Why is one manager's estimate less reliable than the team's?

1. No error cancellation

Any estimate of a future outcome carries error. Some of it is systematic — optimism bias and the planning fallacy push nearly everyone the same way — and some is idiosyncratic: this person had a bad experience with a similar integration, that one is unusually cautious by temperament.

With one estimator, all of that error is in the number, and there's nothing to offset it. With six independent estimators, the idiosyncratic component partly cancels. Someone who over-weights the flaky test suite is balanced by someone who under-weights it.

Two honest caveats. The systematic component doesn't cancel — if the whole team is optimistic, the average is optimistic. And cancellation only happens if the estimates are actually independent, which is why private responses matter. Six people who all heard the tech lead say "we're fine" produce one estimate repeated six times.

2. No variance

This is the part most often overlooked, and it's arguably worth more than the accuracy improvement.

A single estimate is a point. Six estimates have a spread, and the spread carries information the average doesn't. Five people at 4 and one at 2 averages to roughly the same place as six people at 3.7 — and the first is far more interesting, because somebody has seen something the others haven't.

A manager's summary destroys this by construction. Even a diligent manager who has spoken to everyone produces one number, and their job in producing it is to reconcile the disagreement. The reconciliation is exactly the information you wanted.

A leader in our research described wanting precisely this: to know when "someone is seeing something the rest aren't." No single estimate can provide it.

3. Distance and exposure

The third reason is about who the estimator is.

Distance from the evidence. The engineer three weeks into an integration has met the problem. Her manager has heard a description of it, filtered through her judgement about how much to mention. A group project manager described the cost of that gap:

"The problem with being one step removed — if the PM had detected the thing sooner, it would have been helpful."

Exposure to consequences. The manager is usually the person accountable for the date. Reporting reduced confidence means reporting on their own project, to people who will hold them responsible. That's a real incentive pointing in a specific direction, and it operates on honest people.

A head of engineering described where this ends:

"I trusted what I was told by the dev manager. Two weeks before the summit, I was told they weren't going to deliver everything that was promised."

The dev manager wasn't lying. He was doing what people do when they're closest to the accountability and furthest from the evidence: waiting until certain, because being wrong in the pessimistic direction is expensive.

Does the team actually beat the expert?

The honest answer is: under conditions, and not always.

Aggregated judgement outperforms individual expert judgement when the judgements are diverse, independent and mechanically combined — the conditions are set out in can the wisdom of crowds be applied to project status. Where those hold, the aggregate is typically better than most individuals in the group, and often better than any single one.

Where it fails:

When the expert has information the team doesn't. A manager may know that the vendor contract is being renegotiated, or that scope is about to be cut. Their estimate can legitimately be better informed. The answer isn't to discard the aggregate — it's to notice the disagreement between the manager's view and the team's, which is itself informative.

When the team shares a blind spot. Everyone believing the same wrong thing produces a confident wrong average. Independence protects against influence within the team, not against being collectively misled.

When the question is technical rather than predictive. "Will this architecture scale?" is a question for whoever knows most, not for a vote. Aggregation suits questions where distributed local knowledge matters — and "will we hit this date?" is exactly that kind of question, because it depends on many small facts held by different people.

The objection worth taking seriously

One project manager in our research pushed back on the whole idea:

"Not comfortable crowdsourcing status. I want it done in a consistent way."

Worth answering rather than dismissing. The concern is that status becomes a vote rather than an assessment.

Two responses. The alternative isn't an objective measurement — it's one subjective assessment, from the person with the most reason to be optimistic. And consistency is a property of the question, not of the number of respondents: everyone answering the same fixed question on the same scale at the same cadence is more consistent than prose summaries, not less.

How Genchi does this

Genchi asks every team member the same question at a set cadence and aggregates the responses arithmetically, with no summarising step. What you see is the average, the spread of individual values, and the trend — so you get the error cancellation, the variance, and the direction, none of which a single estimate can provide.