Why Averaging Planning Poker Estimates Is a Mistake

The Quiet Math Mistake in Most Planning Poker Sessions

Planning poker is meant to surface disagreement productively: everyone estimates independently, reveals at the same time, and discusses the gaps. But the step that turns those revealed numbers into a single "team estimate" often gets handled carelessly — someone eyeballs the numbers, or worse, a spreadsheet averages them. On a Fibonacci-like scale, that average can be actively misleading.

Why the Fibonacci Scale Breaks Averaging

Story point scales like 1, 2, 3, 5, 8, 13, 21 aren't evenly spaced by accident — the gaps widen deliberately as the numbers grow, because precision matters less for large, uncertain tasks than for small, well-understood ones. That's useful for estimating, but it's exactly what makes averaging unreliable. If three estimators say 3, 3, and 13, the average is 6.3 — a value nobody in the room actually chose, and one that sits awkwardly between two real Fibonacci steps.

Median: The More Honest Middle

The median — the middle value when estimates are sorted — doesn't have this problem. For 3, 3, 13, the median is 3: the value that most of the room actually converged on. For an even number of estimates, the median is the average of the two middle values, which stays close to what people actually said rather than getting pulled by one extreme outlier. This is why the median is the standard recommended approach for consolidating planning-poker rounds, not a simple mean.

When the Spread Itself Is the Signal

A tight cluster of estimates — say 5, 5, 8, 5 — tells you the team has a shared mental model of the task, even if the exact number needs a final nudge. A wide spread — say 2, 3, 13, 21 — tells you something more important: people are not estimating the same task. Maybe one estimator is factoring in a dependency others don't know about, or someone is thinking about a much larger scope. A useful rule of thumb: if the highest and lowest estimate are separated by more than one step on the Fibonacci scale, that gap is worth a two-minute conversation before locking in a number.

What To Do When Estimates Diverge

• Ask the extremes to explain first. The person who estimated highest and the person who estimated lowest usually know exactly why — surfacing that reasoning resolves the gap faster than re-voting blind.
• Check for hidden scope. Wide spreads often mean one estimator assumed extra work (tests, migrations, edge cases) that wasn't stated in the ticket.
• Re-vote after discussion, not before. A second round after the extremes explain their reasoning almost always converges tighter than the first.
• Don't force false consensus. If the team genuinely disagrees on scope, that's a sign the ticket needs to be split or clarified, not just re-voted until numbers align.

Bottom Line

A planning poker tool is only as trustworthy as its math. Median consensus plus an explicit spread check turns a five-minute vote into a genuinely useful signal about whether the team is actually aligned — not just whether the numbers happen to look close on a spreadsheet.