Sunk Costs and the Escalation of Commitment

The idea in plain words

Economics teaches that sunk costs, money already spent that cannot be recovered, should not affect decisions. Only future costs and future benefits matter. If a project needs another million to finish and will return half a million, you should stop, whether you have already spent ten million or nothing.

Almost nobody behaves this way. Having spent ten million, we feel the remaining million as the price of not wasting the ten. The past investment becomes a reason to continue. This is the sunk cost effect, and when it plays out over a series of decisions, each adding more resources to a failing course, psychologists call it escalation of commitment.

Where it comes from

In 1976 Barry Staw ran an experiment with business students. Each was asked to allocate research money to one of two divisions of a company. Later they learned how their chosen division had performed, and were asked to allocate more money. Students whose division had done badly gave it more money than those whose division had done well, and they did so especially when they had personally made the first choice. Responsibility for the original decision made people dig in.

In 1985 Hal Arkes and Catherine Blumer showed the effect in everyday choices. In one study, people who had paid full price for theatre season tickets attended more performances than those randomly given a discount, even though the money was equally spent in both cases. In another, people asked to imagine they had bought two ski trips for the same weekend by mistake, one cheaper but more enjoyable, often chose the more expensive one.

Arkes and Ayton later called a version of this the Concorde effect, after the Anglo-French supersonic airliner whose development continued long after it was clear it would not pay back. Staw and Jerry Ross argued in 1987 that escalation has several layers: psychological (self-justification), social (not wanting to look inconsistent), organisational (projects acquire teams and budgets that defend them) and project-specific (some projects really do pay only at the end).

What it does to the mind

The core is self-justification. Abandoning a project means admitting the original decision was wrong, and admitting it publicly. The person who started a project has the most to lose from stopping it, and is usually the person best placed to decide whether to stop. This is a structural conflict of interest built into most organisations.

Escalation is also fed by framing. Spending more on a failing project is felt as a chance to avoid a certain loss, and, as prospect theory predicts, people facing certain losses become willing to gamble. Each additional investment buys hope that the loss can still be avoided.

Finally, there is ambiguity. Real projects rarely fail cleanly. Each setback can be explained as temporary, each delay as the last. Without a clear stopping rule decided in advance, the evidence never quite becomes decisive.

Where it shows up

Iridium and Itanium are textbook cases: each was a vast, multi-year commitment whose market changed underneath it, and in each case the size of the investment made it harder to stop. Nokia’s long commitment to Symbian, discussed in the Burning Platform memo, shows the organisational layer: a platform with thousands of engineers and a proud history. Netflix’s decision to separate DVD from streaming in 2011, in the Qwikster memo, was in one sense the opposite: an attempt to stop investing attention in a declining business before sentiment made it impossible.

What it asks of someone deciding

Ask the zero-based question regularly: would we start this today? If the answer is no, the burden of proof shifts to continuing.

Set kill criteria before you start. Decide in advance what results, by what date, would lead you to stop, and write them down where they cannot be revised quietly.

Separate the decision to continue from the person who began. A fresh reviewer, without authorship to defend, sees the project as it is.

Respect the objection. Some valuable things look like failures for a long time. The difference between persistence and escalation is usually whether new evidence is being sought and honestly weighed, or explained away.