Your Competitor's Next Move Is a Probability Distribution, Not a Headline

Every competitive-intelligence tool can tell you that your competitor raised $200M. Most can tell you within the hour. None of them can answer the three questions your CRO actually asks in the room: what will they do with it, how much does that cost us, and what should we do first?

Those are not retrieval questions. They are decision-theory questions, and answering them takes a different kind of machinery than a feed — it takes a model of the game you are actually in: incomplete information, beliefs instead of certainties, and payoffs that depend on what both sides do.[2]

Step one: the move is a distribution

PYRAMYD's engine starts by treating "what do they do next?" as a probability distribution over concrete candidate moves — ship an agent tier, cut entry pricing, push into a new segment, acquire the capability they lack. Each move starts from a confidence derived from signals: hiring patterns, filings, release cadence, pricing-page diffs, review language.

Then the distribution gets reweighted by something most CI programs never formalize: Porter's Four-Corner analysis — what the competitor's management is actually incentivised, organised, and funded to do.[1] A move that fits their declared strategy and their org chart can carry up to four times its signal-derived weight; a move that contradicts both can be cut to a quarter. The result is a proper simplex: the moves compete for one unit of belief, so raising one lowers the rest.

Step two: exploitability, not likelihood

Knowing a move is likely tells you nothing about whether you should care. The number that matters is different: if they make this move and you respond well, how much better off are you than if you had done nothing? The engine computes that as the best-responder's expected gain over the status quo, weighted by the posterior — so a near-certain move you cannot profit from scores near zero, which is the correct answer.

A 90%-probable move you can't exploit is trivia. A 40%-probable move with a wide-open counter is the thing to prepare for this quarter.

Step three: the blind-spot amplifier

This is the part with no equivalent in any feed. Porter's third corner — assumptions — asks what the competitor believes. Sometimes the evidence contradicts the belief: they price as if mid-market churn is a cost problem, while their own reviews name onboarding time as the reason people leave. When a predicted move rests on a belief like that, the engine makes it more exploitable, not less — because a competitor committed to a mistake keeps funding it.

Step four: a band, not a point

A single expected payoff hides the number that actually decides whether you act: how badly this can go. The engine runs four thousand seeded draws over the joint move-and-response tree and reports the P5, the median, the P95 — and an explicit probability that you end up net worse. The mean of the simulation converges on the enumerated expectation, which doubles as a built-in cross-check that the tree and the sampler agree.

The output is a ranked list of counter-moves — prescriptions — ordered by posterior-weighted expected exploitable value, each with the moves it blocks and the payoff it forfeits if you skip it. Not a list to read. An order to act in.

The part that keeps it honest

A forecast you cannot grade later is entertainment. Weather forecasting learned this in 1950[3]: write the probability down before the outcome, score it after. Every prediction the engine issues is written to a ledger with the leading indicators that would confirm it, and scored against what actually happened once it resolves. The engine also refuses in the other direction — a move with no assessed payoff contributes exactly zero exploitability rather than an optimistic default, and every synthesis ships with its epistemics attached: this is calibrated decision-support under imperfect information, not certainty.

Run it yourself

The formulas in this post are not a whitepaper abstraction — the production engine's pure core runs client-side, and we put it on a public page with sliders. Drag a move's confidence and watch the posterior renormalize, the exploitability re-band, and the prescriptions re-rank. The scenario is illustrative; the math is the real thing.

Run the engine at pyramyd.ai/game-theory →