Resource Allocation and Capacity Planning: Worked Example
A team’s capacity is the project hours its people can actually give, which is their scheduled hours minus leave, meetings and every other commitment, counted per person and per skill rather than as one team total. Resource allocation then matches planned work to that capacity week by week. The two decisions that make plans fail are treating hours as interchangeable across skills and loading anyone to 100%. When demand still exceeds capacity, the two standard fixes are resource smoothing, which moves work only within float so the finish date holds, and resource leveling, which moves work to fit the people available even if the finish date slips.
| Question | Short answer |
|---|---|
| Capacity formula | Scheduled hours - leave - meetings - other commitments = available project hours, per person per period |
| Pool the hours? | No. Hours are only interchangeable between people who can do the same work |
| Target load | Below 100%. Queueing arithmetic shows delays rising steeply as load nears 100% (table below); the exact target comes from the team’s own history |
| Leveling vs smoothing | Leveling moves work to match resource limits and may delay the finish; smoothing uses only float and never moves the finish |
| Part-time or shared people | Bookings across projects cannot exceed 100% of the person’s hours (80% + 40% = 120%, so one commitment must change) |
The capacity arithmetic
Start from each person’s working calendar for the period, not from a headcount. Subtract in separate lines so nothing is deducted twice: approved leave, recurring meetings, support or on-call duty, and work for other projects. What remains is the available project time. A sample 40-hour week with 8 hours of leave, 4 hours of scheduled meetings and 6 hours of support duty leaves 22 hours before any contingency.
Capacity is a property of a role and a period. The illustrative table below uses a three-person project in a single ordinary week.
| Role | Scheduled hours | Meetings | Other commitments | Available project hours |
|---|---|---|---|---|
| Editor | 40 | 4 | 6 | 30 |
| Designer | 32 | 4 | 8 | 20 |
| Reviewer | 16 | 2 | 4 | 10 |
| Total | 88 | 10 | 18 | 60 |
The project has 60 available hours, but they are not interchangeable. If the plan needs 15 review hours in a week, the reviewer is 5 short while the editor has spare time. Dividing total effort by total team hours would hide that bottleneck. A schedule that names no resources hides it too: the US Government Accountability Office notes in its Schedule Assessment Guide (GAO-16-89G, December 2015) that “a schedule without resource assignments implies an unlimited number of resources and their unlimited availability.”
Three terms are often mixed up, and they should be kept apart in the plan:
- Available capacity: hours left after leave, meetings and other duties (the last column above).
- Focused capacity: the part of that time spent on the deliverable itself. Sprint and iteration plans should draw on this figure.
- Committed load: what the team has agreed to deliver. It should not exceed focused capacity.
Scrum frames the same idea for iterations. The Scrum Guide says the more the Developers know about “their past performance, their upcoming capacity, and their Definition of Done,” the more confident they will be in their Sprint forecasts, and it limits Sprints to “one month or less.” The measured figures from recent iterations are the best evidence for a team’s own interruption allowance, which is why a fixed productivity percentage borrowed from elsewhere is a weak substitute.
Why 100% allocation fails: utilization and waiting time
Work does not arrive at an even pace, and individual tasks vary in size. The simplest standard queueing model, M/M/1, assumes random arrivals, random service times and a single server. For that model, MIT OpenCourseWare’s manufacturing-systems notes derive the average time a job spends waiting in the queue as Wq = λ / (μ(μ - λ)), where λ is the arrival rate, μ the service rate and ρ = λ / μ the utilization. Because 1/μ is the average service time, the same wait measured in average task durations is ρ / (1 - ρ). The table applies that formula to a range of loads:
| Utilization (ρ) | Average wait in queue, in average task durations |
|---|---|
| 50% | 1 |
| 70% | 2.3 |
| 80% | 4 |
| 85% | 5.7 |
| 90% | 9 |
| 95% | 19 |
| 100% | unbounded |
These are computed values for an idealised single-server model, so they show the shape of the curve, not a forecast for any real team. The shape is what matters: raising load from 80% to 95%, about a fifth more work, multiplies the average wait by 4.75, from 4 to 19 task durations. A specialist who reviews, approves or tests work for several people is a single-server queue of this kind.
That curve is the case for a planning target below 100%. No standard sets the number. Teams choose it from their own data: how much unplanned work arrives, how often the specialist is interrupted and how costly delay is. The worked example below uses 85% for illustration. A team with heavy support duty would set it lower, and a team with steady, predictable work could set it higher.
Resource leveling and resource smoothing
Both terms name ways to change a schedule so that demand for people stays within supply. PMI’s PMBOK Guide treats them as two separate resource optimization techniques, and MPUG’s PMP exam-prep explainer sets out the difference. In plain words, resource leveling adjusts start and finish dates to fit the people actually available. It is the tool when resources are limited, available only at certain times or over-allocated, and it can lengthen the critical path and push the finish date back. Resource smoothing works to a fixed finish date: it moves only activities that have float, keeps demand under a set limit and leaves the critical path unchanged. When float is short, smoothing may not clear every peak.
| Resource leveling | Resource smoothing | |
|---|---|---|
| Trigger | Resources are limited, available only at certain times, or over-allocated | Demand peaks above a set limit while the finish date is fixed |
| What moves | Start and finish dates, even on the critical path | Only activities that have float |
| Finish date | May move later | Does not move |
| Critical path | May change | Does not change |
| Float | Absorbed first, before the finish date moves | Reduced on the activities that move |
The GAO guide gives the practical version and its limits. Leveling, it says, “adjusts the scheduled start of activities or the work assignments of resources to account for their availability,” and the organization that controls the resources uses it mainly “to smooth spikes and troughs in resource demands created by the sequencing of activities in the schedule network.” The guide’s own example of an inefficient plan is one that “requires 30 software engineers one week, none the next week, and 25 the next.” It also advises that:
- activities delayed by leveling typically have the greatest free float and the fewest assigned resources;
- “resource leveling should not delay a completion date if possible,” and if it does, resources are “an inevitable constraint” and “an alternative approach to the project plan is needed”;
- leveling should be applied only to detailed schedules with resource estimates supported by historical data, never to summary schedules;
- changes made to resolve resource conflicts should be thoroughly documented and understood by everyone, because they can alter float and critical path calculations.
The float in these rules comes from the schedule logic described in the critical path method guide. Without that network there is no way to tell which activities can move without changing the finish.
A four-week worked example
In this illustrative case, a three-role team (editor, designer, reviewer) is producing a documentation site. Available hours per week come from the capacity table above, except that the designer is on leave for 8 hours in week 3. The planning target is 85% of available hours, and all weekly demand figures are illustrative.
| Wk 1 | Wk 2 | Wk 3 | Wk 4 | Total | |
|---|---|---|---|---|---|
| Editor available | 30 | 30 | 30 | 30 | 120 |
| Designer available | 20 | 20 | 12 | 20 | 72 |
| Reviewer available | 10 | 10 | 10 | 10 | 40 |
| Editor demand | 26 | 24 | 20 | 16 | 86 |
| Designer demand | 10 | 22 | 18 | 8 | 58 |
| Reviewer demand | 2 | 6 | 14 | 16 | 38 |
Team demand is 182 hours against 232 available, a 78% load that looks comfortable. The per-person, per-week view does not.
| Load (demand / available) | Wk 1 | Wk 2 | Wk 3 | Wk 4 |
|---|---|---|---|---|
| Editor | 87% | 80% | 67% | 53% |
| Designer | 50% | 110% | 150% | 40% |
| Reviewer | 20% | 60% | 140% | 160% |
The designer is over capacity in weeks 2 and 3, and the reviewer in weeks 3 and 4, even though the editor is below the 85% target after week 1. The 85% target gives plannable hours of 25.5 for the editor, 17 for the designer (10.2 in week 3) and 8.5 for the reviewer each week.
Step 1: smooth what has float
Assume the designer’s work in weeks 2 to 3 includes tasks that do not depend on approved copy and so carry float. Moving 4 hours earlier into week 1 and 9 hours later into week 4 brings the designer’s plan to 14, 17, 10 and 17 hours, all within the 85% line. The finish date does not move and the critical path is unchanged, which is smoothing in the PMBOK sense. The editor’s 0.5-hour overrun in week 1 is cleared by moving one hour to week 4 (25, 24, 20 and 17 hours).
Step 2: level what has no float
The reviewer cannot review work that does not yet exist. Two hours of outline review can move into week 1, and week 2 review can rise to the 8.5-hour line, but that still leaves the reviewer with 4, 8.5, 8.5 and 8.5 hours, or 29.5 of the 38 needed. The remaining 8.5 hours cannot be absorbed by float, so this is a leveling decision with a real cost. Three options are open:
| Option | Effect on finish | Cost |
|---|---|---|
| A. Bring in a second qualified reviewer for 8.5 hours in weeks 3-4 | None | Needs a named person and a booking agreed with their own project owner |
| B. Add a fifth week for the remaining review hours | One week later | Delivery date moves, and every dependent owner must be told |
| C. Reduce review scope to a risk-based sample | None | Less coverage; the owner must accept the lower assurance in writing |
None of these is the plan to choose automatically. Under the GAO guidance, option B is what resource leveling produces when a resource is a true constraint, and the date change has to be documented. Options A and C change the plan instead of the date. The point of the exercise is that the choice becomes visible in week 1 rather than in week 4.
How to allocate: named people, skills or pools
The GAO guide describes three levels at which a schedule can assign labor, and each has a cost.
| Level | Example | Advantage | Disadvantage |
|---|---|---|---|
| Named individual | ”Bob Smith Jr.” | Tracks one person’s effort and catches overallocation of a scarce expert | Needs constant updating for vacations and sick leave; less flexible to level |
| Skill set | ”software engineer level 3” | A compromise between too much and too little detail; management can rearrange people within a skill | Obscures specific labor rates; slower to set up than organization level |
| Organization | ”agency test department” | Quickest to set up | One unit holds several skill sets, which can complicate matching skills to activities |
Name-level assignment suits the scarce specialists, such as the reviewer above, whose absence would delay the project. Skill-level assignment suits interchangeable roles. The same guide recommends storing resource information in the schedule file, because otherwise “the effects of resource availability must be manually updated, increasing the likelihood of error.”
Shared specialists and the cost of switching
For a person split across projects, record the booking, not only the percentage. “Tuesday afternoon, four hours” tells both owners when a review can happen, while “40%” does not. A person who is 60% on one project and 40% on another has a full allocation; 80% and 40% is an overbooking, and one owner must release hours before either plan is credible. The GAO guide’s Arizona border surveillance case study (GAO-14-368) shows what happens otherwise. Personnel were shared among the surveillance programs, in part delaying them; one program had to share a contracting officer with another; and officials delayed planning and deployment activities because of resource constraints and a lack of dedicated contracting officers.
Splitting also costs capacity that no timesheet shows. Sophie Leroy’s experiments, published in Organizational Behavior and Human Decision Processes (2009, vol. 109, pp. 168-181), found that people who switch before finishing a task keep thinking about it, which impairs performance on the next task. The practical consequence for allocation is to give a person longer blocks on fewer projects where the work allows it, and to put the switching loss into the other-commitments column when a person must be shared.
Putting availability beside dependencies
A weekly total can look adequate while the sequence fails. If the designer’s 20 hours fall on Monday to Wednesday but approved copy arrives on Thursday, the plan is infeasible regardless of the total. Either bring copy approval forward, reserve later design time or change the deliverable. Checking the capacity table against the dependency network, such as the one built in the project timeline estimator or from a work breakdown structure, exposes the conflict before work starts.
Milestones belong in the same check. The GAO guide advises examining resource peaks against major milestones; peaks that fall just after a milestone, rather than just before it, are a sign that resources may need to be reallocated. See project milestone planning for how to set those checkpoints.
Replanning without hiding the change
Compare planned and actual hours at every reporting cycle. If support work used six more hours than expected, record the cause and reduce the remaining forecast by that amount instead of assuming overtime will recover it. Decide which deliverable or date changes and tell the affected owners. A running record of interruptions, skill bottlenecks and review queues is the evidence behind the next utilization target.
A release-level capacity check follows the same arithmetic over a longer horizon: list the skills each milestone needs, total the hours by skill and by period, and mark every period where demand exceeds the plannable hours. Gaps can be closed by hiring, cross-training, moving work earlier into float, reducing scope or moving the date, and the choice should be recorded with the figures that forced it.
Sources
- GAO, Schedule Assessment Guide: Best Practices for Project Schedules (GAO-16-89G, December 2015): Best Practice 3, resource loading and leveling.
- Scrum Guide: capacity in Sprint Planning and Sprint length.
- MIT OpenCourseWare 2.854, M/M/1 queue notes (Fall 2016): utilization and waiting-time formulas.
- MPUG, PMP Prep: Resource Leveling and Resource Smoothing: how the PMBOK Guide separates the two techniques, with worked examples.
- Leroy (2009), attention residue when switching between work tasks: Organizational Behavior and Human Decision Processes 109(2).