An insight isn't yet a recommendation, and mixing the two up is why some synthesis feels sharp but useless: all diagnosis, no prescription. Richard Rumelt's Good Strategy / Bad Strategy gives the cleanest anatomy, and it maps straight onto the top of the So-What Ladder. Rumelt's "kernel" of any real strategy has three parts. There's a diagnosis, what's actually going on, which is the insight. There's a guiding policy, the overall approach to the diagnosed problem, which is the spine of the recommendation. And there's coherent action, the specific, coordinated steps that carry out the policy. A recommendation missing the diagnosis is an arbitrary demand: "grow the premium segment". Why? One missing the guiding policy is a list of disconnected good ideas. One missing coherent action is a slogan. Bad strategy, in Rumelt's phrase, is what you get when synthesis stops at the insight and calls it advice, or leaps to actions with no diagnosis under them.
Advertisement
What are the four parts of a recommendation?
In practice, wrap the kernel in a four-part recommendation you can say in one breath. It's the same answer · because · but · now you used in the 30-second close, now unpacked:
- Answer: the governing thought as an imperative. What to do (or decide, or believe). One sentence.
- Because: the MECE key line. These are the two to four reasons, each an insight, that make the answer the right one. Together they're the diagnosis plus the guiding policy: why this is the problem and why this is the approach.
- But: the risks and their mitigations. Every real recommendation has a way it could be wrong. Naming it raises your credibility, because it proves you looked and it gets ahead of the objection the client was forming. A risk without a mitigation is a worry. A risk with one is a plan. (The go/no-go call in commercial due diligence and the synergy stress test in M&A are this move at full scale.)
- Now: the coherent action. This is the concrete next step, with an owner and a rough timing, that turns advice into motion: "The first thing to do, this quarter, is…" A recommendation with no next step is a philosophy, and the client can't act on a philosophy on Monday.
The "but" and the "now" are what separate a consultant's recommendation from a commentator's opinion. Anyone can have a view. An advisor names the risk to their own view and hands you the first step anyway. That's no coincidence. It's the low self-orientation the trust equation prizes: you're visibly serving the decision instead of defending your cleverness.
{
"skill": "Pressure-testing a synthesis — sharpening the 'so what?' and red-teaming a pyramid for MECE gaps",
"use": "Once YOU have drafted the governing thought and the key line, use a chat assistant as a hostile first reader: ask it to name the weakest link in your argument, the objection a skeptical CEO would raise, and any overlap or gap in your key line — then fix what survives scrutiny. The AI is the sparring partner, never the author of the answer.",
"tools": ["Chat assistants — Claude, ChatGPT, Gemini", "A blank page or whiteboard (the real tool — the pyramid is built by you first)"],
"prompt": "Here is my recommendation and its three supporting arguments: <PASTE YOUR PYRAMID>. Act as a skeptical client CEO who is short on time. (1) Is my top-line answer actually an answer, or a finding in disguise? (2) Are my three supports mutually exclusive and collectively exhaustive — name any overlap or gap. (3) What is the single strongest objection to my recommendation, and have I addressed it? Do not rewrite it for me; interrogate it.",
"verify": "You own the governing thought — never accept an AI-generated 'so what' you cannot defend from your own analysis. Any figure the model introduces or restates must be traced to your primary source before it enters the pyramid (Guardrail 1-5): trace it, match unit/scale/period/entity, recompute anything derived, date-check, cite. A polished sentence from an AI that you cannot source is a liability, not a support.",
"diy": "You must be able to climb the So-What Ladder and build the MECE pyramid by hand first — that is exactly what this module gates. AI can stress-test a structure you built; it cannot install the judgment that built it. AI proposes phrasing; you dispose of the argument.",
"market": "IN"
}