How to Read a Cumulative Flow Diagram (With Examples)
By Sebastiaan Jansen · · 7 min read
A cumulative flow diagram is a stacked chart of how many work items have reached each stage over time. The slope of the done band is your departure rate, and the vertical gap between arrivals and done is work still on the board. A taller band is where work piles up. Read the shape rather than any single point.
People shorten the name to cumulative flow. An item entering a stage adds to a running total, which only falls if someone rewrites history.
What the chart is made of
Time runs left to right, usually one point a day. The vertical axis is a cumulative count, and each stage is a band. The top edge is every item that entered the system you plotted.
With done at the bottom, its top edge is items finished by that day, with earlier stages above it. Some tools stack the other way, so check the legend; the counts don't change.
Thickness is items in that stage that day. The top of the stack is cumulative arrivals, the top of done is cumulative departures, and the space between them is unfinished work.
Three distances matter. Vertically, arrivals minus done is what is still plotted: work in progress on committed stages, or mostly "not started" if a raw backlog is in the stack. Horizontally, the days from a point on arrivals until done reaches that count are a cohort average, not one ticket. For a single item, use its own start and finish, as in Cycle Time in Software Development: Definition, Formula and Benchmarks. The slope of done is throughput, and a flat done line means nothing finished. If the top rises faster than done for long enough, the board is taking on more than it finishes.
What to include
Start at commitment: Ready, Building, Review, Done. An idea backlog would dwarf the active bands. Count items, not points, unless the title says points. Split types that don't share a workflow, because incidents and multi-week projects on one stack match nobody's week.
How to read a cumulative flow diagram
Read the done slope, then arrivals against it, then which band changed thickness, then the policy behind the bend. Five shapes cover most boards.
Parallel bands
Done rises steadily, upper bands keep a similar thickness, and the top tracks done. Arrivals match departures, so no queue is growing. This is the only shape where one average time through the system is fair, because the horizontal gap holds. Parallel doesn't mean fast, though. A wide, stable gap is predictable; speed is that gap with the slope of done.
A band that widens
One band thickens while others hold or shrink, so items enter it faster than they leave. If Review widens while a reviewer is away, that's capacity. If it widens every Thursday, that's batching. Starting more work only feeds the thick band, so cap it and stop starting until it thins. WIP Limits: How to Set Them and What Changes When You Do covers the policy this shape points to.
A flat done band
Done stops while an upper band grows: people are busy and nothing finishes. A flat top is an old queue; a rising top is new work arriving during the stall. One flat day is noise and a flat week is a signal, so don't fold it into the quarter's average.
A step
A jump in the top line is a batch arrival: planning day, a dumped checklist, or an import. A jump in done is a batch finish, like a release or a Friday cleanup. A staircase means lumpy delivery, and one average hides the gaps between steps.
A bulge that then flushes
A thick band thins while done steepens, because the queue left. That slope is the bulge clearing, not a new pace. Wait for parallel bands before you quote a rate. Refill upstream right away and the bulge just moves one column left.
A worked example: ten days, one review bulge
These numbers are illustrative, not a benchmark. Done is cumulative, and arrivals entered Ready. Ready + Building + Review + Done should equal Arrivals. A row that doesn't add up is missing something, double-counting, or using a stage you left off.
| Day | Arrivals | Ready | Building | Review | Done |
|---|---|---|---|---|---|
| 1 | 8 | 5 | 2 | 1 | 0 |
| 5 | 18 | 7 | 4 | 3 | 4 |
| 8 | 22 | 5 | 3 | 8 | 6 |
| 10 | 24 | 3 | 3 | 5 | 13 |
Done goes from 0 to 13. Review climbs from 1 to 8 by day 8 while Building stays near 3 or 4, so Review is the constraint. Ready jumps on day 5, then thins. Done moves about one a day through day 8, then 3 and 4, which is a flush rather than the rate. Unfinished work grows from 8 to 16, and days 9 and 10 only partly clear it.
Hold the next batch out of Ready while Review is over its cap, and stop Building from pulling, so the wait sits upstream. Don't quote a single lead time while the gap is moving and done has two regimes. If you need a number, use Throughput vs Cycle Time: Reading Both Without Fooling Yourself.
Misreads that send you to the wrong fix
The horizontal gap is an average only while arrivals and departures are steady. Expediting the newest item keeps the chart calm while older tickets sit. Aging on the ticket names the stuck item; the chart can't.
A top line from ideas tracks demand, which often should beat done. For load, start at the first committed stage. A renamed or split column kinks every band, so annotate it. Deleting "won't do" rewrites last month's done line, so keep the history. Compare a team with its own past, since smaller slices alone steepen done. Weekend flats are the calendar: plot working days or ignore them.
How it sits with the other kanban flow metrics
This chart is one of the kanban flow metrics, with throughput, cycle time, and work in progress. It shows whether those numbers are stable enough to average.
Little's Law says average work in progress equals average throughput times average time in the system, measured the same way, on a roughly stable system. Parallel bands are what that stability looks like. A widening band still has an average, but don't manage from it yet.
A thickening review band points at the review cap, and percentiles show the tail. Throughput over a calm window tells you what done has been. Monte Carlo Forecasting for Software Delivery Dates turns those finishes into a date range once the bands settle. On a staircase or a widening gap, a wide range is the truthful answer.
The chart describes the system; lead time and cycle time describe items. Cycle Time vs Lead Time: The Difference and Why Both Matter keeps start and end aligned with the gap you're reading. If you need one score, give the done slope you're willing to defend, the thickest band, and when it started thickening.
A ten-minute weekly read
Count what finished. If done flattened, ask what finished, not what people touched. If arrivals beat done, ask who may add work. The band that changed most is the agenda. Mark releases, holidays, renames, and imports. Leave with one policy: a cap, smaller intake, or same-day review.
A launch needs a smooth done band rather than a staircase. The same history can sit next to cycle time, cumulative flow and Monte Carlo forecasts. Jira, Azure Boards, and Falrow all draw the stack. Policy thins a band; extra statuses won't make arrivals and done parallel. Teams that publish launch dates marketing can plan against protect the done slope. If the same stage bulges this month and did last quarter too, the problem is the workflow.
FAQ
What is a cumulative flow diagram?
Running totals by stage, stacked over time. The top edge is entries, the done band is finishes, and thickness is items in that stage, so you see work arriving, waiting, or leaving.
How is it different from a burndown chart?
A burndown tracks remaining scope against a hoped-for date. This chart has no target line. It grows as work arrives and shows which stage holds it. A burndown fits a closed batch; the flow chart fits a board that keeps taking work.
How do you spot a bottleneck?
Look for a band that thickens while done doesn't steepen to match, then check the tickets. A holiday, a batch release, or a new column can fake the shape for a few days. If the band stays thick across two or three reviews, it deserves a WIP change.
Can the chart give you an item's cycle time?
Only as a rough average, read across from arrivals to done while the bands are nearly parallel. A single ticket uses its own start and finish. Use the shape to judge whether an average means anything, then use item times for a percentile or a forecast.