Build a war-game style review that takes one problem, situation, or opportunity and attacks it from many seats and angles before you commit.
A war game is not a brainstorm and not a single devil's-advocate note. You freeze the scenario, assign seats that would actually show up in the real room, and force each seat to press the plan until something breaks or holds. The output you want is a short list of material objections, the evidence each objection needs, and the moves that would survive them.
Set the seating chart
Name the seats before you open a model. Finance cares about cash timing, margin, and what the forecast hides. Counsel cares about exposure, disclosure, and what you cannot say yet. A customer seat cares about switching cost, trust, and what fails in delivery. A competitor seat cares about the cheapest counter and the narrative the market will hear. A skeptic seat cares about the claim that looks strongest and is least sourced. Add an operator seat when execution risk is the real killer: capacity, dependencies, and the first ninety days after the decision.
Two seats are not enough for a war game. Arguing both sides clarifies a binary. A war game needs the leadership seating chart for the actual decision.
Write the seats into the prompt as named roles with goals, not as vague vibes. Example: "Finance: attack cash timing and margin assumptions. Counsel: list disclosure and contract exposure. Customer: state what would break trust in delivery. Competitor: describe the cheapest counter-move. Skeptic: pick the strongest-looking unsupported claim." Keep the same source pack for every seat so disagreements are about judgment.
What to feed the review
- One problem, situation, or opportunity stated without the preferred solution bloating the brief.
- The current recommended move and the two next-best alternatives.
- Numbers, contracts, customer evidence, and competitive facts you can stand behind.
- Hard constraints: budget, legal, timing, people, and political limits inside the company.
- The decision you will make if the war game finds nothing new.
Use sequential passes as the loop
Pingpong runs models in sequence: each later model receives the original question and earlier answers, with instructions to assess the work so far. Use that loop as the war-game engine. First pass: steelman the recommended move. Second: finance and counsel attack. Third: customer and competitor respond. Fourth: skeptic hunts unsupported claims. Final: turn surviving objections into tests, mitigations, or a changed recommendation. Agreement across passes is not proof. Keep the objections that still have evidence gaps.
Pair this with multi-perspective strategy review, red-team review, pre-mortem, and before you commit. Browse more scenarios in the war-game decisions hub. Walkthrough: how to run a Pingpong. Product shape: how it works and multi-model AI review.