Decision making · Article

Planning Fallacy: Why Everything Takes Longer and How to Estimate Better

Why your plans keep running late even when you know they usually do, what the classic studies found, and a five-step way to make estimates you can actually keep, using your own history and the outside view.

Planning Fallacy: Why Everything Takes Longer and How to Estimate Better

Lindiwe is a product designer at a small software company in Cape Town. On Monday her manager asks how long the new onboarding flow will take. She pictures the work: a few screens, a round of feedback, some polish. "Two weeks," she says, and means it.

Her colleague Sipho doesn't argue. He opens the team's project board and filters for the last three redesigns of similar size. One took four weeks. One took five. The checkout redesign took six, because legal wanted changes to the consent screen.

"Every one of those," Sipho says, "you called two weeks."

Lindiwe isn't bad at her job. She knows her past projects ran long. She just believes this one is different, which is exactly what she believed last time. That pattern has a name: the planning fallacy. This guide covers what it is, what the research found, why knowing about it doesn't fix it, and a five-step method for estimates you can actually keep.

What is the planning fallacy?

The planning fallacy is the tendency to underestimate how long a task will take, and often what it will cost, even when you know that similar tasks took longer than planned in the past. You make the forecast from the plan in your head, not from the record of how things actually went.

The key part is the second half. Being wrong about something new is normal. The planning fallacy is being wrong in the same direction, again and again, while holding the evidence that you usually are.

Psychologists Daniel Kahneman and Amos Tversky proposed the idea in 1979. Later, Dan Lovallo and Kahneman widened it in 2003 to cover the tendency to underestimate the time, costs and risks of future actions while overestimating their benefits. It's entry 13 in our guide to cognitive biases, and it often travels with overconfidence and optimism.

The writer Douglas Hofstadter captured it in a joke in his 1979 book Gödel, Escher, Bach, now called Hofstadter's law: "It always takes longer than you expect, even when you take into account Hofstadter's Law."

What the studies found

The thesis study

The best known experiment comes from the psychologists Roger Buehler, Dale Griffin and Michael Ross, published in 1994 in the Journal of Personality and Social Psychology. They asked 37 psychology students to predict when they would finish their senior theses.

The average prediction was 33.9 days. The students also gave a best case, "if everything went as well as it possibly could" (27.4 days on average), and a worst case, "if everything went as poorly as it possibly could" (48.6 days). The average thesis actually took 55.5 days. About 30 percent of the students finished within the time they predicted.

Read that again: the typical student took longer than their own worst case.

Even "99 percent sure" wasn't safe

In a follow-up, the same researchers asked students for three dates for personal projects: when they were 50 percent, 75 percent and 99 percent sure they would be done. Only 13 percent finished by their 50 percent date, 19 percent by their 75 percent date, and 45 percent by the date they were 99 percent sure about.

The tax return that is always late

A Canadian survey on tax returns found that people mailed their forms about a week later than they had predicted. They remembered being late in past years. They just expected to be quicker this time.

Outsiders see it differently

The bias applies mainly to your own tasks. When outside observers predict how long someone else will take, they tend to lean the other way and overestimate the time. So a second opinion is worth something, as long as it comes from someone who isn't doing the work.

Kahneman's own textbook

Kahneman told a story about himself that is the clearest illustration of the bias. In his book Thinking, Fast and Slow, he describes how he once persuaded officials in the Israeli Ministry of Education that high schools needed a course on judgment and decision making. He assembled a team, including Seymour Fox, then dean of the Hebrew University's School of Education and an expert in curriculum development, to design it and write a textbook.

Once the work was underway, Kahneman asked everyone to write down how long it would take to submit a finished draft to the ministry. The estimates, he wrote, "were narrowly centered around two years."

Then he asked Fox, who had written down an estimate like everyone else, how long similar teams had taken. About 40 percent of such teams never finished. Of those that did: "I cannot think of any group that finished in less than seven years," he said, "nor any that took more than ten."

The team carried on anyway. "The book was completed eight years later," Kahneman wrote. By then, the ministry's enthusiasm had faded, "and the textbook was never used."

Here's the part worth noticing: Fox had the right information the whole time. He just didn't use it until someone asked a different question.

Inside view vs outside view

Kahneman's lesson was about two ways of looking at a forecast.

Inside view Outside view
Starting point The details of your plan How similar projects actually turned out
Question it asks "What steps does this take?" "How long do things like this usually take?"
What it sees well Your specific situation Delays, surprises and things you didn't plan for
What it misses Everything not in the plan What's truly unusual about your case
Typical result Close to the best case Closer to what actually happens

The inside view is what everyone does naturally. You picture the steps, and the total feels solid because you can see every piece. But the delays aren't in the plan. Sick days, a stakeholder who goes quiet, the legal review: none of them appear when you imagine the work, but some of them appear every time.

The outside view starts somewhere else: with a group of similar past cases, called a reference class. As Kahneman put it, the base rate gives "a reasonable basis for a baseline prediction," which "should be the anchor for further adjustments." You start from what usually happens, then adjust for what's genuinely different about your case.

Debate it on Debatly“Should you trust your detailed plan or the average of past projects when they disagree?”

Why knowing about it doesn't fix it

You'd think learning about the planning fallacy would cure it. It rarely does. A few reasons:

  • The plan feels like evidence. A detailed plan is vivid and specific. Your past delays are vague memories. Vivid beats vague.
  • Past delays feel like exceptions. Buehler and his colleagues found that people tend to blame past delays on outside factors: the flu, the client, the broken laptop. Each one feels like one-off bad luck, so none of them seems to apply to the next project.
  • We plan for the best case. Asked how long something will take, people tend to imagine how it will go if things proceed as planned. Kahneman said his team's estimates were "closer to a best-case scenario than to a realistic assessment."
  • Social pressure. Saying "two weeks" in a meeting feels better than saying "five." One experiment found that when people made predictions anonymously, the optimistic bias disappeared.
  • Incentives. In big projects, a low estimate helps get approval. Sometimes that's strategy, not self-deception.

Planning fallacy examples

The bias shows up at every scale, from a Saturday errand to a national landmark.

Where The plan What usually happens Outside view question
Work projects "Two weeks for the redesign" Four to six weeks, like the last three How long did our last three similar projects take?
Home renovation "Six weeks for the bathroom" Delays from suppliers, permits and surprises in the walls How long did bathrooms like ours take for people we know?
School "I'll write the paper this weekend" It runs into Tuesday night When did I actually finish my last three papers?
Personal goals "I'll finish the course by the end of the month" Two months, with a gap in the middle How often have I finished a course on schedule?
Everyday "I'll leave at 7:05" Out the door at 7:30 What time did I actually leave on the last five workdays?

The Sydney Opera House

The classic large example is the Sydney Opera House. According to the Opera House's own history, the building was estimated to cost just 3.5 million pounds when construction began in 1959. By the middle of 1962, the estimate had reached 13.7 million pounds. It had been expected to be finished in 1963. It opened in 1973, a decade later, as a scaled-down version of the original design.

Big projects in general

The planning researcher Bent Flyvbjerg studies cost overruns in large infrastructure projects. Building on Kahneman's outside view, he developed reference class forecasting for practical use: forecasting a project from the actual results of comparable past projects. In June 2004, the UK Department for Transport published official guidance based on it for dealing with optimism bias in transport planning.

Even an outside view isn't magic. A 2004 review of a planned tram line in Edinburgh, using that method, raised the original £320 million forecast and estimated that a cost of £400 million gave an 80 percent level of confidence. The line opened three years late, in May 2014, at a final cost of £776 million. A base rate makes estimates less wrong. It doesn't make them certain.

Debate it on Debatly“Is it ever smart to commit to a deadline you know is optimistic?”

How to estimate better: a five-step method

You need your own history and a bit of honesty. This works for a work project, a renovation or a personal goal.

1. Make your usual estimate, and write it down

Don't skip this. Your inside view estimate is useful as a comparison, and writing it down stops you from quietly adjusting it later. Lindiwe's was two weeks.

2. Find a reference class

List three to ten past projects that resemble this one: similar size, similar kind of work, similar people. Your own history is the best source. If you don't have one, ask people who have done it: "How long did your bathroom actually take?" The question is about what happened, not what they planned.

3. Look at the spread, not just the average

Note the fastest, the slowest and the typical one. If the last three took four, five and six weeks, five is your baseline, and six is a realistic bad case. A best case of two weeks isn't on the list at all, which is worth noticing.

4. Adjust for what's truly different, a little

Now use the inside view, but only to move away from the baseline, not to replace it. Is this project genuinely smaller? Is a key dependency already solved? Be strict: "I'll be more focused this time" doesn't count, because you believed that last time too. Move the number by small steps and write down why.

5. Commit to a range and break the work down

Give a range, not a single date: "four to six weeks, most likely five." Then list every part of the task, including the boring ones (reviews, approvals, testing, the handoff). Researchers call this unpacking, and studies suggest it tends to raise estimates, because it brings up parts you skipped when you pictured the whole. Finally, decide when and where you'll start. Experiments on implementation intentions found that people who made specific "when and where" plans started sooner and had fewer interruptions, which reduced their optimistic bias.

Step Question Lindiwe's answer
1. Usual estimate What does my gut say? Two weeks
2. Reference class What are the most similar past projects? Last three redesigns: 4, 5 and 6 weeks
3. Spread Fastest, typical, slowest? 4, 5, 6 weeks
4. Adjust What's truly different? No legal review needed, so maybe a little faster
5. Range and breakdown What will I commit to? Four to five weeks, with research, two feedback rounds, developer handoff and QA listed

Worked example: Lindiwe's onboarding flow

Lindiwe went back to her manager the next morning with a different answer: "Four to five weeks, most likely four and a half. Here's why." She showed the three past projects and the breakdown.

Her manager was relieved to have a number with reasons behind it. They agreed to cut one screen from the first version to aim for four weeks, and to check progress at the end of week two.

At the week two check, user research had taken longer than planned, as it had on the checkout project. Because the estimate had room for that, nobody panicked. The flow shipped at the end of week five. That was later than her best guess, but inside her range, and for the first time in four redesigns nobody had to explain a missed deadline.

The lesson she took away wasn't "add a buffer." It was "look up what actually happened before you say a number out loud."

The planning fallacy's partner: sunk costs

The planning fallacy and the sunk cost fallacy are a common pair. The project runs late, money and months pile up, and because so much has been spent, nobody wants to stop. A check-in date set in advance, like Lindiwe's week two review, gives you a natural moment to ask whether to keep going, re-plan or cut scope.

How to practice this with AI

The planning fallacy is hard to catch on your own, because the person making the estimate is also the most optimistic person in the room. The research on observers points to the fix: get a view from someone who isn't doing the work. A debate gives you several.

On Debatly, several AI models and characters argue different sides of your question over several rounds, you can step in at any point, and a judge gives a verdict with reasoning.

  • Put your estimate on trial. In the debate room, describe the task, your estimate and your past projects. Ask: "Is four to five weeks realistic for this, or am I still being optimistic?"
  • Bring in a skeptic. Viktor, the skeptic, will question each assumption: who has to approve this, what happens if research runs long, why this project would be faster than the last three.
  • Ask for the failure story. Max, the risk analyst, can list what is most likely to slip and by how much. If you want the full version, combine it with a pre-mortem: imagine the deadline was missed by a month and ask what happened.
  • Read the verdict, then the reasoning. Judge Alex explains which arguments held up. If the winning side leaned on "you'll be more focused this time," treat that as a warning sign.

For the wider decision process around an estimate, see our playbook on how to make better decisions with AI.

Debate it on Debatly“Should a team pad every estimate by a fixed percentage, or build it from past projects?”

An honest limit

An AI model doesn't know your team or how long your last projects took unless you tell it, so any number it gives is a rough starting point, not your base rate. Models also tend to go along with what you seem to want (see why AI agrees with you), so ask outright for the case that it will take longer. Use the debate to find the assumptions and delays you missed, then check them against what really happened last time.

Look back before you look ahead

The planning fallacy isn't a lack of intelligence. It's what happens when a detailed plan in your head outweighs a vague memory of how things went. The fix is simple to describe and a little uncomfortable to do: before you say a number, look up how long the last few similar things actually took, and start there.

This week, pick one estimate you're about to give, at work or at home, and run the five steps. If it's a big one, run it as a debate on Debatly and let a skeptic and a risk analyst argue with your plan before reality does. Give it a try. Plans are on the pricing page.

Debate it on Debatly

“Planning Fallacy: Why Everything Takes Longer and How to Estimate Better”

FAQ

It's underestimating how long something will take, even when you know similar things took longer before. You base the estimate on your plan instead of on what actually happened last time. Saying a project will take two weeks when the last three took five is a typical example.
The psychologists Daniel Kahneman and Amos Tversky proposed it in 1979. In 1994, Roger Buehler, Dale Griffin and Michael Ross published a widely cited set of studies on it, including one in which students predicted about 34 days for their theses and took about 56 on average.
A student plans to write a paper over the weekend and is still working on it on Tuesday night. A homeowner expects a bathroom renovation to take six weeks and it takes twelve. On a large scale, the Sydney Opera House was expected to be finished in 1963 and opened in 1973.
Start from the outside view: find several similar past projects and see how long they actually took. Use that as your baseline, adjust only a little for what's genuinely different, and give a range instead of a single date. Breaking the task into all its parts and asking someone who isn't doing the work for their estimate also help.
It's a method that predicts the outcome of a planned project from the actual outcomes of similar past projects. You identify a group of comparable cases, look at the range of how they turned out, and place your project within that range. Bent Flyvbjerg developed it for practical planning, building on Kahneman's idea of the outside view.
#planning-fallacy #cognitive-biases #decision-making #estimation #project-planning