Learn & Understand

The Flaw of Averages: Why a Single Release Date Misleads

The release forecast calculator takes your remaining backlog, divides by average velocity, and returns a number of sprints, and from there, a single confident date. That date is useful, but it conceals a statistical trap so common it has been dubbed "the flaw of averages": plans built on a single average value are, on average, wrong. A forecast that produces one date implies a certainty the underlying uncertainty simply does not support, and treating it as a promise invites trouble.

Why One Number Is the Wrong Answer

Velocity is not a constant; it fluctuates sprint to sprint with the mix of work, interruptions, illness, and discovery. Using its average to project a date effectively assumes every future sprint will be an average sprint, which none of them will be. The result is a single date sitting at the middle of a distribution of possible outcomes, roughly a coin-flip proposition, as likely to be missed as met. Communicated as "we'll ship on the 14th," it is quietly a fifty-fifty bet dressed up as a commitment.

The Flaw of Averages

The broader principle, popularized under the phrase "the flaw of averages," is that plugging a single average input into a calculation rarely gives you the average output when the inputs vary, and it always hides the spread. A river that is on average four feet deep can still drown someone at its deepest point. A project that on average takes ten sprints can, with entirely normal variation, take eight or fourteen, and the single-point forecast erases exactly the range a decision-maker most needs to see.

Forecasting as a Range

The honest fix is to forecast a range rather than a point. The simplest version reruns the calculation with a pessimistic and an optimistic velocity to bracket the likely window. The more sophisticated version, Monte Carlo simulation, runs the project thousands of times using the actual spread of historical velocities and reports dates with probabilities attached, "eighty-five percent likely by the 21st." Both replace a misleading single date with a spread that reflects reality, and both let stakeholders choose how much confidence they want to commit to.

Using the Single Date Responsibly

None of this makes the calculator's output useless, it makes it a starting point. The average-based date is a reasonable central estimate; the error is stopping there and treating it as a guarantee. Present it with an explicit hedge, pad it toward the pessimistic end when the commitment is firm, and remember that the backlog it is dividing is only today's known work. New work discovered later moves the whole distribution later, which is one more reason to quote a range and revisit it often.

The velocity feeding this forecast is derived in the Sprint Velocity Calculator; to protect a committed date against this variability, see the Project Buffer Size Calculator.