and fonts?

Yes. Marp custom themes control layout, colour, and typography, so a client template becomes a reusable theme file applied with a single --theme flag change. McKinsey's own standard pairing, Georgia for titles and Arial for body text, is a straightforward theme configuration, and any house font stack can be substituted the same way.

How long does it take to build a consulting-style deck with this workflow?

Storylining is the slow step, and it falls from hours of manual drafting to minutes once an AI agent converts raw notes into a slide map. Formatting and alignment time disappears almost entirely, because Marp applies the layout rather than requiring manual repositioning of text boxes. Time saved goes to analysis, message refinement, and a human verification pass. The saving compounds at revision, when each round becomes a text edit rather than a full re-layout.

Can I use this workflow on confidential client data?

Yes, provided the AI calls run inside a controlled environment. The Marp Docker image and the ANTHROPIC_INVOKE linting call both support sandboxed execution, so the .md source never has to touch a public service. That matters for teams in regulated sectors such as banking, where client material cannot leave the firm's perimeter.

Are AI-generated consulting decks reliable?

The structure is reliable; the numbers are not, until a human checks them. An AI agent can produce a coherent storyline and enforce action-title rules, but it can also generate a waterfall chart that does not sum its components, or a chart that contradicts the title above it: both undermine the deck's credibility in front of a client. Use the linting pass to catch structural violations, and a separate human review for every numerical claim before delivery to a client.

Build your way.

The AI layer for your entire organization.

Get Started