Cumulative Flow Diagram (CFD) | Agile SM
Cumulative Flow Diagram is a workflow visualization that plots counts of work items over time across states (for example To Do, In Progress, Done) to make flow visible. By showing how WIP accumulates or drains in each band, it helps teams detect bottlenecks, stability issues, and policy violations early, then agree improvements grounded in data rather than opinions. It is commonly used with Kanban systems and can complement Scrum by revealing flow patterns across Sprints. Key elements: time axis, workflow states/bands, WIP per state, throughput trend, lead time cues, trend changes.
How a Cumulative Flow Diagram (CFD) works
A Cumulative Flow Diagram shows how many work items exist in each workflow state over time. The stacked bands make queues visible so teams can inspect how the system behaves, not evaluate individuals, and then adapt policies to improve flow.
Cumulative Flow Diagram uses time on the horizontal axis and work item counts on the vertical axis. Each band represents a defined workflow state (for example Requested, In Progress, Review, Done). When bands remain roughly parallel, the system is relatively stable. When a band widens over multiple samples, work is accumulating in that state, which usually increases waiting time and raises lead time risk. This makes CFDs useful for empiricism: observe the trend, form a hypothesis about the constraint, change one policy or practice, and see whether the band behavior and downstream outcomes improve.
Key Components of a Cumulative Flow Diagram (CFD)
- Time (X-axis) - A consistent timeline (often daily snapshots) that makes trend changes and policy impacts visible.
- Work items (Y-axis) - A count of items tracked through the workflow, using a stable unit of work long enough to learn from the data.
- Workflow states - Explicit states that match how work actually flows, each with clear entry and exit criteria.
- Band width - The amount of WIP in a state at a point in time, indicating queue size and potential waiting time.
- Band slope - The rate of change over time, indicating shifts in arrival rate, completion rate, or both.
Reading a Cumulative Flow Diagram (CFD) for flow signals
Interpret a CFD by looking for sustained changes in band width and slope. The goal is to detect where flow is slowing, what changed, and what to inspect next. Avoid reading meaning into single-day noise.
Common signals in a Cumulative Flow Diagram include the following:
- WIP growth - A band widening over time indicates arrivals exceed departures in that state, increasing queues and typically increasing lead time.
- Throughput change - A sustained change in the slope of completed work suggests the system’s completion capability changed, often due to constraints, quality, demand shifts, or policy changes.
- Constraint location - The state whose band widens persistently is a strong candidate for the current constraint and should be investigated first.
- Waiting and dependency drag - A widening waiting-related state indicates delays from dependencies, decision latency, handoffs, or unclear intake and prioritization.
- Batching patterns - Step-like progress or saw-tooth shapes often indicate batch reviews, batch testing, or batch releases that delay feedback and hide queues.
- Rework pressure - Repeated widening in review or test states can signal unclear acceptance criteria, insufficient discovery, unstable requirements, or quality gaps.
Use these signals to drive questions and experiments. The CFD points to where the system changed; it does not explain why by itself. Pair it with observation, explicit policies (including WIP limits), and complementary measures such as lead time distributions and aging work in progress to confirm root causes.
Building a CFD from workflow data
A CFD is only as useful as the workflow definition and the data behind it. Start by agreeing on workflow states that reflect reality, keep them stable long enough to learn, and define entry and exit criteria so transitions are meaningful. Ensure each work item can be in only one state at a time.
To create a CFD, take periodic snapshots (daily is common) of how many items are in each state. Many tools generate CFDs automatically, but the same logic applies if you build it from exported data.
The steps to create and maintain a useful Cumulative Flow Diagram are:
- Define the workflow - Name states, clarify policies, and decide whether to model waiting, blocked work, and rework explicitly.
- Choose a stable work item unit - Use a consistent item type and avoid mixing very different item classes in one view without filtering.
- Collect snapshots - Capture counts per state on a regular cadence that is frequent enough to reveal trends without creating noise.
- Validate data hygiene - Ensure items are updated promptly and transitions are not backfilled later, which distorts the trend.
- Plot stacked counts - For each date, plot the counts per state as stacked bands from bottom to top.
- Annotate meaningful changes - Record workflow definition changes, policy changes, team boundary changes, and tooling changes so interpretation stays fair.
- Review and learn - Inspect trends, form hypotheses, run small experiments, and check whether the bands and outcome signals move as expected.
If your workflow includes different classes of service or work types, consider separate CFDs (or a filterable view). Otherwise, variability can mask constraints and lead to incorrect conclusions.
Improving flow with a Cumulative Flow Diagram (CFD)
The purpose of a CFD is to improve the system of work through evidence-based adaptation. Use it to test whether changes reduce queue size, stabilize WIP, and improve predictability and quality, then keep what works and adjust what does not.
Typical improvement actions informed by a Cumulative Flow Diagram include:
- Adjust WIP limits and pull policies - When a band widens, reduce new arrivals into that state and focus on finishing work already started before starting more.
- Reduce batch size - Slice work into thinner vertical increments to shorten feedback loops and reduce queue buildup.
- Strengthen upstream clarity - Persistent widening in review or test often signals unclear acceptance criteria or weak refinement; improve shared understanding earlier.
- Reduce handoffs and skill bottlenecks - Repeated queues in one state can indicate specialization constraints; broaden capability and redesign the workflow to reduce waiting.
- Make dependencies explicit and renegotiate - When waiting states widen, clarify decision rights, integrate earlier, and change sequencing to reduce external delays.
CFDs can support forecasting when combined with historical throughput or lead time distributions. Keep forecasts probabilistic and recalibrate frequently. Use forecasts to inform trade-offs and options, not as targets that pressure the system into more WIP and lower quality.
Benefits and limitations of Cumulative Flow Diagram
Cumulative Flow Diagram makes system behavior visible and supports better conversations about constraints. It is especially effective in Kanban contexts, where WIP and flow policies are explicit, and it can complement Scrum by revealing where work queues up across and between iterations, such as around review, testing, or release steps.
Limitations matter. A CFD can mislead if workflow states are unclear, data is stale, or item size varies widely while the chart tracks only counts. It also does not measure customer outcomes by itself; it is a flow visualization. Pair it with outcome measures and quality signals so improvements increase value delivered, not just motion through states.
Misuse and guardrails
A CFD is sometimes misapplied as a performance tool or as justification to push more work into the system. That turns a learning instrument into a pressure mechanism, which typically increases WIP, delays feedback, and degrades predictability.
- People scoring - Looks like using the chart to judge individuals; it hurts by driving defensive behavior and hidden work; do instead use it to discuss system constraints and policy changes with the whole team.
- Throughput pressure - Looks like demanding a steeper Done slope without changing constraints; it hurts by incentivizing shortcuts and rework; do instead change policies, reduce queues, and improve quality so throughput improves as an outcome.
- Unsafe comparisons - Looks like ranking teams by chart shape without accounting for workflow and work-type differences; it hurts by creating false conclusions; do instead compare trends within the same system and make differences explicit when comparing.
- Gaming transitions - Looks like moving items to Done early or redefining states to look better; it hurts by hiding queues and quality issues; do instead keep state definitions meaningful and verify with quality and release evidence.
- Noise-driven changes - Looks like changing policies based on short spikes; it hurts by destabilizing flow; do instead look for sustained trends, connect them to observable causes, then run a small experiment.
Cumulative Flow Diagram (CFD) is a flow analytics chart showing work item counts over time by workflow state to reveal bottlenecks and improve predictability