Actionable Agile Metrics: Predicting 'Done'

The mental model

Stop making promises based on gut-feel estimates and start making forecasts based on historical data. Treat your workflow like a system through which work flows. By measuring how tasks move from start to finish, you can make statistically sound predictions about future work. This answers the perpetual customer question, "When will it be done?", not with a single, fragile date, but with a probabilistic forecast that accounts for real-world uncertainty.

How it works

The system is built on four basic flow metrics. First, Work In Progress (WIP) is the count of all tasks currently being worked on. Second, Cycle Time is the elapsed time a single task takes from start to finish. Third, Throughput is the number of tasks finished per unit of time (e.g., per week). Fourth, Work Item Age is the elapsed time for tasks that are still in progress. By collecting this data, you can create powerful visualizations like a Cycle Time Scatterplot, which plots the cycle time for every completed item. This chart reveals the true distribution of your delivery times, allowing you to move beyond misleading averages.

When to use it

Use these metrics when your primary goal is predictability. If your customers, stakeholders, or team members are constantly frustrated by missed deadlines and unreliable estimates, flow metrics provide a path forward. They are particularly effective for teams using flow-based systems like Kanban, or Scrum teams looking to improve their forecasting capabilities beyond velocity. The goal is to establish a credible Service Level Expectation (SLE), such as "85% of our work items are finished in 10 days or less," which provides a reliable basis for planning.

When not to use it

These metrics are less useful for teams working on highly experimental R&D projects where discovery is the main goal, not predictable delivery. If there's no repeatable process or consistent flow of work, historical data will not be a good predictor of the future. Similarly, if your organization is unwilling to define and track work from a clear "start" to a clear "finish" point, the metrics will be garbage-in, garbage-out. They require discipline.

One canonical example

A team wants to answer "How long will this new feature take?". Instead of an estimation meeting, they look at their Cycle Time Scatterplot. They see that 85% of similar features in the past were completed in 15 days or less. They can now tell their stakeholder: "Based on our historical data, there is an 85% probability we will deliver this feature within 15 days of starting it." This is not a guarantee, but a probabilistic forecast that manages expectations far better than a single-date promise. This 85th percentile becomes their Service Level Expectation (SLE).