You said the report would take two hours. You remember saying it, because you said it out loud, to a colleague, with confidence. That was Tuesday. It is now Thursday afternoon, the report is maybe two-thirds done, and the strangest part is not that you were wrong — it's that you've been wrong in exactly this way dozens of times before, and it changed nothing. Next week you will estimate something else, and you will estimate it the same way: cleanly, optimistically, and too low.
This is not a personal failing. It is one of the most reliable findings in the psychology of judgment, and it has a name: the planning fallacy.
A Bias With a Long Paper Trail
The term comes from Daniel Kahneman and Amos Tversky, who described the pattern in a 1979 paper on intuitive prediction: people systematically underestimate how long their own projects will take, even when they know — in the abstract — that similar projects have run long before.
The cleanest demonstration came later, in a 1994 series of studies by Roger Buehler, Dale Griffin, and Michael Ross. They asked university students working on their honors theses to predict when they would finish. The students' average estimate was about 34 days. The actual average was around 55. Fewer than a third finished by the date they had named. And here is the detail that should make you sit down: when the researchers asked students to predict their finish date assuming everything went as badly as it possibly could, even those doom-laden estimates still fell short of how long the theses actually took.
One more twist from that research program: people are much better at predicting other people's timelines. Hand someone a description of a stranger's project and they'll forecast delays, obstacles, ordinary human slippage. Ask them about their own project and all of that evidence quietly leaves the room. The bias is not about ignorance. It is about perspective.
Why Experience Doesn't Cure It
If underestimating were simply a lack of data, twenty years of missed deadlines would fix it. It doesn't, and the reason is the distinction Kahneman later made famous: the inside view versus the outside view.
When you estimate a task from the inside, you build a little mental movie of doing it. Open the document, pull the numbers, write the sections, done. The movie is a plan, and plans are best-case scenarios by construction — they contain the steps you intend to take and none of the things that will actually happen. The movie does not include the meeting that runs over, the file that won't export, the afternoon you lose to a colleague's emergency, or the section that turns out to require a decision you don't have the authority to make. No specific obstacle is likely. Some obstacle is nearly certain. The inside view has no way to represent that.
Memory conspires too. Buehler and his colleagues found that when people recall past projects that ran long, they tend to explain each overrun as an exception — the flu, the surprise client request, the one weird bug. Every delay gets a story, and a delay with a story doesn't feel like evidence about the future. So the past gets filed under "unusual circumstances" and the next estimate starts fresh, unburdened by history.
This is why the planning fallacy is oddly egalitarian. It catches optimists and pessimists, novices and veterans. You can be gloomy about life in general and still be sunny about the specific movie playing in your head.
The Outside View: The Fix That Actually Works
The correction Kahneman and Tversky proposed is almost insultingly simple: stop asking how will this task go and start asking how long do tasks like this take.
That second question forces you out of the mental movie and into your own track record. The last three reports: how long, really? Not how long they should have taken — how long they took, door to door, including the interruptions and the stalls. That number is your reference class, and it will almost always be larger and more accurate than anything the inside view produces.
This approach — reference class forecasting — has been developed most rigorously by the planning researcher Bent Flyvbjerg, who studies why large infrastructure projects blow through budgets and schedules with such monotonous regularity. His prescription for billion-dollar rail lines is the same one that works for your Thursday report: locate similar past projects, take their actual outcomes as your baseline, and adjust only if you have concrete reasons to believe this case differs. The inside view gets a vote, not a veto.
The practical version for knowledge work: before estimating anything that matters, name one or two past tasks that resemble it and write down what they actually cost you. If the honest answer is "the last thing like this took two days," then "two hours" is not an estimate. It is a wish wearing an estimate's clothes.
Unpack the Task Before You Price It
A second evidence-backed correction: break the task into its parts before you estimate, not after. In studies published in 2004, Justin Kruger and Matt Evans found that asking people to unpack a task into its component steps before predicting its duration led to longer, more accurate estimates. The mental movie, it turns out, skips scenes. Forced to list them — draft, revise, get feedback, incorporate feedback, format, send — you remember that the task is not one action but seven, and several of them involve waiting on other people.
Unpacking also exposes the estimate's softest spot: dependencies. Anything that requires another person's reply, approval, or calendar is not a task you control; it is a task plus a lottery. When a step depends on someone else, your reference class should include how long that person usually takes, which is a number the inside view never thinks to consult.
Plan for the Day You'll Actually Have
There is one more quiet assumption inside every optimistic estimate: that the hours in your plan are hours you will actually get. "This will take four hours" silently becomes "this will take four consecutive, undisturbed hours," and those are different currencies with a brutal exchange rate. A four-hour task scattered across a day of meetings and messages does not take four hours. It takes four hours of touch time plus the cost of every restart, and it often spills into tomorrow — which is exactly the overrun you'll later explain away as an exception.
So the last correction is to estimate in realistic hours: look at tomorrow's calendar, subtract the meetings, subtract the predictable interruptions, and see what's genuinely left. Most people find the honest answer is two or three deep hours a day, not eight. Sizing your plans to that number feels like pessimism for about a week. Then things start finishing on time, and it starts feeling like the first accurate map you've ever used.
Seeing Your Own Reference Class
The planning fallacy survives because the evidence against it is scattered — a missed deadline here, a long Thursday there, each one explained and forgotten. The outside view only works if you can actually see your track record: how long focused work really takes you, how much uninterrupted time a day really contains, what happened to the last ten estimates you made. That's the quiet case for keeping a record of your attention rather than relying on memory, which we've seen is a motivated witness. Reclaim exists for exactly this — it guards your focus time so the hours you planned are hours you get, and it shows you where your time actually went, so your next estimate can be built from evidence instead of optimism. If you're tired of being surprised by your own weeks, you can start reclaiming them at reclaim.lumenlabs.works.