Are We Really Bad at Estimating? Surprising Data Shows We’re Better Than You Think

Translator’s note: Over many years of experience I have seen teams become discouraged by poor estimates and even question the value of estimating. This article uses simple examples, data, and research to offer a reassuring conclusion, with some surprisingly interesting points.

The author, Mike Cohn, co‑founder of the Agile Alliance and Scrum Alliance and author of Scrum: The Art of Agile Development , provides an original translation with permission.

Are We Really Bad at Estimating?

Estimating is a contentious topic. Teams may avoid it for fear that managers will use the numbers against them. Stakeholders always want to know delivery dates, yet new or unfamiliar work makes accurate estimates difficult, leading some to wonder whether they should skip estimating altogether.

We’re Actually Pretty Good at Estimating (Some Things)

We are certainly bad at estimating some things, but we are quite adept at others. For example, I can estimate that writing a blog post will take about two hours, give or take a half hour, which is a useful plan for the afternoon. When I taught Certified ScrumMaster® courses, I estimated a typical room setup would take 45 minutes, based on extensive experience.

We routinely estimate everyday tasks—dinner, driving, grocery shopping—with reasonable accuracy because we are familiar with them. We are less skilled at estimating unfamiliar work.

Research supports this view. A review by University of Oslo professor and Simula Research Laboratory chief scientist Magne Jørgensen found most estimates fall within 20‑30% of actuals, and he did not observe a systematic tendency for software estimates to be too low.

The large number of time‑prediction failures throughout history may give the impression that our time‑prediction ability is very poor, but this is an unfair evaluation. Human ability to predict time usage is generally impressive and extremely useful, though it sometimes fails us.

One Answer Lies in the Projects We Never Start

Consider a boss who asks for an estimate before approving a project that will actually require 1,000 hours. The team estimates 500 hours, the boss approves, but the work ends up taking the full 1,000 hours, causing a delayed, memorable project.

In a parallel scenario the team estimates 1,500 hours. The boss, hearing the large number, cancels the project, so the over‑estimate never becomes visible. Under‑estimated projects are more likely to be approved, creating the perception that teams are always late.

Our Overconfidence Often Leads to Incorrect Estimates

Overconfidence is a major source of error. In an exercise I asked participants to give a range they were 90% confident would contain the correct answer to ten questions. Even when the range was wide, most people got the questions wrong.

For example, estimating Elvis Presley’s birth year yields a correct range of 1930‑1940, and the answer is 1935. However, for less familiar questions—such as iPad sales in 2019 or the number of athletes in the 2016 Olympics—people tend to give overly narrow ranges, reflecting overconfidence.

Can We Get Better? The Data Suggests It’s Possible (with Feedback)

When estimators see evidence of their overconfidence, they improve. In a software‑development study, programmers’ correct‑estimate rate rose from 64% on the first ten items to 70% on the next ten, and to 81% after further feedback.

Helping estimators recognize misplaced confidence encourages collaborative estimation practices. Those who believe their estimates are infallible are unlikely to engage constructively in estimation discussions.

What Do You Think?

Do you think the view that we are bad at estimating is unfair? How does your team improve its estimates? Share your thoughts in the comments.