The money is already gone.

Two pieces of paper are held together by a binder clip. On top is a funding request for the third tranche of a project: four million dollars to finish what has been started. Underneath, clipped to it as supporting evidence, is a paid invoice for the first two tranches, nine million dollars, stamped and settled. The pair in this essay is invented, but anyone who has sat on an investment committee has seen it. The invoice is meant to make the request feel safer. It proves that a payment occurred. It says nothing at all about whether the next four million will earn anything.

That gap is the subject of this essay. My argument is that a decision to continue should compare what happens next under each available choice: continue, stop, or change course. Two kinds of money appear on the committee's table. One is money already spent, which cannot be recovered whatever the committee decides. The other is future costs and benefits that change with the choice: what it will take to finish, what the finished thing will be worth, what can be salvaged if the work stops, what it will cost to walk away. Only the second kind can tell the committee what to do. The first can tell it something about how it got here.

This is not an argument that persistence is foolish, or that history is irrelevant. Past spending often leaves behind things that matter a great deal going forward: half-built assets, signed contracts, hard-won knowledge, people who learned the work. Those are real, and they belong in the comparison. Nor is it an argument against reviewing the original decision; that review is valuable. The claim is narrower. Nine million dollars of spending, considered as spending, is not a reason to spend four million more, and a funding request that leans on it has not yet made its case.

Two questions on one page

Part of the trouble is that a funding request usually tries to answer two different questions at once. The first is about accountability: was the original decision sound, given what was known when it was made? The second is about allocation: given where things stand today, what should we do with the next dollar? Both questions are legitimate. But they have different evidence, different audiences and, most importantly, different stakes for the people answering them.

The person presenting the request is often the person who made the original case. For that person, stopping the project can feel like a verdict on the first decision, so the allocation question quietly becomes a way of answering the accountability one. In 1976 the organizational psychologist Barry Staw tested exactly this. He gave 240 business students a case in which they allocated research money to one of two divisions of a company, and then, after five years of simulated results, asked them to divide a further 20 million dollars between the two. Students who had personally chosen a division that then performed badly put an average of 13.07 million dollars back into it, more than in any of the other conditions. Staw called the pattern escalating commitment, and his explanation was self-justification: people invest further to show that the first choice was not a mistake.¹

The practical lesson is to put the two questions on separate pages. The review of the original decision can be thorough and even uncomfortable, and it can be done by someone other than the sponsor. The decision about the next tranche should be made as if the project had been handed to the committee this morning, complete with whatever it has built and whatever it owes.

What the experiments show

The best-known demonstrations come from Hal Arkes and Catherine Blumer, who in 1985 described the sunk cost effect as a greater tendency to continue an endeavor once money, effort or time has been invested in it.² In one problem, subjects imagined holding a 100 dollar ticket for a ski weekend in Michigan and a 50 dollar ticket for the same weekend in Wisconsin, which they had been told they would enjoy more. Neither ticket could be sold or returned. Only 46 percent chose Wisconsin, the trip they preferred.³

A second problem is closer to a boardroom. Subjects played the president of an airline that had spent 9 million of a 10 million dollar budget developing a plane invisible to radar, when a competitor began selling a faster, cheaper plane that did the same thing. Asked whether to spend the last million to finish, 85 percent said yes. Another group was asked whether to spend one million dollars on the same plane, with the same competitor already on the market, but with nothing spent so far. Only 17 percent said yes. The future consequences of spending the million were identical in the two versions; the only difference was the nine million in the past.³

Arkes and Blumer also ran a field study. Customers buying season tickets to a university theater series were randomly given either the normal price or one of two discounts. Those who had paid full price attended more performances in the first half of the season, although the plays, the seats and the evenings were the same for everyone.² Their interpretation was that people dislike appearing to waste what they have paid for. These were students and theatergoers, not executives, and the studies are evidence of a mechanism, not an estimate of how often real projects are continued for the wrong reason. But the mechanism does not need a large budget to work, and nothing about a boardroom switches it off.

Two projects with the same future

Here is the test I find most useful. Imagine two projects. One has spent nine million dollars; the other has spent one million. In every other respect they are identical from today: four million dollars to finish, the same value when finished, the same amount recoverable if the work stops, the same contracts to unwind. Should the committee's decision differ between them?

If the answer is yes, the committee is letting the past decide. If the answer is no, then the nine million cannot be doing any work in the decision, and it can be moved off the comparison and into the accountability review. The worksheet below makes the test mechanical. Move the spent-to-date line and watch the comparison stay put; then move any of the forward lines and watch it change.

At the starting values, finishing the project as planned is worth one million dollars more than it costs, stopping recovers half a million, and a smaller redesign that reuses what has been built comes out ahead of both. None of these numbers describes a real project. The point is the procedure: every line that can change the answer is a line about the future.

Where the past still counts

The clean version of the rule invites a fair objection: surely the nine million bought something. Usually it did, and this is where the hard cases live. What it bought counts, but it counts as an asset or an obligation in the present, not as a payment in the past.

Partly finished work. A half-built system, a trained team or a licensed design has value that depends on what happens next. If a redesign can reuse most of it, that lowers the redesign's cost. If stopping lets the components be sold or redeployed, that raises the value of stopping. Either way it enters the comparison as a forward number.

Exit costs. Cancellation fees, severance, penalties for breaking a lease and the cost of unwinding a migration are future cash that depends on the choice, so they belong on the stop line. They are often larger than people expect, and leaving them out biases the comparison toward stopping.

Uncertainty and the option to learn. The future value of a project is a forecast. Sometimes the best choice is neither to finish nor to stop but to spend a small amount to learn something that changes the forecast: a pilot, a technical spike, a customer test. A pause of that kind has its own cost and its own value, and it can be the right answer when the other two are close.

History as evidence. The past does carry information. A project that has run 50 percent over budget for two years is telling the committee something about how much the remaining work is likely to cost. That is a legitimate use of history: as evidence about the future, not as a debt owed to the past. The economists Preston McAfee, Hugo Mialon and Sue Mialon have argued that much behavior labeled the sunk cost fallacy can be rational once information, reputation and financial or time constraints are taken into account.⁴ Their argument is a useful corrective, and it fits the rule here rather than overturning it. Each of their reasons is a consequence that lies ahead of the decision.

Objections, considered

Stopping can destroy something valuable. It can. A project abandoned at 90 percent may forfeit a capability that would have been cheap to complete, and a team disbanded may be expensive to reassemble. But those are forward costs of stopping, and the worksheet has a line for them. The objection is an argument for counting carefully, not for counting the past.

Commitments and reputation matter. They do. An organization that walks away from partners whenever a forecast dips will find fewer partners, and a public commitment broken has a cost in trust. These are real consequences, and they should be written down and estimated like any other. What does not belong on the list is the sponsor's reputation for having been right. That is a cost to a person, and it is precisely the pressure Staw found driving escalation.¹

Calling everything sunk becomes an excuse. This is the most serious objection. A team that has overspent can use the language of sunk costs to wave away the overrun: the money is gone, so let us not dwell on it. That is the opposite of the point. The accountability review should dwell on it, and should feed what it learns into the forecast for the remaining work. The discipline is to keep the two questions apart, not to answer only the convenient one.

A paid invoice is a record of the past, not a reason for the future.

Before you approve the next tranche

• State the remaining cost. What it will take from today to finish, with the evidence from the project's own history about how reliable that estimate is likely to be.
• Name the alternatives. Stopping, pausing to learn, and at least one redesign, each with its own forward cost and value, including exit costs and what can be reused.
• Show the incremental benefit. What the organization gains by continuing compared with the best alternative, not compared with doing nothing.
• Say what would change the recommendation. The evidence that, if it arrived next month, would lead the sponsor to propose stopping.

Back to the binder clip. There is nothing wrong with the invoice; it is an accurate record of nine million dollars well or badly spent, and it belongs in the file for the review of how the project got here. But it should come off the funding request. What remains is a single page with five blank fields, and the four million dollars depends entirely on what gets written in them.