Capacity-Blended Multi-Team Forecasting (Monte Carlo): How It Works

Capacity-Blended Multi-Team Forecasting (Monte Carlo): How It Works

This page explains how Release Management, Roadmaps & Product Portfolio for Jira produces a single probabilistic release date when a Fix Version spans more than one team. It is a neutral reference doc; for the story behind why single-date rollups fail, see the related blog post linked at the bottom.

What "capacity-blended" means

When a release draws issues from several teams (for example frontend, backend, and platform), each team has its own delivery rate. Capacity-blended forecasting models each team's historical throughput — the number of issues it actually completes per sprint or per week — separately, then combines those distributions into one forecast for the whole release rather than averaging the teams into a single fictional team.

How the forecast is built

  • Per-team throughput history. The app reads each contributing team's completed-issue history at a consistent granularity (per sprint or per week).

  • Monte Carlo simulation. It runs thousands of simulated futures, each one randomly sampling from what every team has actually delivered before. Randomness is constrained by real history, not optimism.

  • Roll-up into one date. Each simulated run resolves the entire remaining scope across all teams; the spread of outcomes becomes the release forecast.

  • Percentile output. Results are expressed as p50 / p85 / p95 confidence dates for the combined release — one number set, not one date per team.

Reading the result

The p85 date is the commonly used planning date: the release finishes on or before it in roughly 85% of simulations. Because the model keeps each team's distribution intact, you can also see which team is the constraint pulling the later percentiles out — the slowest or least predictable team, not the sum of all teams, sets the date.

The combined release is only as "done" as its bottleneck team. Adding more scope to a fast team does not move the date if a different team is the constraint.

Prerequisites

  • Each contributing team has enough completed-issue history for a stable sample (a handful of sprints is usually enough).

  • Issues are assigned to the release's Fix Version so they are inside the forecast scope.

  • Throughput is read at a consistent cadence across teams.

Related