In 30 seconds
- Forecasts use what the team actually finishes, not hopes.
- Always give a range, ideally with a confidence level.
- Update the forecast every sprint as data comes in.
Three common methods
| Method | How |
|---|---|
| Velocity range | Remaining points divided by lowest and highest recent velocity |
| Throughput | Remaining items divided by items finished per week |
| Monte Carlo | Simulate thousands of futures from past throughput; read off 50% and 85% dates |
A quick example
120 points remain. Recent velocity ranges from 20 to 30 per sprint. Forecast: 4 to 6 sprints. Stakeholders hear a range they can plan around, not a single date that will probably slip.
Communicating it
- State the range and how confident you are
- Show what would move it, such as scope added
- Re-forecast every sprint, as the cone of uncertainty narrows
Myth vs reality
Myth · Forecasts are commitments.
Reality · A forecast is a best estimate; it changes with new data.
Myth · You need story points to forecast.
Reality · Item throughput often forecasts just as well.
Myth · Monte Carlo is too complex.
Reality · Free spreadsheets and tools make it a few minutes' work.
The AI angle
After a team adopts AI tools, its delivery history stops being a reliable guide. Re-baseline with a few sprints of new data and forecast in ranges until the pattern settles.
FAQ
How much history do we need?
Even three to five sprints or a few weeks of throughput helps.
What confidence level should we quote?
85% is a common choice for planning commitments.
How do we handle added scope?
Track it separately so its effect on the date is visible.