Parallel tasks: each one on time is not enough
Five parallel tasks each 80% likely to finish by day 20 give a project only a 33% chance of finishing by day 20. Merge bias, worked out exactly.
A project waits for its slowest branch
When several tasks run in parallel and a later task needs all of them, the project can only start that later task when the last branch finishes. That makes the project's finish the maximum of the branch finish times, and the maximum of several uncertain numbers is later than the typical value of any one of them. Schedulers call this merge bias.
The exact arithmetic for independent branches
If each of k independent parallel branches has probability p of finishing by a given day, the chance that all of them do is pk. Take a deadline that each branch meets with 50% probability, then 80%:
| Parallel branches | Each at 50% | Each at 80% | Per-branch confidence needed for 80% overall |
|---|---|---|---|
| 1 | 50.0% | 80.0% | 80.0% |
| 2 | 25.0% | 64.0% | 89.4% |
| 3 | 12.5% | 51.2% | 92.8% |
| 5 | 3.1% | 32.8% | 95.6% |
Three branches that are each as likely as not to finish by day 20 give a project with only a one-in-eight chance of reaching a day-20 milestone. Even when every branch is 80% likely, five of them together are only 32.8% likely. To be 80% confident of the merge, each of five branches needs about 95.6% confidence on its own.
Why a plain sum of task durations misses it
Adding the most likely durations along one path ignores the other paths entirely, and adding every task's duration treats parallel work as sequential. Neither shows the shape of the merge. A simulation handles it directly: in every run it samples each task, computes finish dates through the dependency network, and records the project finish. The percentiles of those project finishes are the deadline confidence levels.
What to do with it
- Promise a date from the project distribution, not from a single branch. The date you can commit to at 80% confidence is later than any branch's own P80.
- Look for near-critical branches. A branch with a short expected duration but a wide range can drive the tail more than a longer, steadier one.
- Reduce the number of tight merges. Starting an uncertain branch earlier, splitting a risky branch, or removing a dependency lowers the effect directly.
- Check assumptions about independence. If the branches share people or a client approval, they tend to slip together, which changes the picture in both directions.
Read dependencies and elapsed time for the staffing side of the same problem, and deadline confidence without padding for how to state a date with a known confidence. For a single task, the deadline confidence calculator shows the one-branch case.
Try the calculation or follow the workflow
- Calculator
Deadline confidence calculator
Check the chance of finishing within an elapsed-day deadline from one duration range.
- Calculator
Project capacity calculator
Compare required project effort with team capacity. Calculate an hours shortfall, minimum team size and resource-only duration from availability assumptions.
- Guide
Software project estimating: effort, dependencies and risk
Build a software delivery estimate from tasks, three-point ranges, dependency chains, availability and discrete risks. Includes a concrete schedule example.