Game theory tells you what position you must hold. Statistics tells you the odds your plan actually gets you there. Almost no company computes either.
Ask a Fortune 500 strategy team what their plan is, and you will receive a narrative: a market thesis, a three-year roadmap, a base-case financial model with bull and bear sensitivities bracketing it like decorative bookends. Ask them a different question such as what is the probability this plan reaches the position the company must hold to survive? and you will receive silence, because nobody in the room has computed it, and in most companies, nobody ever has.
That silence is the most expensive habit in corporate America. It killed eight servicemen at Desert One in 1980, when mission planners launched eight helicopters against a six-helicopter requirement without ever computing that the demonstrated reliability of the fleet of helicopters selected for the mission, gave them roughly a one-in-three chance of reaching their start line. It cost Kroger a $2.6B impairment in 2025, when a site-by-site demand analysis run seven years after the capital was committed, revealed that three of its eight automated fulfillment centers had been built in markets that never had a defensible probability of reaching breakeven volume. And it may be operating right now, at trillion-dollar scale, across an AI data center buildout whose builders have yet to publish a single probability statement about the demand required to earn a return on the capital.
Those are not three stories. They are one story, told three times: an organization that knew its requirement, felt its margin, and never computed its odds. I have written about each case individually. This article is about the method that connects them because what emerges from the wreckage is not a cautionary tale but a strategy methodology, and I would argue the most valuable one available to any operator willing to use it.
I call it probability of sufficiency, and it consists of two questions asked in strict sequence.
The Two Questions
Question one is game-theoretic: what position must this company hold? I have argued before that game theory is the only legitimate foundation for strategy, because strategy is by definition interactive…your outcome depends on the moves of competitors, suppliers, and disruptors who are simultaneously optimizing against you. Game theory defines the requirement: the market position, network density, technology capability, or cost structure below which you lose the game regardless of how well you execute. It is the business counterpart of the mission arithmetic that told Eagle Claw’s planners the answer was six helicopters when in fact the answer was seventeen. In war, the requirement comes from the mission; in business, it comes from the game. Either way, it comes first.
Question two is statistical: what is the probability that your current plan reaches that position? Not the base case. Not the forecast. The probability computed from demonstrated rates, against a confidence threshold set before you see the answer, with kill conditions written in advance.
Most companies answer neither question. The sophisticated ones answer the first and then substitute conviction for the second. The methodology is the pairing, and the pairing is non-negotiable: a requirement without a probability is a wish, and a probability without a requirement is trivia.
For readers who want the pairing’s theoretical pedigree, game theory itself discovered this gap and named it. A Nash equilibrium identifies the stable outcome of a game when every player optimizes against every other and it defines the position you must hold. But classical equilibrium analysis carries a hidden assumption: that players execute their chosen strategies perfectly. Reinhard Selten, who shared the 1994 Nobel Prize with John Nash, built his career on removing that assumption. His concept of trembling hand perfection asks which equilibria survive when players have a small probability of executional error, and a strategy that is optimal only under flawless execution is, in Selten’s framework, not a robust strategy at all. Probability of sufficiency is trembling hand perfection made operational. Eagle Claw’s plan may have been the right strategy in the game against Iran; it collapsed at the first tremble of a hydraulic pump. Seventeen helicopters was the trembling-hand-robust version of the same strategy. Nash tells you the game’s stable outcome. Selten asks whether you can hold it with a trembling hand. The binomial math is the tremble, priced.
One boundary, stated honestly: the second question is not itself game theory, because most of the failure probabilities it prices — hydraulic pumps, grocery demand curves, GPU depreciation schedules — come from nature, and nature is not a strategic player. Dust storms do not study your playbook. A complete strategy must survive two kinds of uncertainty: what your opponents do, and what the world does to your plan, and the two questions assign one discipline to each.
Stated as a discipline, the second question decomposes into five steps, and they fit on one page:
State the requirement as a threshold, not a target. Six helicopters. Breakeven orders per week. The freight density below which your network is a rounding error. If the plan has a hard minimum, and every real strategy does, name it in writing.
Price the per-unit probability from demonstrated rates, not conviction. The RH-53D’s maintenance record existed and nobody pulled it. Amazon’s logistics expansion rate is not a scenario; it is a measured historical sequence. Use what the world has demonstrated, not what your plan requires the world to do.
Set the confidence level before computing the answer. A hostage rescue demands 99 percent. A billion-dollar facility deserves at least 75. Choosing the confidence level after seeing the result is how base cases get laundered into business cases.
Compute the probability of sufficiency — and believe the output. If the math says the plan as designed cannot be resourced to the required confidence, the answer is not “proceed and hope.” It is redesign. Eagle Claw’s honest answer was seventeen helicopters on a deck that held twelve, which was not an argument for twelve, it was an argument for a different mission architecture.
Write the kill conditions in advance, and honor them. Colonel Charlie Beckwith decided months before launch that fewer than six flyable helicopters meant abort, and he honored that criterion in the Iranian desert under pressure that would have broken most executives in most boardrooms. Kroger’s kill condition arrived seven years and $2.6B late. The discipline is not the number; it is the pre-commitment.
“This Is Just Real Options Rebranded”
Let me address something head-on: sophisticated readers will say this is real options analysis wearing a new jacket, and before them, that it is Howard Raiffa’s decision analysis, or Monte Carlo simulation, or scenario planning. Nothing in the mathematics here is new. Binomial sufficiency calculations are undergraduate statistics. Decision trees have been in the literature since the 1960s. Real options theory earned its place decades ago. I didn’t claim that I’ve invented new math. Anyone claiming to have invented the math is selling something.
What I have done, however, is combined multiple disciplines into a methodology that doesn’t live anywhere else except on my laptop. I’m not disputing the past, present or future of real options, I’m simply critiquing its flaws.
Real options is a valuation technique. This is a decision governance discipline. Real options answers the question “what is this flexibility worth?” and it prices the value of the ability to expand, defer, or abandon an individual investment. Probability of sufficiency answers a different question entirely: “what is the probability my plan reaches the threshold it must reach?” One assigns a dollar value to optionality; the other computes the odds of survival. A company can run flawless real options valuations on every project in its portfolio and still never once compute whether the portfolio, in aggregate, gets it to the position game theory says it must hold.
Real options, in practice, is applied after the proposal to justify it. This is applied before, to constrain it. Here is the dirty secret of real options in the wild: it became the corporate rationalization tool of choice precisely because “strategic option value” can be invoked to justify almost anything. The dot-com era ran on it; every land-grab was an option on the future. When the valuation technique arrives after the executive team has already fallen in love with the deal, it produces whatever number the deal requires. Probability of sufficiency inverts the sequence: the requirement, the confidence level, and the kill conditions are committed before the analysis is run which is exactly the feature that makes it uncomfortable, and exactly the feature that makes it work. The methodology’s power is not in the arithmetic; it is in the timestamp.
Real options assumes graceful payoffs. This is built for thresholds. Options mathematics presumes a continuous payoff of more demand is worth more, less is worth less, and the curve is smooth. But the decisions that destroy companies are threshold systems: six helicopters or no mission; breakeven volume or a fixed-cost crater; sufficient network density or acquisition at a distressed price. In a threshold system, “close” is worth nothing, expected value is actively misleading, and the only number that matters is the probability of clearing the line. That is not a real options calculation. It is a different question about a different geometry of risk.
The predecessors lack the enforcement mechanism. Scenario planning generates stories without probabilities. Monte Carlo generates probabilities without committed thresholds. Decision analysis has lived in textbooks for sixty years and dies in boardrooms every day, because none of these tools arrive bundled with the governance that gives them teeth: confidence levels set in advance, kill conditions in writing, and an institutional commitment to honor the abort criterion when every incentive screams to proceed. Beckwith’s abort call is the component every prior framework is missing. The methodology is the integration plus the pre-commitment.
So no, this is not real options rebranded. Real options is one instrument in the orchestra. This is the score, plus the discipline to stop playing when the music says stop.
Application One: The 3PL Industry Should Be Running This Math on Amazon Right Now
Amazon’s logistics expansion is the single most consequential demonstrated rate in American commerce. In roughly a decade, Amazon went from zero to the largest parcel carrier in the United States, and its June 2026 entry into LTL freight extends the pattern into a segment the 3PL industry had comfortably assumed was structurally protected. Every mid-tier 3PL executive has an opinion about this. Almost none of them has a probability.
Here is what the methodology demands of them:
The game-theoretic requirement: define the survival threshold of the combination of network density, technology stack, vertical specialization, and customer diversification below which the firm stops being a strategic acquisition and becomes a distressed one. That threshold exists for every 3PL in the Armstrong & Associates Top 50, whether or not its board has named it.
The demonstrated rate: Amazon’s encroachment is not a scenario to be debated; it is a measured sequence…parcel, then middle mile, then fulfillment services for external shippers, and now LTL with observable intervals between segment entries. Price the distribution from the record, not from reassuring conference-panel consensus about what Amazon “won’t bother with.”
The computation: what is the probability that the organic plan reaches the survival threshold before Amazon’s demonstrated rate reaches this firm’s segment? When that probability falls below roughly 60 percent, M&A stops being one strategic option among several and becomes a mathematical requirement, and the same calculation tells you which side of the transaction you are on, and approximately by what date.
This is the methodological spine beneath consolidation analysis in the 3PL space. The existential question facing every mid-tier operator is not “should we consider M&A?” It is “what does the sufficiency math say?” Run honestly, the calculation converts the industry’s most emotional debate into arithmetic which is precisely why so few boards want to run it.
Application Two: Retailers, Autonomous Vehicles, and Buying the Distribution
The retail industry’s autonomous vehicle problem is a timing problem, and timing problems are where point-estimate strategy fails most reliably. Walk the floor at any retail conference and you will hear two camps: AVs are imminent, or AVs are a decade of hype away. Both camps are making the same analytical error of treating an adoption distribution as a date.
AV delivery economics are a distribution over cost-per-mile decline curves, regulatory timelines, and geographic rollout sequences. The methodology reframes the retailer’s question accordingly:
Not “when will autonomous delivery be ready?” but “at what probability that AV delivery cost crosses below my store-fulfillment cost by year X does my logistics infrastructure investment clear its hurdle rate?” That is a computable number, and it converts an ideological debate into a trigger.
When the distribution is wide, the correct strategy is not to bet a point on it — it is to buy exposure to the whole distribution. This is the statistical case for orchestration. A retailer that hard-wires its network around a single AV bet of one partner, one vehicle architecture, one rollout geography, is drawing once from a wide distribution and praying. A retailer that plugs into a carrier-agnostic orchestration layer holds an option across every draw: whichever AV operator, timeline, or geography wins, the retailer’s demand flows to it. Orchestration neutrality is not a philosophical preference; it is the mathematically dominant position under uncertainty, for exactly the reason a wide distribution punishes point bets.
And the kill conditions write themselves: the cost-per-mile figure, the regulatory milestone, the date at which the retailer shifts capital from store-based fulfillment to AV-integrated infrastructure. Committed in advance, those triggers replace the two worst outcomes in retail strategy: moving years too early on conviction, or years too late on consensus.
Why Nobody Does This
If the methodology is this cheap…no consultants, no software, and one page of math, the obvious question is why it remains so rare. The answer is not analytical. It is organizational, and it has three parts.
First, incentives: base cases get funded and probabilities get interrogated. An executive who presents a plan with “82 percent probability of sufficiency” has volunteered an 18 percent probability of failure to a room that will remember it, while the executive presenting a confident narrative has volunteered nothing. The methodology punishes honesty in any organization that hasn’t decided to reward it.
Second, the org chart: probability lives with the CFO and risk teams, strategy lives with the narrative builders, and the methodology dies in the gap between them. The companies that run this math institutionally, and a handful of the best capital allocators quietly do, have fused the two functions at the point of decision.
Third, and most fundamentally: kill conditions are career conditions. Writing an abort criterion in advance means creating a document that can later prove you should have stopped. Beckwith did it anyway, honored it in the desert, and was vindicated by history precisely because the criterion predated the pressure. Most executives would rather keep the criterion vague which is to say, most executives would rather launch eight helicopters.
Strategy, as practiced, is a narrative with a spreadsheet attached. Strategy, as it should be practiced, is a probability statement: a named threshold, a demonstrated rate, a confidence level chosen in advance, and a written commitment about what happens if the math says no. Game theory supplies the first question. Statistics supplies the second. Everything else…the frameworks, the off-sites, and the three-year roadmaps, are commentary.
The planners at Desert One had the requirement and skipped the probability. Kroger had the capital and skipped the threshold. The data center builders have the conviction and, so far as anyone can see, have skipped the confidence level. The pattern is forty-six years old, it has never once been the math’s fault, and it will keep collecting its debts until boards start asking the only question that matters:
What is the probability this plan is sufficient, and who in this room computed it?