Using Cumulative Flow Diagrams to Predict Project Delivery Dates

Project delivery dates are often treated as fixed commitments long before the work has generated enough evidence to support them. A Cumulative Flow Diagram (CFD) offers a more reliable view by showing how work moves through a workflow over time. Instead of relying on optimism, status colours or a single estimate, teams can examine completed work, work in progress and emerging bottlenecks.

A CFD is especially useful for digital product teams, engineering groups and operational improvement projects where work arrives continuously and passes through several stages. It can reveal whether the system is stabilising, whether unfinished work is accumulating, and how long items typically take from entry to completion. Those patterns create a practical basis for forecasting a likely delivery window.

This approach aligns with the applied research explored through Project Visit, where digital visual planning, Kanban, enterprise systems and lean processes connect research with industry practice. For Australian organisations working across Melbourne, Sydney, Brisbane or regional sites, a flow-based forecast can also account for public holidays, supplier delays and changing customer demand without disguising uncertainty.

Reading The Shape Of Flow

A cumulative flow diagram usually displays time on the horizontal axis and the number of work items on the vertical axis. Each workflow state appears as a coloured band, such as ready, analysis, development, testing and done. The height of a band represents the amount of work in that state, while the distance between the start and finish of an item provides evidence about flow time.

A healthy system generally shows bands that remain reasonably parallel and move upwards at a steady rate. The completed band rises consistently, indicating a dependable throughput. A widening development or testing band signals that work is entering the stage faster than it is leaving. If the testing band expands for several reporting periods, the delivery date is likely to move even when individual task estimates appear reasonable.

The diagram must reflect the actual workflow rather than an idealised process. If a team records “done” before customer acceptance, the forecast will be artificially positive. If blocked work is hidden in a separate spreadsheet, the visual system will understate congestion. Clear policies for starting, pausing and completing work are therefore as important as the chart itself.

Building A Forecast From Real Data

The first requirement is consistent historical data. Each item should have a recorded start point, completion point and relevant workflow transitions. Teams do not need a complex platform to begin; a Kanban board with reliable timestamps can provide enough information. The valuable signal comes from repeated observations, not from a large volume of administrative detail.

A delivery forecast can begin with throughput. If a team completes between six and ten items per fortnight, a forecast for the next forty items should use that observed range rather than an average that conceals variation. Percentiles are often more informative than a single mean: the 50th percentile shows a typical result, while the 85th or 90th percentile offers a more cautious planning boundary.

Lead time provides a second perspective. It measures how long an item takes to travel from an agreed starting point to completion. When lead-time data is plotted as a distribution, stakeholders can see whether most work finishes within two weeks or whether a small number of long-running items create substantial risk. A CFD helps explain the causes behind that distribution by showing where work spends its time.

Practical Signals For Australian Teams

The same flow pattern can have different operational causes depending on the local setting. A Sydney software team may pause work while a client’s legal or security review is conducted in another time zone. A manufacturer in Geelong may face a supplier constraint, while a Brisbane construction technology project may lose capacity during a site shutdown or seasonal weather event. The CFD does not identify every cause by itself, but it makes the effect visible.

Australian planning also needs to distinguish ordinary variation from calendar effects. An Easter shutdown, state public holiday or end-of-financial-year workload can temporarily reduce throughput. Treating that period as a permanent change would distort the forecast; ignoring it could produce a delivery promise that falls directly into a known capacity gap. Teams should annotate such events on the chart and explain whether the impact is expected to recur.

Useful checks before accepting a forecast include:

• Confirm that “done” means the same thing across every reporting period.
• Mark public holidays, planned shutdowns, releases and major supplier interruptions.
• Separate urgent expedite work from the normal delivery stream.
• Investigate widening bands before adding people or increasing work in progress.
• Compare forecasts with actual delivery outcomes at regular review points.

The diagram is also valuable in organisations with distributed teams. A product group in Melbourne may depend on a testing partner in Adelaide and a platform team in Sydney. Their combined CFD can expose queues created between locations, while separate views can show whether the constraint belongs to a particular service or handover. This is more useful than assigning blame to whichever team owns the visible backlog.

Turning Lead Time Into Delivery Ranges

A CFD can support a delivery-date forecast, but it does not produce a guaranteed calendar day. The sensible output is a range linked to a stated confidence level. For example, a team might forecast that a feature set is likely to finish between 12 and 18 weeks, with the later boundary representing a higher level of confidence. This is more honest and more actionable than saying “it will be ready in 15 weeks”.

Little’s Law provides a useful relationship between work in progress, throughput and average lead time: average lead time is approximately work in progress divided by throughput. The relationship is not a magic prediction formula, because it depends on a reasonably stable system. However, it explains why adding more items to an already crowded workflow usually extends delivery time rather than accelerating it.

For larger initiatives, teams can use historical throughput to run a Monte Carlo simulation. The simulation repeatedly samples actual past delivery results and creates many possible future sequences. The resulting range can show the probability of completing a defined number of items by a particular date. Since it uses the team’s own evidence, it captures variation that a carefully debated estimate may miss.

Forecast quality depends on the unit being forecast. If one work item can represent a day of effort and another can represent three months, throughput becomes misleading. Teams should split work into items of broadly comparable size or use a clear class-of-service model. Forecasts should also state what is included, such as completed customer stories, approved change requests or tested production increments.

Using Forecasts In Planning Conversations

A reliable forecast changes the tone of planning meetings. Instead of asking whether a team feels confident about a date, leaders can discuss the evidence behind a range, the assumptions supporting it and the decisions that could change it. This creates a shared view of risk without turning uncertainty into a performance failure.

Forecasts are particularly helpful when negotiating scope. If a fixed launch date is immovable, the CFD can support a conversation about which items are essential and which can follow later. If the scope is fixed, stakeholders can examine the likely date range and decide whether additional capacity, reduced dependencies or a different release strategy is justified. The chart makes the trade-off visible.

A practical review can focus on these management signals:

• Completed work rising steadily over several reporting periods.
• A growing band showing that work is entering a stage faster than it exits.
• A high proportion of aged items compared with the normal lead-time range.
• Throughput falling after a policy, dependency or staffing change.
• Frequent urgent work interrupting the planned delivery stream.

Visual flow ideas also transfer beyond software. An organisation coordinating community services, training or student support can examine registrations, preparation, delivery and follow-up as connected stages. The same logic appears in after-school academic support, where demand, available places and service completion can be considered as a flow rather than as isolated activities. The context differs, but queues and capacity still shape reliable commitments.

Making The Forecast Operational

The CFD becomes valuable when it is part of the operating rhythm rather than a report prepared after a delay. A team can review the chart weekly, identify the largest change in band width and select one constraint for investigation. Leaders can review longer-term trends monthly, comparing forecast ranges with actual outcomes and checking whether process changes have improved flow.

Teams should also protect the integrity of the data. Changing workflow definitions halfway through a project, reopening completed items without recording why, or excluding blocked work makes historical comparisons weak. When the process changes, the chart should show the boundary between the old and new system. Forecasts can then use the most relevant period instead of mixing incompatible evidence.

A delivery forecast is strongest when it combines three views: the CFD for system behaviour, lead-time data for item-level experience and throughput history for completion capacity. Together, they help an Australian team plan around real constraints, communicate ranges responsibly and decide where intervention will have the greatest effect.

The next concrete step is to export the past eight to twelve weeks of workflow data, draw the cumulative bands, and record the first forecast range for the next agreed delivery milestone.