Flow Metrics Fundamentals: WIP, Throughput, Cycle Time
A board full of columns isn't a flow system by itself. Before any number from it is comparable week to week, the team needs to agree on where work starts, where it ends, and what each state actually means.
Six questions to agree on first
- • Value unit — what actually flows through the system?
- • Start / end — where exactly do we measure the time between?
- • States — which stages does work pass through?
- • WIP control — how many items can be in a stage at once?
- • Explicit policies — when and how does work move forward?
- • Service expectation — what does past data say about duration?
Kanban is built on visualizing and defining this flow, actively managing items in progress, and continuously improving it — in that order. Without the first step, the metrics below aren't comparable across weeks or teams.
Four core flow metrics
WIP (work in progress)
The count of items started but not yet finished. The earliest signal of lost focus and growing queues.
Throughput
The number of items completed in a given period. Shows the system's actual completion rate.
Cycle time
How long an item takes from start to finish. Only calculated for completed work.
Work item age
How long an item still in progress has been running. Surfaces at-risk work early.
Watching all four on one board
Picture a five-person support team's board on a Monday morning. The “In Progress” column holds exactly 9 cards — that's the WIP at that moment. Over the previous week (Monday to Friday), the team moved 7 cards into “Done” — weekly throughput was 7.
The cycle times of those 7 cards, in days, sorted smallest to largest: 1, 2, 2, 3, 3, 4, 15. That sums to 30, for an average of about 4.3 days. But look at the median — the middle value, 3 days — and a different story appears: six cards finished in 4 days or less, and a single card, probably one stuck waiting on a third-party approval, dragged on for 15 days and pulled the average up on its own. Someone who only reads “our average cycle time is 4.3 days” misses most of the picture; the median and the full spread, read together, tell you what actually happened that week.
Now go back to the 9 cards still in WIP. One of them has been sitting “In Progress” for 11 days — it hasn't finished yet, so it doesn't show up in throughput or the cycle-time average at all, but the work-item-age report surfaces it today, in real time. If the team doesn't notice it for another week, it will either quietly drag down next week's throughput or, once it finally finishes, distort the cycle-time average with another outlier — exactly like the 15-day card above.
The practical takeaway for this team: pushing throughput up isn't the most valuable first move. Asking why the 11-day card is stuck is — that's a decision the age signal can trigger today, weeks before the problem would show up in any other number.
Cycle time and work item age complement each other this way in every dataset, not just this one: cycle time tells you what already happened to finished work, while work item age tells you what's still at risk right now.
Now practice it
Take a set of degrading flow metrics, separate what the data says from what you're reading into it, and design a small reversible experiment — with AI rubric feedback.
Open the PracticeLab →Frequently asked questions
What's the difference between cycle time and work item age?
Cycle time is measured on finished work — how long it took from start to done. Work item age is measured on unfinished work — how long an item has been in progress right now. They complement each other: one shows what already happened, the other shows what's at risk today.
Is a higher throughput always better?
Not on its own. Rising throughput alongside rising WIP and a flat or worsening quality signal usually means work is being rushed or split artificially, not that the system got faster.
Do we need special tooling to track these four metrics?
No — any board with clear start/end boundaries and consistent state names can produce all four. The hard part is agreeing on the definitions, not the tooling.