Estimates that ignore baseline frequency
People underestimate time to completion even when they have done similar tasks before and know those tasks ran over. A project that took three weeks last time is estimated at two weeks. The fallacy persists even after multiple overruns. The person estimating knows intuitively that projects run late but still predicts early, as if this project is the exception.
The mechanism is a focus on the specific task at hand, often called an inside view. You think through the steps, estimate each step, and sum them. But you ignore (or heavily discount) the outside view: how long did the base class of projects actually take? How many unexpected obstacles came up last time? What interruptions happened? Your specific estimate ignores these patterns because they feel like distractions from the analysis, yet they are the most reliable predictors.
Correcting through reference class thinking
The antidote is reference class forecasting: before estimating, ask what happened with similar past projects and anchor your estimate there. If the average similar project took six weeks, your estimate should start at six weeks unless you have specific reasons this project is smaller or simpler. Then adjust upward, not downward, for added uncertainty and past underruns.
This requires separating estimation from planning. Planning focuses on the specific task. Forecasting focuses on the base rate. The two should work together: use the reference class to set your time bound, then plan the work to fit that bound. The planning might reveal the work does not fit six weeks comfortably, which means the project is bigger than the reference class or more complex, information you need before committing to a deadline.