Every project has two progress curves: a planned/target and an actual curve. Each week it can be measured how fast the project is developing compared to the plan: faster, slower, or on track?
​

The forecast uses this past development to answer: If the team continues working as it has been — when will the project be finished?
​

Instead of giving a single answer (which would only be an estimate), the model calculates 50,000 simulated future trajectories, each of which represents a plausible development of the project from today.

Each week is compared to see how much progress was actually achieved relative to what was expected according to the plan at that point in time. This produces a performance ratio for each week:

The model pools all these weekly ratio values together. Planned downtime (public holidays, winter break) is also taken into account: during these weeks the simulation — just like the plan — remains unchanged.

Starting from the current actual progress, the model simulates the further course week by week. For each week, one of the historical performance ratios is randomly selected and multiplied by the planned progress.

Since the selection is random, each simulation plays out slightly differently: some faster, some slower.

After 50,000 runs, it is evaluated when each simulation is completed. This produces a distribution of possible completion dates.

The forecast fan curve shows the range of possible developments, based on past project performance.

The wider the fan, the less predictable the project development has been in the past.

The blue dashed line shows the median of the forecast: at any given point in time there is a 50% probability that the project is above or below it.

The forecast is only displayed once at least 8 weeks of progress data are available. With fewer data points, no forecast is generated.

This minimum ensures that the forecast is based on sufficient history to be reliable.