BCG's Case for Multi-Agent War Gaming: Test the Strategy Before the Market Does
A strategy that has never been attacked is a guess. BCG’s Elton Parker, Nick D’Intino, Amy MacDougall and Max Struever argue that war gaming — participants playing competitors and other stakeholders — stress-tests strategy before reality does, and that AI makes the exercise faster, broader and deeper.
Their design advice is specific: use a multiagent setup with separate agents for competitors, customers, vendors and regulators, rather than one model playing everyone. Before a session, AI compresses weeks of preparation and can generate dozens of future worlds where an analog game covers a handful. During the session, it adds information and simulates extra actors while humans supply judgment. Afterward, it tests moves across a wider range of scenarios.
The examples carry the argument. A heavy equipment manufacturer’s agents showed that its baseline strategy consistently underperformed more ambitious moves, and surfaced contingent moves tied to triggers. A global health care company modeled more than 50 scenario permutations, ran more than 350 simulations and tested over 11,000 actions with more than 400 stakeholder agent versions. Its earlier pre-AI workshop had covered four stakeholder groups.
The success factors are plain: focus on one pivotal decision, and give it executive ownership. BCG’s three principles are to make war gaming an always-on capability, test assumptions early, and keep humans in control of the exercise.
Our read: the point is outcomes, not simulation volume. The agents widen what the room gets to argue about; they don’t replace the room. Pick one decision that would be costly to get wrong, name the executive who owns it, and run the first exercise before the plan is locked — while a bad assumption is still cheap to change.