A board we sat in on last year approved a forty-million-dollar commitment to a new manufacturing capability in roughly forty minutes. The deck was good. The team had done the work. The CFO had signed off on the cash plan. The vote was unanimous. Eighteen months later, after the capability had shipped, after the first customers had taken delivery, after the second cohort had quietly declined to renew, the company wrote down half the investment and the executive sponsor took a different role inside the company.
In the post-mortem, the question we asked the board was not what had gone wrong with the bet. The bet had gone roughly the way bets of that size, in that domain, go on average — which is to say, somewhat worse than the inside view had predicted. The question we asked instead was: what was the smallest version of this commitment you could have made, eighteen months earlier, that would have taught you most of what you needed to know to size the big one correctly?
The board took a while to answer. The eventual answer was something like four million dollars, twelve months, two pilot customers, and a single line on a contract-manufacturer's existing capacity. Roughly a tenth of the eventual write-down. The smaller bet, had it been made, would have produced the information that made the larger bet obviously wrong — or, equally plausibly, that justified an even larger commitment with more confidence. Either way, the four million was the cheapest piece of capital the company never spent.
This is the question, in our view, that every board should be required to put to every major commitment before the vote. It is almost never asked, and the cost of not asking it shows up, on average, several times a decade, in the form of write-downs that were foreseeable from the analytical posture rather than from the underlying business.
The skip from analysis to commitment
The pathology has a specific shape. The team works the analysis hard. Months of work, sometimes a year, produce a thick deck and a clear recommendation. The recommendation is presented as a binary: commit, or don't. The board, having read the deck, evaluates the binary. It approves the commitment, or it sends the team back to work the analysis harder.
What is missing from the choice architecture is the third option — the intermediate experiment. The thing you could do between the analysis and the commitment that would, for a fraction of the eventual capital, produce the piece of evidence the analysis cannot. This option exists for almost every major decision. It is, almost without exception, not on the slide.
The reason it is not on the slide is not that the team doesn't see it. The team often sees it; sometimes the team has explicitly considered it and rejected it. The reasons it gets rejected are revealing and worth naming.
The first reason is that the intermediate experiment looks, to the executive sponsor, like a vote of no confidence. To propose a four-million-dollar pilot in place of the forty-million-dollar commitment is to admit, in the room, that the case for the forty is not yet earned. Executives whose careers depend on appearing to lead with conviction rarely make this proposal voluntarily. The pilot is, organizationally, a humiliation.
The second reason is that the intermediate experiment, in most companies, has no institutional home. The forty-million commitment fits the capital-allocation process. The four-million pilot fits no process — it is too big for an operating budget, too small for a capital request, and too short-term for strategy. The team that proposes it has to invent the wrapper themselves, and most teams don't have the bandwidth.
The third reason is cultural. Boards reward bigness. A four-million pilot is harder to put on a board slide than a forty-million commitment. The CEO who walks into the boardroom with the small bet looks, in many cultures, less serious than the CEO who walks in with the large one. The incentives, perversely, run toward the larger commitment even when the smaller one would clearly produce the better expected outcome.
Douglas Hubbard, in How to Measure Anything, makes the point cleanly: the value of additional information is calculable, and is almost always highest for the variables on which the decision hinges most heavily. The expected value of the pilot, properly priced, frequently exceeds the expected value of the analysis-without-pilot. Companies do the analysis anyway, because the analysis fits the calendar and the pilot doesn't.
Designing a minimum-information bet
The discipline we recommend, and have installed in roughly thirty engagements, is straightforward to describe and slightly harder to actually do. Before any commitment over a defined threshold — pegged, for most companies, to about one percent of revenue or one quarter of operating profit, whichever is smaller — the team is required to answer a single supplementary question on the cover memo: what is the smallest version of this bet that would teach us most of what we'd need to know to make the full commitment with appropriate confidence?
The question forces three sub-questions.
What is the load-bearing assumption? For almost every major commitment, there is one assumption — sometimes two — that, if wrong, breaks the case. The customer adopts at this rate. The capability ships in this window. The competitor responds in this way. The unit economics scale in this direction. The team that has done its work knows which assumption it is. The team that hasn't, finds out by being asked.
What is the cheapest test of that assumption? The cheapest test is almost never the full commitment. Sometimes it is a small contract with a single customer. Sometimes it is a pilot run on existing capacity. Sometimes it is a six-month period of paid market research against a price point that has not yet been offered. The form varies. The principle does not: the cheapest test is the one that distinguishes between the assumption being right and being wrong, with the smallest possible expenditure of capital, time, and political weight.
What would the test tell us, and on what timeline? The test has to produce a binary, or at least a tightly-bounded range. If we see X by month nine, the central assumption holds and the full commitment is justified. If we see Y, the assumption is broken and we walk. The test that produces ambiguous results is the test that gets re-litigated forever; design accordingly.
The output of this exercise is, in our experience, almost always one of three things. Either the team designs a small bet that subsequently produces the information needed to size the big one. Or the team realizes that no small bet exists for this decision, in which case the full commitment is genuinely binary and the board now knows that explicitly. Or — and this is the most common, the most uncomfortable, and the most valuable — the team realizes that the case for the full commitment has not yet been earned, and that the small bet is what should be on the slide in place of the large one.
The minimum-information bet is the most under-bought option in corporate strategy. Companies that build the habit of buying it routinely outperform companies of comparable analytical sophistication, because the analytical sophistication has, in their case, been pointed at the load-bearing assumption rather than at the deck.
A worked example
Anonymized, but close to a real engagement we ran in 2024.
The proposal: a one-hundred-million-dollar acquisition of a software company adjacent to the buyer's core platform. The thesis: cross-sell the acquired product into the buyer's existing customer base, producing roughly twenty-five million in incremental ARR over three years.
The load-bearing assumption: that the buyer's existing customers would adopt the acquired product at a meaningfully higher rate than they were currently adopting it as a third-party integration.
The deck contained a sophisticated market sizing, a competitive analysis, an integration plan, and three case studies of analogous acquisitions. It did not contain a single piece of evidence on the load-bearing assumption.
The minimum-information bet, when we asked the team to design it: bundle the third-party product into the buyer's standard contract for six months, at a defined discount, with explicit measurement of the lift in attach rate. Cost: roughly four hundred thousand dollars in foregone margin, run on existing rails, no integration work required. Expected timeline to a clean read: five months.
The bet was made. The attach rate did not lift. The acquisition was not pursued. The four hundred thousand dollars saved, by a conservative reckoning, somewhere between forty and sixty million dollars of write-down. More importantly, it saved the eighteen months of organizational attention that the integration would have consumed, and the political cost of either continuing to defend or eventually unwinding the deal.
The team's view, in the post-mortem, was that the minimum-information bet had been institutionally easier than they expected. The proposal looked, in retrospect, like an obvious step. It had not been obvious in advance because the decision architecture had been pointed at the binary — buy or don't buy — rather than at the intermediate experiment that the binary depended on.
Where this discipline doesn't apply
Two honest caveats.
The minimum-information bet does not apply to decisions that are genuinely time-sensitive — where the option to make the small bet has a meaningful cost in foregone optionality on the large one. Some acquisitions, in particular, cannot be piloted because the target will be acquired by someone else within the pilot window. For these, the question becomes a different one: how much would you pay for the information the pilot would have produced, and is there any cheaper proxy available within the window? The answer is often yes, and the framing produces a different deck.
The minimum-information bet also does not apply to decisions where the small version of the bet would, itself, foreclose the option to make the large one. Some markets reward first movers heavily enough that announcing your intent through a pilot gives competitors the signal they need to respond. In these cases, the pilot has a strategic cost, and the analysis has to be done in private. The discipline still applies — the question becomes, what is the smallest experiment that does not give away your hand? — but the answer space is smaller.
What the board can do
A board that wants to install this discipline has one move available, and it is a powerful one: require, for every major commitment, a single page in the deck titled The Smaller Bet We Considered. The page describes the minimum-information bet that the team considered as an alternative to the full commitment, and the reasoning for not making that bet instead. If the team did not consider one, the deck is incomplete, and the vote is deferred.
The page is structurally cheap to require, structurally hard to fake, and structurally changes the conversation. The team, knowing the page will be required, designs its analysis differently — points it at the load-bearing assumption from the beginning rather than at the deck's overall persuasiveness. The board, reading the page, has a credible alternative against which to evaluate the proposed commitment. The discussion stops being binary. The conversation about confidence becomes, finally, the conversation about how much confidence the analysis can actually support.
We described, in a related essay, the broader shift from declarative to bet-framed strategy. The minimum-information bet is the operational tool that makes that shift bite at the moment of the commitment. The strategic plan can be written in the language of bets all year. If, at the moment forty million is on the table, the room reverts to the binary, the language has not done its job.
If the next nine-figure commitment in your calendar is being framed as a binary, and you cannot, on reading the deck, identify the page on which a smaller bet was considered, that is the right conversation to have before the vote — not after.
Writes about decision quality at Bayeseon. Reach the team at hello@bayeseon.com.