
Project management
Critical path method: a 39-day project, step by step
How to find the critical path by hand: build the network, run the forward and backward pass, calculate float, and use the result to decide where extra effort actually shortens a project.
Project planners use the critical path method (CPM) to calculate the earliest finish supported by a set of activity durations and dependencies. The critical path is the longest duration path through that network. A delay on it moves the finish unless the team changes the plan.
The worked example follows eight activities in an expense-system rollout. Calculate the dates by hand, then use the results to compare delays and shorter activity durations.
The example project
Say you're rolling out a new expense management tool across a company. Eight activities, durations in working days:
| ID | Activity | Duration | Depends on |
|---|---|---|---|
| A | Agree requirements | 5 | none |
| B | Select the vendor | 10 | A |
| C | Configure workflows | 8 | B |
| D | Integrate with ERP and HR systems | 12 | B |
| E | Write policy and training material | 6 | A |
| F | Pilot with one team | 5 | C, D |
| G | Train everyone | 4 | E, F |
| H | Go live and support | 3 | G |
The activities total 53 days. Because the team can do some work in parallel, we need the dependencies to calculate the earliest finish.
Step 1: draw the network
Each activity becomes a box, and each dependency an arrow. This is the "activity on node" style most tools use.
Review the network with the people doing the work. In this example, the pilot depends on configuration and integration. If the real pilot also needs the policy, add that link before calculating its dates.
Step 2: the forward pass
Start at the beginning and move right, calculating the earliest each activity can start (ES) and finish (EF).
Forward pass
ES = the largest EF among its predecessors (0 for the first activity)
EF = ES + duration
- A starts at 0 and finishes at 5.
- B and E both start at 5. B finishes at 15, E at 11.
- C and D both start at 15. C finishes at 23, D at 27.
- F needs both C and D, so it can't start until the later of the two: max(23, 27) = 27. It finishes at 32.
- G needs E (day 11) and F (day 32), so it starts at 32 and finishes at 36.
- H runs from 36 to 39.
The earliest finish under these assumptions is day 39. At F and G, use the later predecessor finish because both required activities must be complete.
Step 3: the backward pass
Now start at the end and move left, calculating the latest each activity can start (LS) and finish (LF) without pushing the finish past day 39.
Backward pass
LF = the smallest LS among its successors (the project finish for the last activity)
LS = LF − duration
- H must finish by 39, so it must start by 36. G: 32 to 36. F: 27 to 32.
- C and D both feed F, which starts at 27. So D must start by 15, and C by 19.
- E feeds G, which starts at 32, so E can start as late as 26.
- B feeds both C (latest start 19) and D (latest start 15). The smaller wins: B must finish by 15.
- A feeds B (latest start 5) and E (latest start 26), so A must finish by 5.
At split points (B, and A), the backward pass takes the earlier successor. It's the mirror image of the forward pass.
Step 4: calculate float
Float
Total float = LS − ES (or LF − EF)
Free float = earliest ES of its successors − its own EF
Total float is how long an activity can slip without delaying the project. Free float is how long it can slip without delaying anything, not even the next activity's early start.
| ID | ES | EF | LS | LF | Total float | Free float |
|---|---|---|---|---|---|---|
| A | 0 | 5 | 0 | 5 | 0 | 0 |
| B | 5 | 15 | 5 | 15 | 0 | 0 |
| C | 15 | 23 | 19 | 27 | 4 | 4 |
| D | 15 | 27 | 15 | 27 | 0 | 0 |
| E | 5 | 11 | 26 | 32 | 21 | 21 |
| F | 27 | 32 | 27 | 32 | 0 | 0 |
| G | 32 | 36 | 32 | 36 | 0 | 0 |
| H | 36 | 39 | 36 | 39 | 0 | 0 |
The critical path is A → B → D → F → G → H: 5 + 10 + 12 + 5 + 4 + 3 = 39 days. Here's the same thing as a Gantt chart, with the float drawn in:
What the numbers let you do
Recalculate the same network after each proposed change:
| What happens | Effect on the finish | Why |
|---|---|---|
| Configuring workflows (C) takes 3 days longer | None, still day 39 | C has 4 days of float |
| The ERP integration (D) takes 3 days longer | Day 42 | D is critical |
| The training material (E) takes 25 days longer | Day 43 | E only had 21 days of float |
| Add a second integrator, cutting D from 12 to 9 days | Day 36 | D was critical, so the saving counts |
| Cut D to 7 days | Day 35, not 34 | The path through C (35 days) is now the critical one |
After cutting D to 9 days, C has only 1 day of float left. Reducing D below 8 days leaves the path through C controlling the finish. Recalculate both passes before paying for a further reduction in D.
The other lesson is where not to spend. Adding people to write the training material (E) would cost money and change nothing, because E wasn't holding the project back.
Five planning mistakes to check
- Treating float as the activity's spare time. Float belongs to the project, not the task. If C's team uses its 4 days early, nothing is left for surprises later on that path.
- Ignoring near-critical paths. An activity with 1 or 2 days of float can become critical after a small delay. Compare its float with the uncertainty in its remaining work.
- Assuming unlimited people. Classic CPM assumes every activity can start the moment its predecessors finish. If the same two engineers are needed for C and D, they can't happen in parallel, and your real critical path is longer. That's what resource levelling (and methods like critical chain) deal with.
- Missing dependencies. A forgotten link, like "the pilot needs the policy", quietly makes the plan look shorter than it is.
- Calculating it once. The critical path moves as work finishes early or late. Recalculate at every status update, not just at kickoff.
When durations are uncertain
CPM uses one duration per activity. When you genuinely don't know, a common approach from PERT is to estimate three values and blend them:
Three-point (PERT) estimate
Expected duration = (Optimistic + 4 × Most likely + Pessimistic) ÷ 6
For integration, optimistic 9 days, most likely 12 and pessimistic 21 gives (9 + 48 + 21) ÷ 6 = 13 days. The higher pessimistic input pulls this blend above 12. Using 13 in CPM still gives one finish date; it does not calculate a probability of meeting that date. The schedule-risk example shows how to explore duration uncertainty across this same network.
Sources
- Kelley, J. E. & Walker, M. R. (1959). Critical-path planning and scheduling. Proceedings of the Eastern Joint IRE-AIEE-ACM Computer Conference.
- Kelley, J. E., Walker, M. R. & Sayer, J. S. (1989). The origins of CPM: a personal history. PM Network, PMI.
- U.S. Government Accountability Office (2015). Schedule Assessment Guide: Best Practices for Project Schedules (GAO-16-89G).
- Kramer, S. W. & Jenkins, J. L. (2006). Understanding the basics of CPM calculations. PMI Global Congress.
The project and its numbers are illustrative.