Is a project delivery prediction a probability?

Two ways to predict a deadline

Statistical forecasts

Tools built on Monte Carlo simulation take a team's historical throughput (how many items it actually finishes per week), simulate the remaining work thousands of times, and report the results as percentiles: 50% likely by one date, 85% likely by a later one. The output is a genuine probability, grounded in what the team has really delivered.

The limits come from the same source. A statistical forecast assumes the future looks like the recent past. It needs enough history to sample from and a backlog that reflects the real scope. And it can only see what's in the tracker. A vendor quietly slipping, a requirement nobody understands, or a key engineer about to go on leave won't appear in the numbers until they've already cost you time.

The team's prediction

A confidence vote asks everyone on the project one question on a regular cadence: how confident are you the team will achieve their goal by the deadline? The answers are combined into a score per project and tracked over time.

Its strength is everything the tracker can't see. The people doing the work know when a dependency is shaky or a task is bigger than its ticket suggests, usually well before the plan of record admits it. And the prediction moves the moment someone's belief changes, rather than waiting for the work to slow down.

Its limit is that it's a judgement. People can be optimistic about their own work, and a whole team can be confidently wrong about something none of them can see.

What the score actually measures

Each vote is one of five answers: Doubtful, Unlikely, Uncertain, Likely or Certain. The labels describe confidence, not odds. Combined per project, the votes give a score that's comparable over time and across projects, but it isn't calibrated to a probability, so it shouldn't be read as one.

In practice, two things matter more than the absolute number. The first is the direction of the trend: a project that has drifted from a 4 to a 3 over three weeks is telling you more than one that has sat steadily at 3. The second is the spread of individual votes: a team that agrees at 3 is in a different position from a team where most people say 5 and one person says 1. That lone low vote is often the most valuable data point on the page. More on reading movement in why is a trend more useful than a status?

Why it's still a prediction

A prediction is a forward-looking claim about whether something will happen. That's exactly what each vote is: one person's judgement about whether the project will land, made before the outcome is known. Combining many independent judgements is a well-studied way of producing a better prediction than any single one.

Companies have tested this directly. At Siemens, an internal market predicted that a software project would fail to deliver on time, even as the traditional planning tools said the deadline could be met. Google and Hewlett-Packard ran similar experiments with their own employees. The evidence, and where it's weaker, is set out in does asking the team actually work?

What makes it a different kind of prediction is its source. A statistical forecast predicts from the work. A confidence vote predicts from the people doing it.

When the two disagree

If you run both, the most useful moments are the ones where they diverge.

The forecast says on track, the team is worried. This is usually the project to look at first. The throughput looks healthy because the hard part hasn't started yet, or because the problem lives somewhere the tracker can't see. The team is telling you something the numbers don't know.

The forecast says at risk, the team is confident. The team may know something the history doesn't reflect: a scope cut that hasn't hit the backlog, or a blocker that just cleared. Or they may be optimistic. Either way, it's worth a short conversation to find out which.

Neither signal is the right one by default. The disagreement is the signal: it tells you where to go and ask.

What it can't tell you

• A date or a percentage. If you need "what's our chance of finishing by November 30?", that's a job for a statistical forecast.
• What nobody can see. If no one on the project knows about a problem, no amount of asking will surface it.
• Why confidence is falling. The score tells you something is drifting. Comments, blockers and a conversation tell you what. See what do I do when the trend goes down?