How to use the agile delivery date forecaster
- Paste the number of items finished in each past week, one number per week. Weeks with 0 count.
- Enter how many items are left and the date the first forecast week starts.
- Choose the number of simulated futures and a seed.
- Read the weeks needed at each confidence level and the finish date at the end of that week.
Formula
draw one past week at random, with replacement, and add its items until the backlog is done
weeks for that future = number of weeks drawn
confidence level p = percentile of the simulated weeks, rounded up to a whole week
Repeating this 10,000 times shows how long the backlog takes when the future resembles a random reshuffling of the past. The dates count weeks of 7 days from the start date.
Twelve past weeks and 60 items
The past weeks finished 4, 6, 3, 5, 7, 2, 5, 6, 4, 0, 5 and 6 items, an average of 4.42 a week. A simple estimate of 60 / 4.42 is 14 weeks. Simulating 10,000 futures with seed 1 gives a median of 14 weeks, 15 weeks at 70 percent, 16 weeks at 85 percent and 17 weeks at 95 percent.
With a start date of Monday, November 2, 2026, the 85 percent finish falls by the end of Sunday, February 21, 2027, and the 95 percent finish by the end of Sunday, February 28, 2027. The fastest simulated future took 10 weeks and the slowest 22.
Why not just divide by the average
The average gives one date that has about a 50 percent chance of being met. Weeks vary, and a few slow weeks in a row push the finish out. Resampling the real weekly counts shows how much later the finish could be and how likely each date is.
The 85 percent row is a common planning choice because it leaves a margin without adding a lot of buffer.
When the forecast breaks
The method assumes the next weeks look like the weeks you entered. If the team grows, the type of work changes or the backlog grows, the history no longer fits. Counting items also assumes they are of roughly similar size.
Use recent history, at least 6 to 12 weeks, and refresh the forecast as new weeks arrive. For a plan of named tasks with dependencies, use the Monte Carlo Project Duration Simulator.
Assumptions and limits
- Each past week is an equally likely sample of a future week.
- The backlog does not grow during the forecast, and items are of similar size.
- A week is seven calendar days, and holidays are not skipped.
- The same seed always gives the same answer. Percentiles are rounded up to whole weeks.
- Between 5 and 1,000 past weeks and 100 to 100,000 simulated futures are accepted.
Frequently asked questions
How many past weeks should I use?
Use enough to include normal ups and downs, usually 6 to 12 weeks, but not so many that old conditions dominate. At least 5 weeks are needed here.
Can a week with no finished items be included?
Yes, and it should be. Zero weeks happen, and leaving them out would make the forecast too optimistic.
What does 85 percent mean?
In 85 percent of the simulated futures the backlog was finished within that many weeks. It is a statement about the simulation, not a guarantee about the real project.
Why must at least one week be above zero?
If every past week finished nothing, no backlog could ever be completed, and the simulation would never end. The page asks for real data instead.