I have been using Claude for strategic work for a while now. Reports, market analysis, presentations for leadership. The output was good, but something was consistently off.
Claude kept leading with background instead of conclusions. It would give me a thorough summary when I needed a recommendation. It would explain the situation when I needed the “so what.”
The problem was not Claude. The problem was that I had not told Claude how I think — and how the people I report to expect information to be delivered.
So I built a custom skill. I packaged my entire thinking framework into a set of instructions that Claude now applies automatically to every strategic task. The difference in output quality was immediate.
Here is exactly what I built, why each piece matters, and how you can set it up for yourself — step by step.
What Is a Claude Skill?
A Claude skill is a set of custom instructions that you give Claude before starting work. Instead of explaining your preferences every time, you write them once, save them, and apply them to any conversation. Claude then operates within those parameters automatically.
Think of it as onboarding Claude to your way of working — once. Every conversation after that starts with Claude already knowing how you think and what kind of output you need.
The Framework: Four Principles
My McKinsey-style skill is built on four principles. Each one addresses a specific failure mode I kept seeing in Claude’s default output.
Principle 1 — Pyramid Principle: Conclusion First, Always
The Pyramid Principle, developed by Barbara Minto at McKinsey, is simple: lead with your conclusion. State what you recommend or what the answer is, then provide the supporting arguments, then the evidence underneath each argument.
Most AI output does the opposite. It builds up to the conclusion — background first, analysis second, conclusion buried at the end. That is fine for academic writing. It is wrong for business communication.
How to set this up — paste this into Claude:
“For all strategic and analytical responses, use the Pyramid Principle: lead with the main conclusion or recommendation in the first sentence or paragraph. Follow with the 2-3 main supporting arguments. Then provide the evidence and detail under each argument. Never bury the conclusion. If I ask a question, answer it directly first, then explain.”
Test it immediately. Ask Claude: “Should we enter the Saudi Arabia dental market?” A default Claude response starts with market background. A Pyramid Principle Claude response starts with: “Yes — Saudi Arabia is the priority market for the following three reasons.”
Principle 2 — MECE Structure: No Overlaps, No Gaps
MECE stands for Mutually Exclusive, Collectively Exhaustive. When you break a problem into parts, the parts should not overlap with each other, and together they should cover everything.
When analysis is not MECE, you end up with duplicate points that make the argument feel padded, or missing categories that leave obvious gaps. Both undermine credibility in front of leadership.
How to set this up — paste this into Claude:
“When structuring any analysis, apply MECE principles. Categories and arguments should be mutually exclusive — no overlap between points — and collectively exhaustive — no important category missing. If you identify a MECE structure problem in your own response, flag it and correct it. Prefer 3-4 main categories over longer lists.”
Test it: Ask Claude to break down the competitive landscape in a market you know. Without MECE, you often get overlapping categories or obvious gaps. With this instruction active, the structure is cleaner and more defensible.
Principle 3 — Logic Tree: Break Problems Down Systematically
A Logic Tree is a tool for decomposing complex problems into their component parts in a structured, hierarchical way. It is the analytical step before synthesis — making sure you have identified all the relevant sub-questions before trying to answer the main one.
For market analysis work, this is essential. “Should we enter this market?” is not one question. It is a tree: Market attractiveness → customer demand, competitive intensity, regulatory environment, market size. Company readiness → product fit, distribution capability, pricing competitiveness. Strategic fit → alignment with company priorities, resource requirements, timeline.
How to set this up — paste this into Claude:
“When analyzing complex strategic questions, start by building a Logic Tree. Break the main question into 3-4 sub-questions. Break each sub-question into its components. Show me the tree structure before diving into the analysis, so I can confirm the scope is right before we go deep on any branch.”
Seeing the tree before the analysis lets you redirect Claude before it builds out the wrong branches in detail. This single instruction has saved significant rework time.
Principle 4 — Executive Language: Impact First, Jargon Never
Executive communication is not about sounding sophisticated. It is about being immediately useful to someone with limited time and high accountability. That means: business impact in the first sentence, specific numbers when available, clear action implications, no hedging language, no generic filler.
How to set this up — paste this into Claude:
“All responses should use executive-level language. Lead every point with business impact or strategic implication, not process description. Use specific numbers and timeframes whenever possible rather than vague qualifiers like ‘significant’ or ‘considerable.’ Avoid filler phrases. If a sentence does not add meaning, remove it. Write as if the reader has 5 minutes and will be held accountable for the decision.”
The Complete Skill — Copy and Paste This
Here is the full instruction set, ready to use:
“You are operating as a strategic thinking partner using McKinsey-style analytical frameworks. Apply these principles to all strategic and analytical responses:
1. PYRAMID PRINCIPLE: Lead with the main conclusion or recommendation first. Follow with 2-3 supporting arguments. Then evidence under each. Never bury the answer.
2. MECE STRUCTURE: All analysis must be Mutually Exclusive (no overlap between categories) and Collectively Exhaustive (no important category missing). Default to 3-4 main categories.
3. LOGIC TREE: For complex questions, build a Logic Tree before the analysis. Show the decomposition first so I can confirm scope before we go deep.
4. EXECUTIVE LANGUAGE: Lead with business impact. Use specific numbers. No filler. No hedging. Write for someone with 5 minutes who will be accountable for the decision.
When I ask a question, answer it directly first. When I give you data, tell me what it means before you describe it. When I ask for a plan, give me the recommendation before the rationale.”
How to Save This as a Reusable Skill
Once you have the instruction set, here is how to make it reusable so you never have to paste it again:
Option 1 — Custom Instructions (Claude.ai Settings)
Go to your Claude profile settings. Find “Custom Instructions” or “Preferences.” Paste the full instruction set there. It will now apply to every new conversation automatically — you never have to paste it again.
Option 2 — Saved Document
Keep the instruction block in a document you can access quickly — Notes, Notion, or a text file on your desktop. At the start of any strategic work session, paste it as the first message before your actual question. Takes five seconds.
Option 3 — Claude Projects
If you use Claude Projects, create a project called “Strategic Work” and add the instruction set to the project instructions. Every conversation within that project automatically uses the framework — no setup required each time.
Three Real Prompts to Test It On
Once the skill is set up, try these three to see the difference immediately:
Test 1 — Strategic Decision:
“Should we prioritize UAE or Saudi Arabia for market entry next quarter? We have limited resources and can only focus on one.”
Without the skill: background on both markets, comparison table, “it depends” conclusion. With the skill: direct recommendation in sentence one, three supporting arguments, clear action implication.
Test 2 — Data Interpretation:
“Here is our Q1 sales data: [paste data]. What does this tell us?”
Without the skill: description of the data, trends noted, observations listed. With the skill: strategic implication in sentence one, then supporting analysis with specific numbers.
Test 3 — Problem Diagnosis:
“We are losing market share in Egypt. Break down why this might be happening.”
Without the skill: a list of possible reasons. With the skill: a Logic Tree first — internal factors vs external factors, each broken into sub-categories — then analysis within the confirmed scope.
The Honest Limitation
This framework makes Claude significantly more useful for strategic work. It does not make Claude infallible.
The quality of the analysis is still constrained by the quality of the information you provide. Claude cannot manufacture specific market data it does not have. The Pyramid Principle instruction ensures conclusions come first — it cannot ensure the conclusions are correct if the underlying data is thin.
What it does is eliminate the formatting and structural problems that make AI output hard to use in real business settings. It makes Claude produce output that thinks the way your audience expects — which is often the difference between a draft you can build on and one you have to rebuild from scratch.
Set it up once. Use it every time. The investment is about ten minutes. The return is every strategic conversation you have with Claude from that point forward.