By: WKFGo Editorial · ·
What is a cumulative flow diagram?
A cumulative flow diagram (CFD) plots the cumulative number of work items that have reached each workflow state over time. Each colored band represents a state such as Backlog, In Progress, Review, or Done. The vertical thickness of a band approximates work in that state, while the slope of the Done boundary reflects throughput.
Unlike a board screenshot, a CFD shows how the system changed. It can reveal a growing review queue weeks before a deadline report turns red.
1. Band width: work in progress
The vertical distance between two boundary lines shows how many items occupy that state at a point in time. A widening Review band means items enter Review faster than they leave it.
Width alone is not automatically bad. A larger team or a planned batch can increase WIP. The question is whether the band remains controlled and whether downstream delivery also increases.
2. Slope: arrival and departure rate
The top boundary reflects incoming work; the Done boundary reflects completed work. If arrivals rise faster than completions, total unfinished work grows. Parallel, roughly stable boundaries suggest a more stable flow.
A flat Done line means no items complete during that period. Investigate whether work is blocked, batched, oversized, or simply not being updated.
3. Horizontal distance: elapsed time
The approximate horizontal distance between an item's entry and completion boundaries relates to lead or cycle time. On an aggregate CFD, this is an approximation—not a substitute for item-level cycle-time data—but a widening horizontal gap indicates slower flow.
Common CFD patterns
| Pattern | Likely interpretation | Questions to ask |
|---|---|---|
| One band steadily widens | bottleneck in that state | capacity, policy, or oversized items? |
| All active bands widen | inflow exceeds delivery | too much demand or low throughput? |
| Done rises in large steps | batch completion or data updates | real batch policy or stale board? |
| Band disappears suddenly | workflow changed or cards moved in bulk | was history preserved correctly? |
| Bands cross or shrink unexpectedly | data correction or invalid transitions | does the chart handle reopened/deleted work? |
| Stable parallel bands | relatively stable system | are service expectations still met? |
Treat these as diagnostic prompts, not automatic conclusions. A chart cannot explain why a queue formed.
A worked example
Suppose a six-week CFD shows:
- In Progress remains near five items;
- Review grows from two to eleven items;
- Done changes from five items per week to three;
- incoming work remains eight items per week.
Adding more developers would probably worsen the Review queue. Better experiments include smaller changes, reviewer rotation, clearer acceptance criteria, automated checks, or dedicated review capacity. The CFD points to the constrained state; item-level data and team observation identify the cause.
CFD, WIP, throughput, and cycle time
These measures are related but not interchangeable:
- WIP: unfinished items currently inside the workflow;
- throughput: items completed per unit of time;
- cycle time: elapsed time from the chosen start to finish points;
- work-item age: elapsed time for an item still open.
Little's Law connects average WIP, throughput, and cycle time in a sufficiently stable system. A CFD helps assess stability, but teams should not force a formula onto a period with major scope changes, irregular arrivals, or inconsistent policies.
Choose the chart boundaries carefully
A CFD is only meaningful when states have clear definitions. Decide:
- which item types are included;
- what event starts the measurement;
- what counts as Done;
- whether canceled and deleted items remain in history;
- how reopened work is represented;
- whether blocked is a state, flag, or separate analysis;
- which dates and filters define the period.
Changing these rules mid-chart can create apparent improvement that is only a data-definition change. Annotate workflow changes.
How to review a CFD each week
- Use the same filters and a consistent time window.
- Look for a band whose width or slope changed materially.
- Compare it with individual item age and blocked reasons.
- Ask the people working in that stage what changed.
- Choose one intervention with an owner.
- Annotate the intervention date.
- Wait long enough for the workflow to respond before declaring success.
The meeting can take 15 minutes. The goal is not to explain every pixel; it is to make one evidence-based flow decision.
Inconsistent state updates
If people move cards to Done once a week, the chart displays artificial completion batches. Fix board discipline before drawing operational conclusions.
Mixed item sizes and classes
A tiny bug and a multi-week feature both count as one item. Segment by work type or complement the chart with size and cycle-time distributions.
Workflow changes without annotations
Adding a new Review state changes band shapes. Preserve history and mark the date rather than pretending the before-and-after chart has identical definitions.
Hidden work
Requests handled in chat, spreadsheets, or untracked branches never reach the CFD. A stable chart can coexist with overloaded people if work is missing from the system.
Individual performance interpretation
A CFD describes a system. It cannot fairly rank people. Using it for individual evaluation encourages gaming and hides collaboration at bottlenecks.
CFD versus other charts
Burndown: shows remaining planned work, usually within a sprint. Useful for scope and sprint trajectory; it does not show where work waits.
Control chart: shows cycle-time distribution per completed item. Better for variability and service-level expectations.
Aging chart: shows how long current open items have remained unfinished. Better for immediate intervention.
Throughput chart: shows completed-item counts by period. Better for capacity trends.
Use them together. A CFD identifies the stage pattern; aging names the current risk; cycle-time data quantifies predictability.
A practical intervention matrix
| CFD signal | Possible experiment |
|---|---|
| Review band widens | smaller PRs, reviewer rotation, automated checks |
| Testing band widens | earlier test design, environment capacity, fewer batches |
| In Progress widens | WIP limit, pair to finish, split oversized work |
| Backlog grows rapidly | replenishment policy, demand triage, capacity decision |
| Done arrives in spikes | reduce batch size, update states continuously |
An experiment should state the expected signal. “Add a reviewer for two weeks; Review WIP and 85th-percentile age should decline without increasing escaped defects” is testable.
Cumulative flow in WKFGo
WKFGo can derive flow analysis from board history and pair it with work-item aging and project reporting. Keep the workflow definitions stable, use a consistent filter, and connect each intervention to a recorded decision. The chart provides evidence; it does not replace team judgment.
For setting the constraint that often follows a CFD review, read the practical guide to Kanban WIP limits.
Primary references and verification
This list supports further verification; inline links remain the source for a specific claim.