Throughput
Throughput is the exact count of work items finished during a defined unit of time, such as a day, week, sprint, or month.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/throughput/
01What it means
Throughput counts how many work items cross a defined finished point per unit of time. It is an observed completion rate, not an estimate of effort or a count of work started. The unit might be items per week or features per month, but the item population and finish rule must remain understandable and reasonably consistent.[1][2]
02Why it matters
A history of completions supports capacity conversations and probabilistic forecasts without requiring teams to convert every item into hours or points. Throughput trends can also show instability, while comparison with arrival rate helps reveal whether demand is accumulating faster than the system completes it.
03When to use it
Use throughput for recurring delivery systems with enough completed items to form a useful history. It is especially helpful for answering how many items may finish in a period, feeding Monte Carlo forecasts, and checking the effects of workflow changes. Segment radically different work when one combined count would conceal important behavior.
04How it works
- Define the work-item population, finished state, counting interval, and rules for reopened or split items.
- Count only items that actually cross the finish boundary during each interval.
- Visualize the time series and distribution; retain zero-throughput periods because they are real observations.
- Investigate changes alongside WIP, cycle time, blockers, demand, and quality rather than assuming a cause.
- Use representative historical samples for forecasting, revisiting the sample when the system materially changes.
05A practical example
A content platform counts production-ready experiments completed each week. Over 16 comparable weeks, throughput ranges from two to seven, including one week with none. Instead of planning from the average alone, the team uses the empirical weekly samples in a forecast and examines whether review capacity explains the low weeks.
06Where it breaks
Raw counts treat items as countable units and can mislead when their nature changes sharply. Throughput does not measure outcomes, quality, effort, or individual productivity. Incentives tied to higher counts encourage splitting, easy-item selection, or premature completion. Small samples and major process changes weaken forecasts based on historical throughput.
Common mistakes
- Counting started or nearly finished work instead of items that crossed the agreed finish point.
- Equating throughput with story-point velocity even though one counts items and the other sums estimates.
- Rewarding higher counts without checking value, quality, item mix, and downstream consequences.
07Sources
- The Kanban Guide (May 2025)Kanban Guides · Official guideThe Kanban Guide (May 2025)
- The Kanban Pocket Guide: The Basic Metrics of FlowProKanban.org · Professional guideThe Kanban Pocket Guide: The Basic Metrics of Flow