Psychology and biases
The planning fallacy
People predict their own projects will finish sooner and cost less than almost every comparable past project did.
Daniel Kahneman and Amos Tversky coined the term in a 1979 paper on intuitive prediction; Roger Buehler, Dale Griffin and Michael Ross ran the first direct experiments (1994).
You know your last three projects ran late, and you still believe this one will land on the date you just wrote down. That is the planning fallacy. It is not general optimism: the same people predict other people's timelines reasonably well, and they can recite their own track record if asked. Kahneman and Tversky's explanation is that a plan is a story about the future, and an estimate built from the story is a best case, because the story contains no illness, no blocked dependency and no change of scope. Buehler, Griffin and Ross showed this with think-aloud protocols: people forecasting a task talk almost entirely about the steps ahead and almost never about how long similar things took before.
The fix is to change what you estimate from. Kahneman calls the plan-based estimate the inside view and the alternative the outside view: find the class of similar past projects, start from how they went, then adjust. Flyvbjerg turned this into reference class forecasting for public infrastructure, and his database gives the base rates. Nine out of ten megaprojects run over budget. The Channel Tunnel came in 80 percent over in real terms, Boston's Big Dig 220 percent, and the Sydney Opera House 1,400 percent. Rail projects in his sample average a 44.7 percent cost overrun together with a 51.4 percent demand shortfall. Against numbers like these, an estimate built from your own schedule is a guess about a best case.
Two cautions. First, the fallacy is partly incentive, not only cognition: Flyvbjerg points to the tactical use of low estimates by project advocates, because the cheapest-looking plan wins the funding. Correcting your own bias does not fix a tender process that rewards it. Second, the outside view needs a real reference class. If your project is genuinely unlike anything in the data, a base rate from the wrong class is a false comfort. In practice the useful move is modest: write down the predicted date, write down what happened to the last five similar things, and ask what would have to be true for this one to differ.
Example
In 1994 Roger Buehler, Dale Griffin and Michael Ross asked 37 psychology students at the University of Waterloo, all in the final semester of an honours thesis, to predict when they would submit. Their best estimates averaged 33.9 days; the theses actually took 55.5 days on average. Only 29.7 percent finished by the date they had predicted, and only 10.8 percent met their 'if everything went as well as it possibly could' date. Fewer than half (48.7 percent) finished even by their worst-case date.
Where it applies
Before committing to a date, list the last five comparable tasks and how late each ran, then estimate from that list rather than from your plan. For a data migration or a model fine-tuning run, keep the wall-clock times of previous runs in a file and look at the distribution, not at the one that went well.
Think about it
Pick the project you are most confident about finishing on time: how did your last three projects of that kind actually go?
Applied it to something real? Log it.
Test yourselfWhy does estimating from your own plan produce a best case?
A plan is a story of the steps ahead and contains none of the problems that will happen; estimating from similar past projects (the outside view) puts them back in.
Related models
- Pre-mortemBefore you start, imagine the project has already failed and ask everyone to explain why.
- Probabilistic thinkingThink in odds rather than certainties: estimate how likely each outcome is, and judge a decision by those odds rather than by the single result that happened.
- Margin of safetyLeave a wide gap between what you expect to need and what you build or pay, so that your errors and bad luck do not sink you.
- AnchoringAny number you see before estimating pulls your estimate towards it, even when you know the number is random.
- Redundancy and slackSpare parts and spare capacity look like waste until the day they are the only thing between a fault and a failure.
Sources and further reading
- Roger Buehler, Dale Griffin and Michael Ross, Exploring the Planning Fallacy: Why People Underestimate Their Task Completion Times, Journal of Personality and Social Psychology 67(3) (1994)
- Bent Flyvbjerg, What You Should Know About Megaprojects and Why: An Overview, Project Management Journal 45(2) (2014)