War Gaming

War Gaming

When to use

Use before a major move where competitor or market reaction is genuinely uncertain and the stakes are high enough that "we'll figure it out" isn't good enough — pricing moves, market entry, product launches into a contested category. Also use to upgrade a narrative-only scenario exercise ("here are three things that might happen") into an actual decision tool with probabilities and payoffs.

What it does

Structures the decision as a tree: your strategic options as the first branch, each followed by chance-node scenarios (the ways the market/competitors could respond) with assigned probabilities and payoffs. The bundled calculator computes expected value per option and separately checks the maximin (best worst-case) option — because the EV-maximizing choice and the safest choice are sometimes different, and that's a risk-tolerance decision the business needs to make consciously, not one a spreadsheet should make silently.

Method

1. Define the strategic options being compared (the same discipline as strategic-options — genuinely distinct paths, not variations on one theme).
2. For each option, build a comparable set of response scenarios — ideally drawing on competitive-intel's capability/incentive/constraint profiles so the scenarios are grounded in what a competitor would plausibly do, not a brainstormed list.
3. Assign probabilities to each scenario within an option, summing to 1.0. Where real data doesn't exist, use structured expert elicitation (ask multiple people independently, then reconcile) rather than one person's gut number — and label the confidence level of the probabilities explicitly.
4. Assign a payoff to each scenario in a consistent unit (usually $ impact over a defined time horizon) — include downside scenarios with real negative payoffs, not just "less upside," or the tree can't do its job.
5. Run the bundled calculator (scripts/decision_tree_ev.py) to compute expected value per option, plus worst-case and best-case bounds, and a maximin comparison.
6. When EV-best and maximin-best disagree, treat that as the central finding, not a footnote: ask explicitly whether the business could survive the EV-maximizing option's worst case. If not, expected value is the wrong decision criterion here regardless of how favorable the average looks.
7. Stress-test the probabilities: rerun with a plausible pessimistic shift in probabilities (more weight on the bad scenarios) to see if the ranking is robust or fragile to being wrong about the odds.
8. Translate into a monitoring plan: name the early signals that would indicate which scenario is actually unfolding, so the team can react in real time rather than waiting for the full outcome.

Inputs

• Strategic options under consideration
• Plausible response scenarios per option (ideally sourced from a competitive-intel profile)
• Probability estimates per scenario (labeled with confidence level)
• Payoff estimates per scenario, including realistic downside cases
• Config saved as JSON matching the format documented at the top of scripts/decision_tree_ev.py

Output format

Decision tree with options, scenarios, probabilities, and payoffs; EV, worst-case, and best-case per option; explicit flag and discussion when EV-best and maximin-best diverge; probability-stress-test result; monitoring plan tied to early signals per scenario.

Example

An aggressive price cut shows the highest expected value (470,000) but a worst case of -200,000 if the competitor matches; holding price and adding value shows a lower EV (420,000) but a worst case of +300,000. The calculator flags the divergence explicitly. If a -200,000 outcome would breach a covenant or force a difficult follow-on fundraise, that changes the decision even though the aggressive option "wins" on paper — a conversation the raw EV number alone would never force.

Common pitfalls

• Running scenario planning as a narrative exercise with no probabilities or payoffs, so it can't actually be compared to an alternative.
• Defaulting to the highest-EV option without checking whether the business can survive its worst case.
• Assigning probabilities without labeling their confidence, letting a guess be treated with the same weight as measured data.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Natan-Mohart
• Source: Natan-Mohart/24-strategy-skills-for-claude
• License: MIT

Install and usage instructions live in the source repository linked above.