The worst decisions an organisation makes rarely look bad at the time. They look like progress. They are approved by sensible people, backed by a business case, and often they work as advertised. They fail later, somewhere else, and usually at someone else's expense.
That gap, between where a decision looks better and where its cost lands, is the subject of our booklet The Price of Better. Part One, the part that is free to read online, is built on four pieces of well-studied evidence. They are worth setting out here, with their caveats.
A road that made everyone later
In 1968 the mathematician Dietrich Braess showed that adding a road to a congested network can lengthen every driver's journey.1 Nobody errs. Each driver takes the route that is fastest for them given what everyone else is doing. The new road offers each a better choice, they all take it, and the network slows. In the textbook version, 4,000 drivers who each took 65 minutes before the new road each take 80 minutes after it, and no driver gains by switching back alone.
The reverse has been reported. Stuttgart in 1969, New York's 42nd Street on Earth Day 1990, and Seoul when it took down the Cheonggyecheon elevated highway in the early 2000s are the usual examples, where closing a road did not produce the gridlock people expected.2 It is fair to add the caveat. Reviews of road closures have found many cases of reduced traffic, and in many of them the saving is better explained by drivers changing how much they travel than by a true Braess effect.1 The paradox is real. It is also rarer than the anecdotes suggest. What survives is the design lesson: when every unit is free to choose the best option for itself, the total can be worse than if no one had been given the option.
A game in which nobody is foolish
The Beer Game was built at MIT's Sloan School from Jay Forrester's work on industrial dynamics, and Peter Senge told it in The Fifth Discipline. A retailer, wholesaler and brewery each see only their own stock and orders and may not talk. Demand steps up once and stays up. The result, played thousands of times by very different people, is shortage, backlog, panic orders and a glut. The players blame each other. Nobody has made a foolish decision.
John Sterman's 1989 experiment found that players underweight the supply line, the orders already placed but not yet received, and keep ordering for a shortage that is already on its way.3 The real-world version is the bullwhip effect. Lee, Padmanabhan and Whang described how at Procter & Gamble, babies used diapers at a steady rate while orders swung wildly further up the chain.4 The lesson is not that delay excuses every result. It is that blame aimed at people misses the structure that made the result likely.
Why the fashionable choice wins
An organisation that buys a technology because its peers have bought it is not being foolish. It is responding to forces that have been studied for fifty years.
- Imitation under uncertainty. When nobody knows what works, copying respected peers is the safest way to look competent. DiMaggio and Powell called it mimetic isomorphism: the aim is legitimacy, and it gets mistaken for effectiveness.5
- The cascade. Each adopter takes the previous adopter's choice as evidence and sets aside their own doubts. After a few rounds, nobody in the queue is acting on information of their own.6
- The career. The sponsor is rewarded for launching and rarely penalised when it fails to pay back. Being the executive who did not adopt is riskier than adopting and failing alongside everyone else.
- The solution looking for a problem. In Cohen, March and Olsen's garbage can model, solutions arrive first and go looking for problems to attach to.7 "We need an AI use case" is that sentence in modern form.
The parity premium
Most fashionable adoptions do not damage the organisation. The assistant works, the dashboard loads, the platform runs. That is why they are hard to challenge: nothing has visibly gone wrong. What went wrong shows in what the organisation did not do while it was busy: the data clean-up that was deferred, the pricing error left another year, the managers on the steering committee with less time for their teams.
In the booklet we call the difference the parity premium: what an organisation pays, in money, attention and foregone options, to own the same thing as its peers, minus the value that can actually be traced to owning it. Some parity is legitimate. A regulator may require a capability, or a professional norm may make it the defensible choice. The failure is adopting without knowing which of those reasons applies.
The question that matters is not "will this make us worse?" It is "what did we stop doing to make room for it, and who decided that?"