Red-Team Your Strategy Doc from Three Hostile Seats

Workflow area: Strategy & Roadmap · Tier: Core · Hands-on time: ~30 minutes

The order matters more than the prompts here: the strategy is written by you first, attacked by AI second. Reversing it — having AI draft strategy and then polishing — produces the product-management theater this track keeps warning about: fluent documents with no judgment inside. A red-team pass on your own thinking is the opposite use of the model: it can't know your context well enough to write the strategy, but it's relentlessly good at finding the sentence where your logic leans on hope. Three separate hostile seats beat one generic critique because each seat has a worldview; "give me feedback" gets you politeness, "you are the competitor's CEO" gets you a plan of attack.

When to use — and when not to

• Use it on any strategy or big-bet doc before wide reviewannual strategy, a major pivot memo, the roadmap narrative for exec staff.
• Use it before pb-14's rehearsalthe red-team finds the weaknesses; the rehearsal practices defending them. In that order.
• Use it when the doc has had only friendly readersteams converge; three hostile seats simulate the readers your reviewers are too collegial to be.
• Skip it on AI-drafted strategyred-teaming a document the model wrote is the model grading its own homework in a different voice. The doc under attack must contain your judgment or the exercise is circular.
• Skip the fourth, fifth, and sixth seatsafter three passes the critiques repeat and the marginal finding is noise. Ship the revision instead of collecting objections indefinitely.

The recipe

1. 1.

   Finish the human draft first. The doc should state, in your own words: the bet, the why-now, what you're explicitly not doing, and the numbers underneath. Include the parts you're least sure of — sanding them off before the red-team defeats the purpose.

2. 2.

   Seat one: the competitor. Fresh conversation:

   Copy

   You are the CEO of [COMPETITOR or "our strongest competitor"], a
   [one-line description]. You have obtained our strategy document.
   
   Read it as an adversary:
   1. What does this doc tell you about our weaknesses and blind spots?
   2. Which of our assumptions would you attack first, and how —
      concretely, with the move you'd make this quarter?
   3. Where does the doc underestimate you?
   4. What in this strategy do you genuinely fear? (Be honest — this
      calibrates the rest.)
   
   Quote the specific lines of the document you're reacting to.
   
   Document:
   [PASTE DOC]
   

3. 3.

   Seat two: the CFO. New conversation, same doc:

   Copy

   You are a skeptical CFO reviewing this strategy for funding. You
   have seen a hundred product strategies overpromise.
   
   1. List every number, and mark each: SOURCED / ASSUMED / MISSING.
   2. Find the softest number — the one the whole case leans on —
      and stress it: what happens at half the assumed value?
   3. Where are costs understated or absent (headcount, opportunity
      cost, maintenance, AI/infra unit costs)?
   4. What would you need to see to fund this, and what would you
      cut first if the first two quarters disappoint?
   
   Quote the lines you're challenging.
   

4. 4.

   Seat three: the pre-mortem. New conversation:

   Copy

   It is 18 months from now and this strategy has failed. You are
   writing the honest internal post-mortem.
   
   1. The three most probable failure narratives, each traced to a
      specific assumption in the doc (quote it).
   2. For each: the early warning signal we'd see by month 3, and
      the point of no return.
   3. The failure mode the doc's authors are LEAST likely to see
      coming, given what the doc emphasizes.
   

5. 5.

   Adjudicate in a findings table — this is the human step. One row per distinct attack across all three seats:

   Copy

   FINDING (seat)                        | VERDICT   | ACTION
   CFO: expansion revenue assumes 130%   | ACCEPT    | re-model at 110%,
   NRR, our actual is 108%               |           | update doc §4
   Competitor: they'd bundle X free      | ACCEPT    | add counter-move
                                         |           | to risks section
   Pre-mortem: hiring plan slips 2 qtrs  | PARTIAL   | add trigger +
                                         |           | checkpoint at Q1
   Competitor: claims our segment is     | REJECT    | our win-rate data
   shrinking                             |           | says otherwise — noted
   

   Every ACCEPT changes the doc. Every REJECT gets a one-line reason — the rejects are your prepared answers when a human raises the same point in review.

6. 6.

   Revise, and keep the strongest objections visible. Fold accepted findings into the doc, and add a short "risks we're accepting with eyes open" section built from the best rejected-but-real attacks. A strategy doc that acknowledges its sharpest counterarguments reads as stronger, not weaker — and it pre-empts the meeting where someone lands the same punch.

Make it yours

• Localize the seats: swap the CFO for whoever actually kills proposals in your org — the CTO on feasibility, the head of sales on sellability, a board member on market timing. Attack from the seats that vote.
• Two-model red-team: run the same seat in Claude and ChatGPT and keep the union of findings — different models reliably find different soft spots, and the overlap tells you which weakness is most visible.
• The roadmap edition: run the same three seats against next quarter's roadmap one-pager before planning review — smaller doc, same discipline, thirty minutes.

Quality check

• The attacked doc was human-writtenif AI drafted the strategy, the red-team is theater; go back and write the judgment sections yourself first (pb-23's discipline).
• Findings quote the docan attack that doesn't cite a specific line is a generic objection; tighten the prompts until every finding points somewhere.
• The verdict table exists and rejects have reasonsaccepting everything means you had no strategy; rejecting everything means you weren't listening. Both patterns are diagnostic.
• At least one number got re-modeledthe CFO seat almost always finds a soft number; if it didn't, it wasn't skeptical enough — rerun with a harder system framing.
• The doc changeda red-team that produces no revision was either run too late to matter or read too generously to count.