Critical chain method
The critical chain method schedules a project on a network that respects the dependencies between activities and the capacity of the resources. The critical chain is the longest sequence of activities in this levelled network. The method takes the safety margin out of each estimate and pools part of it in a project buffer, placed after the last activity of the chain. During the project, the team tracks how much of this buffer is consumed; no activity receives a finish date. Eliyahu Goldratt published the method in 1997 as an application of the theory of constraints to projects; the full set of its practices is called CCPM (critical chain project management).
Goal
The critical chain method produces a schedule that accounts for the people and equipment available, then protects the finish date with a project buffer. It tells the project which finish date to promise with the resources it actually has, then whether that date holds along the way.
It targets projects that run late although every estimate contained a margin. The safety held in each activity gets lost along the way: an activity that finishes early rarely hands back the time it saved, while an activity that runs late passes its delay on. The method takes this safety out of the activities and pools it where it protects the project date.
The deliverable is a critical chain schedule: the levelled network, the critical chain marked on it, the project buffer and the feeding buffers. During execution, the team adds the buffer tracking to this schedule. The PMBOK Guide lists the method among the tools of the processes that develop and control the schedule.
Usage
When to use it
- Resource contention: several activities need the same team, specialist or test bench.
- Repeated delays despite cautious estimates: take the safety out of the activities and pool it in a buffer.
- A single delivery date promised to a client: protect that date rather than the intermediate milestones.
- Staff spread across several projects: reduce the multitasking that stretches every activity.
When not to use it
- No resource contention: the chain coincides with the critical path; apply the critical path method.
- A culture that penalises any estimate overrun: teams put the safety back into their durations; keep the critical path and a contingency reserve.
- Order of work revised at every iteration: a frozen network cannot keep up; order a backlog and plan by time boxing.
Description
Critical chain and critical path
The critical chain method builds on the dependency network and the longest-path analysis of the critical path method. The critical chain is the critical path of a resource-loaded, levelled network: the PMBOK Guide treats the two terms as synonyms. The critical path method calculates the schedule from the dependencies alone; the planner levels afterwards, in a separate step. The critical chain method levels within the calculation of the chain itself. The two methods also differ on where the safety sits and on what the team tracks during execution.
| Question | Critical path | Critical chain |
|---|---|---|
| What the calculation respects | The dependencies; levelling follows, separately | The dependencies and the capacity of the resources, in the same calculation |
| Where the safety sits | In each estimate | In buffers, at the end of the chain |
| What the team tracks | The float and the dates of each activity | The consumption of the project buffer |
| Dates promised | A finish date per activity | The project finish date |
| Activities off the longest path | Scheduled as early as possible | Scheduled as late as possible, protected by a feeding buffer |
Building the levelled network
The team builds the network as it would for the critical path: activity list, durations, dependencies. It then assigns each activity the resource it consumes. Two parallel activities that need the same team cannot run at the same time. Resource levelling places them one after the other, which creates a resource link between them that the dependency network did not show.
The critical chain follows both the dependency links and the resource links. It can therefore run through an activity to which the critical path calculation gave float.
The order chosen between two conflicting activities changes the length of the chain. Scheduling software proposes an order; the team checks it.
Removing the safety from the estimates
A safe estimate covers contingencies: the activity almost certainly finishes within it. Goldratt describes two behaviours that consume this safety without protecting the date. Student syndrome puts off the start of the work as long as the deadline seems far away. Parkinson's law holds that work expands to fill the time allowed: an activity finished early is rarely reported as such. Multitasking comes on top: a person who switches between two ten-day activities finishes the first around day 20 instead of day 10.
The method replaces each safe duration with a duration within which the activity finishes half the time, called the 50% duration. Goldratt's original rule cuts each safe estimate in half. A team that has records of the actual durations of past projects estimates the median directly. By construction, an activity overruns its 50% duration half the time; the buffer absorbs these overruns.
Pooling the safety in buffers
The project buffer sits after the last activity of the critical chain. The date promised to the client is the end of this buffer. The PMBOK Guide specifies that the buffer replaces the safety in the activities and does not add to it. A contingency reserve placed on durations that are already safe doubles the safety.
The buffer pools the variability of the activities. The delays of some and the early finishes of others partly cancel out. The buffer can therefore be shorter than the sum of the margins it replaces. Goldratt's original rule, known as cut and paste, sets it at half the safety removed from the chain. When the safe durations were cut in half, this comes to half the length of the chain at 50% durations.
Sizing remains a decision. A buffer that is too short triggers emergency measures to hold the date; a buffer that is too long produces overpriced bids and lost contracts. The statistical rule known as RSEM (root square error method) sets the buffer at the square root of the sum of the squares of the margins removed. The half rule grows with the chain, RSEM with its square root: on a long chain, the half rule gives a longer buffer, which She (2020) finds excessive. The team records the rule it chose and reviews it against the actual durations of completed projects.
A feeding chain is a sequence of activities outside the critical chain that joins it at one point. A feeding buffer sits at this junction, between the last activity of the feeding chain and the critical activity that waits for it. It absorbs the delay of the feeding chain before that delay reaches the critical chain. The cut and paste rule sets it at half the safety removed from the feeding chain. The method schedules feeding chains as late as possible, so as not to commit work before it is needed, then brings them forward by the length of their buffer.
Bringing a feeding chain forward by the length of its buffer is impossible when its predecessor finishes too late. The buffer then pushes back the critical activity it protects: the critical chain changes. After placing the buffers, the team checks the chain again.
Executing without activity dates
The schedule gives the activities no finish date. An activity on the chain starts as soon as its predecessor is complete. Its owner works on it until it is done, then hands the result to the next activity without waiting. The project manager alerts the critical chain resources a few days before their turn. This advance notice is called the resource buffer: an alert that adds no day to the schedule.
Every day or every week, the activity owners report the remaining duration of the work in progress, not the percentage complete. The remaining duration of the chain, added to today's date, gives the projected finish. The share of the chain complete equals the length of the chain minus its remaining duration, divided by the length of the chain. The share of the buffer consumed equals the gap between the projected finish and the planned finish of the chain, divided by the length of the buffer.
Tracking the buffer
The PMBOK Guide's buffer protection index (BPI) compares the share of the project buffer that remains with the share of the critical chain that remains. The SPI (schedule performance index) of earned value management measures the deviation from the schedule baseline; the BPI shows whether the finish date is still protected.
The team plots each progress point on the fever chart, the chart that tracks the buffer: share of the chain complete on the horizontal axis, share of the buffer consumed on the vertical axis. It sets two thresholds in advance, two straight lines that rise less steeply than the diagonal. They tolerate early consumption of the buffer and tighten as the chain progresses. The team therefore reads each point against the thresholds, never against the diagonal. Below the first threshold, the project manager lets the team work. Between the two, the project manager prepares a recovery plan. Above the second, the project manager carries it out.
Buffer consumption also sets priorities from day to day. When a resource has to choose between two activities, it takes the one whose buffer is most depleted.
The team recalculates the chain when the scope, the network or the resources change. When only a delay occurs, the buffer absorbs it.
Pitfalls
Safety creeps back into the durations. A team that has been blamed for overrunning a 50% duration pads the next one, and the schedule carries the safety twice: in the activity and in the buffer. Management commits to judging no one on the overrun of a 50% duration, then keeps that commitment.
The buffer becomes a reserve for other work. A change request that adds work lengthens the chain. It goes through change control, which sets a new finish date and a new buffer.
Activity dates survive. By default, scheduling software publishes a finish date for every bar of the schedule, and teams read it as a deadline. The project manager gives the teams the order and the durations and publishes a single date: the project's.
Multitasking returns. A critical resource assigned to a second project "for a few days" stretches the chain without the network showing it. Management decides on these assignments based on the buffer consumption of both projects.
AI considerations
Each time it recalculates, scheduling software levels the network and finds the chain without error. A language model adds nothing there.
On an exported schedule, the model spots parallel activities that need the same resource and missing buffers at a junction point. It extracts the remaining durations from the activity owners' messages and prepares the update. It drafts the weekly report: share of the chain complete, share of the buffer consumed, activities that ate into it. On a history of completed projects, it compares estimated and actual durations and helps calibrate the sizing rule.
The team sets the 50% durations: a model proposes plausible durations for work it does not know, and a buffer calculated on them protects nothing. The project manager makes the recovery decision, which commits people and budget. Tracking by resource shows the workload and the delays of named staff: that is personal data under the Federal Act on Data Protection (FADP). The project manager obtains the agreement of the data protection advisor before sending the schedule to an online service.
Examples
A cantonal population office puts the application for a certificate of residence online. The form and the interface with the residents' register fall to the same development team.
| Id | Activity | Resource | Safe duration | 50% duration | Predecessors |
|---|---|---|---|---|---|
| A | Requirements analysis | Analyst | 20 | 10 | none |
| B | Form development | Development team | 24 | 12 | A |
| C | Interface with the residents' register | Development team | 16 | 8 | A |
| D | Writing the online help | Technical writer | 12 | 6 | A |
| E | Acceptance testing | Analyst | 12 | 6 | B, C, D |
| F | Go-live | Operations | 4 | 2 | E |
On the 50% durations, the critical path A-B-E-F lasts 30 days and leaves C 4 days of float. The development team cannot build B and C at the same time: levelling places C after B, and the critical chain A-B-C-E-F lasts 38 days.
The original rule sets the project buffer at half of the 38 days of safety removed from the chain: 19 days. The office promises the service online on day 57. RSEM would give √(10² + 12² + 8² + 6² + 2²) = √348 ≈ 18.7 days: on a chain of five activities, the two rules converge. Feeding chain D, scheduled as late as possible, ends on day 27; its 3-day buffer separates it from testing, which starts on day 31. The same network, levelled on the safe durations, finishes on day 76.
Visualisations
| Day | Report | Remaining chain duration | Projected finish | Chain complete | Buffer consumed | Zone |
|---|---|---|---|---|---|---|
| 13 | A finished 3 days late | 28 | 41 | 10/38 = 26% | 3/19 = 16% | Let the team work |
| 24 | B: 3 days remaining | 19 | 43 | 19/38 = 50% | 5/19 = 26% | Let the team work |
| 34 | C: 8 days remaining instead of 1 | 16 | 50 | 22/38 = 58% | 12/19 = 63% | Prepare recovery |
On day 34, the buffer is consumed faster than the chain progresses for the first time: 63% against 58%. The point falls between the two thresholds, which stand at 44.7% and 64.7% at this level of progress.
Cost
| Dimension | Level | Rationale |
|---|---|---|
| Preparation | High | The network, the resource assignment and two estimates per activity come on top of management's agreement to drop activity dates. |
| Execution | Medium | Each owner reports a remaining duration at every progress point; the project manager updates the chart. |
| Documentation | Low | The schedule, the fever chart and the recovery decisions suffice. |
Tooling
A whiteboard and sticky notes are enough to build the network and spot resource contention in a workshop. A distributed team does the same work on an online whiteboard such as Miro or Mural.
A spreadsheet holds the buffer tracking and draws the chart. With about twenty activities, the team levels the resources by hand in the spreadsheet. A spreadsheet is enough for a first project.
Microsoft Project and Oracle Primavera P6 level resources. The team models the buffer there as an activity without a resource and hides the activity dates when it distributes the schedule. Specialised online software such as ProChain or Exepron identifies the chain, places the buffers and plots their consumption.
Sources
- Project Management Institute (2025), A Guide to the Project Management Body of Knowledge (PMBOK Guide), 8th edition: §2.3.2.2 Develop Schedule, §2.3.2.3 Monitor and Control Schedule and §2.3.3.6 Scheduling Approaches and Methods for the place of the method; Section 5, entries "Critical chain method" and "Critical chain project management (CCPM)" for the definition, the identity between critical chain and levelled critical path, the project buffer that replaces activity safety and the buffer protection index; entry "Theory of constraints (TOC)" for the link with the theory of constraints; entry "Resource optimization technique", sub-entry "Resource leveling", for levelling.
- Goldratt, Eliyahu M. (1997), Critical Chain, North River Press: the original book on the method, written as a business novel.
- Theory of Constraints International Certification Organization (TOCICO), "Critical Chain": the certification body for the theory of constraints; CCPM as the replacement of per-activity safety with buffers.
- She, Bingling (2020), Critical chain identification and buffer sizing for efficient project management, PhD thesis, University of Warwick: after Goldratt (1997), student syndrome, Parkinson's law and the cut and paste rule; late scheduling of feeding chains; the resource buffer; the RSEM rule and the excessive buffer of the half rule on long chains; the effect of inserting feeding buffers on the chain; the cost of a buffer that is too short or too long.