The Planning Fallacy: Why Tasks Take Longer Than Planned
Students predicted 34 days and took 56. What the planning fallacy research shows, and how the outside view produces a more honest estimate.
Most time estimates fail in the same direction. People predict less time than a task will take, even when they know similar tasks ran late before. Psychologists call this the planning fallacy, and the proposed fix is not to try harder or add a vague cushion. It is to change the question you ask before you estimate.
Daniel Kahneman and Amos Tversky proposed the planning fallacy in 1979. In 2003, Dan Lovallo and Kahneman expanded the definition: people tend to underestimate the time, costs, and risks of a future action while overestimating its benefits.
The thesis study that made the pattern hard to ignore
The clearest demonstration comes from Roger Buehler, Dale Griffin, and Michael Ross, published in the Journal of Personality and Social Psychology in 1994. The researchers asked 37 psychology students to estimate how long they would take to finish their senior theses.
The numbers tell the story plainly:
- Average prediction: 33.9 days
- Average prediction if everything went as well as possible: 27.4 days
- Average prediction if everything went as poorly as possible: 48.6 days
- Average actual completion time: 55.5 days
Only about 30% of the students finished by the time they had predicted. The average student finished later than the average worst-case estimate. In other words, even the pessimistic version of the plan was optimistic.
A related study in the same research program asked students for dates by which they were 50%, 75%, and 99% sure they would finish a personal academic project. Only 13% finished by their 50% date, 19% by their 75% date, and 45% by their 99% date. A date someone treats as almost certain should be missed rarely. Here it was missed more than half the time.
This was not a case of forgetting their own history. In a 1997 survey, Canadian taxpayers mailed their tax forms about a week later than predicted, with no misconceptions about their past filing record. People can hold a pessimistic view of their past predictions in general and an optimistic view of the specific prediction in front of them.
At larger scales, the pattern grows more expensive. Bent Flyvbjerg’s 2003 analysis covered 258 transport projects across 20 countries. Average cost overruns were 28% for roads, 34% for bridges and tunnels, and 45% for rail, and forecasts had not improved over the prior 70 years in that data set. The Sydney Opera House, cited in the 1994 paper, was expected to open in 1963 for $7 million. A scaled-down version opened in 1973 for $102 million.
Inside view, outside view
Kahneman and Tversky explained the gap with two ways of looking at a task.
The inside view focuses on the specific case. You list the steps in this project, picture them going in order, and judge the plan on its own terms. The steps you can picture get counted. Interruptions, waiting, rework, and the ordinary friction of other commitments mostly do not, because they are not part of the story of the task itself.
The outside view treats the task as one member of a class. Instead of asking how long this project should take if it goes well, you ask how long similar projects actually took, for you or for other people. That distribution, sometimes called the reference class, becomes the starting point for the estimate.
Buehler, Griffin, and Ross found direct evidence for this split. When participants explained their predictions aloud, they focused mainly on future plans, not on past experience. When they did recall past misses, they tended to explain them with external, temporary, and specific causes: that semester was unusual, that one time was different. Together, those explanations make the past seem irrelevant to the next estimate.
The contrast with observers is telling. In Study 5 of the 1994 paper, observers asked to predict someone else’s completion time overestimated it and made greater use of that person’s past experience. Without a hopeful plan of their own to defend, they reached for the base rate.
Why knowing about the bias is not enough
A common response is to assume that awareness fixes the problem. The later research is less reassuring.
Rebecca Friesdorf tested this question in her 2020 doctoral dissertation at Wilfrid Laurier University, supervised by Buehler. In Studies 1 and 2, with 322 participants, people reported on average that they had finished past projects slightly later than predicted. That awareness did not lead to less optimistic predictions for a current project. Knowing you usually run late, stated as a general belief, is different from using a specific past time when you make a specific new estimate.
What worked better in Buehler’s original work was a tighter connection. In Study 4 of the 1994 paper, the optimistic bias was eliminated for participants who were instructed to connect relevant past experiences directly to their predictions: recall a similar task, recall when they expected to finish it, and recall when they actually finished it, then use that link to judge the current task. Merely recalling the past in a loose way did not have the same effect.
Friesdorf found a similar pattern for reference class forecasting, the formal version of the outside view. In Studies 3 through 6, with 1,425 participants, instructions to recall past completion times using reference class forecasting produced less optimistically biased predictions in three of four studies. The effect was clearest when past times were recalled in relation to the original predictions for those tasks, not just in relation to a deadline. The result was promising rather than uniform, which fits the broader picture: the outside view helps when it is applied concretely, and helps less when it stays abstract.
A second, simpler method has support as well. Research by Justin Kruger and Matt Evans, published in 2004, found that unpacking a task into its component steps produces longer estimates and reduces the fallacy. Listing the sub-steps surfaces work that a single overall judgment skips. Unpacking alone is not a complete correction, because each sub-step can still be estimated optimistically, but it makes hidden work visible before the work starts.
A practical way to estimate
For an everyday project, the research points to a short procedure rather than a personality change.
First, write down your inside-view estimate, then set it aside. Next, name a reference class you can check: the last three times you wrote a similar report, renovated a similar room, or shipped a similar feature. Write the predicted and actual times for those cases, if you have them.
Then start from the typical actual time in that class, not from your new plan. Adjust only for a difference you can name and defend, such as a step that is genuinely absent this time. Finally, unpack the task into sub-steps and check whether the total still fits the outside-view estimate. If the unpacked list implies more time, the longer figure deserves more weight.
Two habits support this. Keep a brief record of estimates and actuals for recurring work, because memory of past durations is itself unreliable. And ask someone not invested in the plan for their estimate. Observers in the research leaned toward pessimism, so treat their figure as the other end of a range. A range built from a base rate is more useful than a single confident date.
None of this removes uncertainty. A reference class can be too broad or too narrow, and genuinely new tasks remain hard to estimate. The claim in this literature is narrower: for tasks that resemble something done before, the distribution of past outcomes is usually a better anchor than a fresh plan built from intentions.
Bottom line
The planning fallacy is a specific, documented pattern: in the thesis study, a 33.9-day prediction became 55.5 days in practice, and even worst-case estimates fell short. The cause is not carelessness. It is a focus on the plan for this task at the expense of evidence from similar tasks. Start your next estimate from what similar work actually took, connect that record to the new task, and unpack the steps before you commit to a date. When an estimate still goes wrong, the durable fix is usually a routine rather than more resolve; how habit loops work explains the cue-and-friction side of that.