Chapter 19: Plan Capacity, Roles, and Resources

The plan that added up

It is month seventeen at Meridian, and the resource review has produced a plan that adds up, which is exactly what worries Dana Okafor. The go-live calendar is fixed: clinic one went live in month fifteen and is in hypercare, clinic two goes live in month eighteen, clinic three in month twenty-one, and the grant pays on access outcomes, so the calendar is also the funding. The resource plan is a table of person-weeks, four rows, four months, and every number fits. Four trainers, sixteen person-weeks of capacity a month, and the heaviest month, clinic two’s go-live, demands eight. Six clinical super-users releasable from the existing clinics, twenty person-weeks, and the heaviest month demands seventeen. Marcus Chen’s platform team, twelve person-weeks, against nine at configuration peaks. The privacy reviewer, three person-weeks, against two. Every row under its capacity, every month under its ceiling, the whole plan green.

Nora, who owns operational acceptance, asks the question that turns the review. “This says we have enough people. Do we?” Dana does what the plan does not: she opens the calendar, then the roster, then the floor. Clinic two’s go-live needs four super-users on the floor for the first two weeks, full days, at the satellite site on the eastern edge of the network. The two super-users whose home clinics sit closest to that site are Amina and Tunde, and they are the named leads at clinics three and four, the two existing clinics whose floors cannot run without them on their busy days. Their releasable capacity is two weeks a month, not four, and the plan’s twenty person-weeks quietly assumed those four would fall inside the go-live’s window. The last week of the month is audit week at every existing clinic, claims, stock, and compliance, and two of the four full-time releasables must be home that week, so the effective floor drops from five to three. The plan says the go-live month needs seventeen person-weeks against twenty available. The calendar says the go-live’s first two weeks run at zero slack, and the audit week runs short by two.

The plan added up. The project does not. And this is the whole subject of the chapter: a resource plan is not a wish that fits in a table. It is a set of explicit choices about capability, availability, focus, and ownership, made at the level where the constraint actually lives, which is weeks, locations, and shifts, not months, headcount, and totals. Chapter 17 built the schedule, chapter 18 built the money, and both landed on the same sentence: the eastern acceleration at BlueLine needs gangs and specialists the corridor has not confirmed, the cash curve depends on capacity that must exist. This chapter is that sentence. It replaces wishful resource plans with plans that survive contact with calendars, rosters, and the people who are already busy.

People are not the only resource

The first discipline is to see the full inventory, because a project can have every person it needs and still stall for want of something that was never in the plan. People are the resource that talks back, and they are not the only one. The clinic program needs equipment, the cardiac monitors and the pharmacy shelving that the fit-out schedule assumes will arrive when the crews are on site. It needs facilities, the training room that every go-live reserves at the same time, and the satellite site’s floor space where the four super-users must actually stand. It needs materials, the clinical supplies for the new clinics, the printed workflow guides, the hardware for the record platform. It needs data, the migration files, the test datasets, the reference tables that the privacy review gates on. It needs environments, the test instance of the patient-record platform that the configuration team books, the sandbox where the integration is rehearsed. Each one has a capacity, an availability, and a lead time, and each one fails differently: the equipment arrives, the facility is booked, the environment is down, and the project waits anyway.

BlueLine’s corridor is the standing proof. The phased opening that chapter 17’s decision produced, the central and northern segments at month twenty-five, concentrates four different kinds of capacity into the same three months: the civils gangs finishing the remaining pours, the systems and ticketing integration specialists working against the frozen design, the commissioning crews who can only commission what is physically ready, and the acceptance inspectors from the transport authority, a fixed number of bodies with a fixed number of working days, who must certify the segments before the bus operator can take them. The resource plan that only counts people misses the acceptance inspectors entirely, because they are not on the project’s payroll, they belong to Marta Reyes’s authority, and their calendar is not the corridor’s calendar. The corridor can hire gangs. It cannot hire acceptance capacity, it can only request it, and the request has a queue.

So the inventory is the first artifact, and its minimum viable form is a list, not a model: every category of resource the work depends on, people, equipment, facilities, materials, data, environments, licenses, and the external capacity that is requested rather than owned, with for each one the same three facts that apply to people: what it can do, when it is available, and who owns the schedule for it. The list feels administrative until the day a project discovers its test environment is double-booked for the fortnight before the release, and then it is the whole project. The chapter’s standing rule: if the plan names only people, the plan is not a resource plan, it is a headcount table.

Headcount is not capability

The second discipline is the one the case lens of this chapter was built on: Meridian has enough trainers by headcount and not enough clinical super-users at the required locations and times, and the two facts are not in tension because headcount and capability are different currencies. A headcount is a body with a name. Capability is what that body can actually do, at what proficiency, in what context, and the difference is where resource plans go to die.

The trainers are the easy case. Four trainers, each capable of running the six-week go-live support window, and the demand is two per go-live, so the numbers work. But capability is not uniform inside the group: one of the four is the only one who has configured the platform’s workflow engine, and the plan that treats the trainers as interchangeable will discover the difference in the week before clinic two’s go-live, when the workflow standard from chapter 8’s first gate needs a change and the config-capable trainer is on leave. The super-users are the hard case. Ten clinical super-users across the six existing clinics, six releasable for go-live support, and the plan reads “six” as six identical units of support. They are not. Amina and Tunde are the only two who have run a full go-live floor before, clinic one’s, which makes them the only two qualified to lead clinic two’s floor. Two of the other four are platform-proficient but thin on the clinical workflow side. One is superb at training and weak at troubleshooting. The capability matrix is the instrument that turns “six super-users” into what is actually available, and its minimum viable form is a table with a row per person, a column per capability, and a level per cell, with the levels defined before the table is filled, because “can do” and “has done under pressure” are different grades, and the go-live floor needs the second one.

The matrix earns its name only if it is used for the decisions it serves. It decides assignment, who can be placed where, and it decides development, who is one step from being able, and it decides risk, which capabilities exist in exactly one person. A matrix that is filled and filed is a headcount table in costume. A matrix that is walked at the resource review, with the red cells, the single-holder capabilities, read aloud, is the difference between the plan that adds up and the plan that works. The decision each row supports is who, exactly, and the question that exposes the wishful plan is always the same: name the person. “We have trainers” is a claim about the payroll. “Amina, Tunde, Fatima, and Sam run the go-live floors, and Amina is the only one who can also configure” is a claim about the project.

Net capacity: the arithmetic of availability

The third discipline is arithmetic, and it is the one that turns the month-17 plan red. Capacity is not headcount times the calendar. It is headcount times capability times availability times focus, and each multiplier is smaller than one. The plan that omits any of them overstates what the project can actually do.

Availability is the multiplier that breaks Meridian. A person’s month is not a person’s project month. The super-user’s month contains twenty-two working days, and it also contains the two days of leave, the training day the network mandates, the monthly clinical meeting, the admin day that the payroll system quietly assumes, and the other programs, the quality initiative, the audit preparation, the roster cover at the home clinic, that have claims on the same person. Netting it down: twenty-two working days at 80 percent availability is 17.6 person-days a month, against the 20 the plan’s four-week assumption silently used, and the plan has overstated the entire resource base by 12 percent before anyone has looked at a calendar. That is not a rounding difference. It is the difference between a plan with slack and a plan at zero, and it is exactly the difference the go-live floor found.

The arithmetic of availability has three rules. First, net before you plan: every person’s month starts at the working days, not at the project’s need, and the project claims a share, not the whole. Second, availability is a property of the calendar, not of the person: the audit week, the quarter-end, the school holidays, the rainy season, the partner’s release window, all of it lives on a calendar, and the plan that averages it away will be surprised by it. Third, the level of detail must match the level of the decision: monthly totals are fine for the question “do we have enough people this quarter,” and they are actively dangerous for the question “can four super-users be on the eastern floor on the third Tuesday,” and the resource plan must be re-sliced to the week, the location, and the shift whenever the decision is that specific. The month-17 plan is green at the monthly level and red at the weekly level, and the red is the information, not the artifact.

The demand-versus-capacity heat map is the instrument that makes the red visible, and its minimum viable form is the table the review started with, demand over capacity, by role and by period, with the periods chosen for the constraint. The monthly version at Meridian:

Every cell is under capacity. The highest utilization is the super-user row at month eighteen, 17 over 20, 85 percent, which sounds like a project that is busy but fine. Then the heat map is re-sliced to weeks, which is the level where the go-live actually happens, and the picture changes. Clinic two’s floor needs four super-users full time for the first two weeks, one more for clinic one’s hypercare evening shift, five person-weeks of demand a week. The effective capacity in a normal week is five, four full-time releasables plus the two capped leads contributing one person-week between them, so the go-live runs at exactly full, zero slack, every week, for the first two weeks, and a single illness, a single audit conflict, a single late roster change, tips it over. The audit week cuts capacity to three, two of the four full-time releasables bound to their home clinics’ claims review while the two capped leads still contribute their combined one person-week, and the demand stays at five: short by two person-weeks in the single week that the clinic cannot afford to be short. And the location column of the map shows why the total of twenty person-weeks is a fiction: the super-users whose home sites sit nearest the satellite location are the two capped leads, Amina and Tunde, and their floors bind on the busy days, the reason their releasable capacity is two weeks a month and not four. The plan assigned them to the floor. The calendar reassigned them home.

The heat map teaches its lesson by construction: it is only as good as its periods and its rows, and the craft is choosing both for the constraint. The row must be the bottleneck role, the role whose absence stops the chain, which for Meridian is the super-user, not the trainer, and the period must be the period of the constraint, the week, not the month. A heat map with the right roles and the wrong period is the month-17 plan: green everywhere, and wrong.

Utilization is the number that the heat map produces and the number that must be read with suspicion. A utilization of 85 percent on the bottleneck role sounds healthy, and it is a queue in disguise, because under variability, waiting time grows nonlinearly as utilization approaches capacity, the result that chapter 17 built from the queueing literature, and the same logic applies to people. A super-user at 85 percent average utilization is a super-user who has no slack for the audit week, the illness, the request that arrives, and the slack is the only thing that absorbs the variability. The practical rule that follows: the bottleneck role should be planned below the ceiling, with deliberate slack, because the slack is not waste, it is the project’s shock absorber, and the go-live that runs its critical role at 100 percent is a go-live that has already spent its contingency without opening a register.

The cost of multitasking belongs in the same arithmetic, because focus is the fourth multiplier. The double-booked QA specialist at KijaniPay, two people, two urgent programs in November, the pilot follow-up and the fraud-control volume test, is not half a person on each; the switching costs more than the split, because every context switch carries a re-entry cost. The evidence is concrete: in a 2008 study of interrupted knowledge work, Gloria Mark, Daniela Gudith, and Ulrich Klocke observed that participants took an average of about 23 minutes to return to a primary task after an interruption, and reported more stress, frustration, and time pressure. Twenty-three minutes is the re-entry cost of one switch, and the project that loads a person into two full-time commitments has built a machine whose job is switching, not working. The chapter 17 lesson holds in the resource plan as much as in the schedule: a person at 100 percent in two plans is 200 percent of a person, and the plan is where that double-booking should have been visible before November arrived.

Roles: ownership you can point at

The fourth discipline is roles, and the distinction that matters is between a job title and a role. A title says what someone is. A role says what someone is accountable for, what they may decide, and what interfaces they own, and it is the third part, the decision rights, that most resource plans omit, because only the governance of chapter 8 can answer them. Meridian’s blank cell is the longitudinal proof. At the first gate, the charter’s operational ownership cell was empty, and chapter 8 named the cost of the blank: the clinical workflow standard, the decision about how a clinic will schedule, book, check in, escalate, and document, went undecided while the platform team waited. By month seventeen the cell has a name, Nora, and the go-live acceptance decision sits with her, with Hana the clinical director, and the evidence comes from Dana and Marcus. The role exists on paper. And the role’s authority does not reach the resource plan: Nora can refuse to accept a go-live, and she cannot move a single super-user onto the floor, because the super-users report to the clinic managers, the clinic managers answer to operations, and operations has its own calendar, its own audit week, its own floors. The go-live readiness decision and the go-live capacity decision live in different hands, and the resource review is where the two must be reconciled, which is why the review is a governance meeting, not a scheduling exercise.

The responsibility assignment matrix, the RACI, the Responsible, Accountable, Consulted, Informed matrix, is the standard instrument for naming roles, and it deserves both its popularity and its limits. Its minimum viable form is a table with the decisions and deliverables as rows and the roles as columns, with exactly one A per row, the account-holder who answers for the outcome, and the R’s, C’s, and I’s around it. The matrix earns its place by forcing the single accountable owner, the one name the row can be escalated to. A row with three A’s is a row with no A, because the escalation has nowhere to land, and a row with no A at all is a row whose outcome will be decided by whoever shouts loudest in the last meeting. The go-live acceptance row at Meridian is the clean example: A is Nora, with Hana, and the R’s are Dana and Marcus and the go-live floor, and the single A is what makes the acceptance decision possible to make.

The limits of the matrix are as important as its use, and they are four. First, it cannot see workload: the super-users’ column can hold six R’s and the matrix will look perfect while the floor is short, because it records who is responsible, not how much is on them. Second, it cannot see authority: Nora’s A can sit next to a decision she cannot enforce, because her accountability runs through operations and the capacity sits in operations’ hands. Third, it is a snapshot, not a flow: it names who owns the decision, not what happens when it is contested, and the escalation path lives in chapter 8’s governance, not in the matrix. Fourth, it is only as good as its rows: a matrix built from the deliverable list misses the decisions that have no deliverable, the release date, the priority call, the resource allocation, the blank cell in another costume. The honest use is to build the matrix from the decision list, the questions the project must actually answer, and then to walk it for the three failures, the row with no A, the row with three A’s, the column that is overloaded. Those three walks find the role problems the meetings cannot.

The responsibility map is the companion instrument, and it answers the question the matrix cannot: for each role, what can this person decide without asking, what must they escalate, and who is their escalation, in one sentence each. Its minimum viable form is the matrix plus one line per role, written in the role’s own language. Amina’s line: “Runs the go-live floor, can reassign super-users within the clinic’s roster, cannot change the go-live date, escalation to Dana.” Nora’s line: “Accepts go-lives on the readiness evidence, can delay an acceptance, cannot move capacity, escalation to Elena Marchetti.” The map turns the matrix from a table into an operating system, the place where each role’s authority meets its limits, and it answers the question the month-17 review could not: who decides, and what can they actually do about it.

Smoothing, leveling, and the priority meeting

The fifth discipline is the demand side, because capacity planning is two-sided, and the project that only manages supply is managing half the problem. The demand side has two instruments, and they are different in exactly the way the schedule’s two models of time were different in chapter 17. Resource smoothing moves work within its float, the slack that chapter 17’s network found, shifting an activity to a less-loaded week without moving the end date. It costs nothing but coordination. Resource leveling moves work that has no float, pushing an activity out to match the capacity that exists, and it moves the end date, and it is a schedule decision, which means it is a commitment decision, which means it belongs to the governance of chapter 8, not to the resource planner. The distinction is the whole craft: smooth what has float, level what has none, and know which one you are doing, because the plan that levels quietly, without a decision, has changed the date without telling anyone.

BlueLine is the case, and its numbers are the corridor’s. The phased opening concentrates the systems and ticketing integration, the commissioning, and the acceptance inspection into months twenty-four through twenty-six, and the demand curve for those three roles spikes exactly where the eastern segment’s work is still running, because the eastern culvert package, committed early to beat the wet season in chapter 18, needs civils gangs in months twenty-one and twenty-two while the wet-season window is open. The resource plan’s job is to see both curves on one page, the demand from the schedule and the capacity from the people, and then to work the moves in order: smooth the commissioning within its float, pull the integration work earlier against the frozen design, which chapter 17 already did for the schedule and the resource plan must do for the people, and then level only what remains, with the date consequence priced and presented. The gangs and specialists that chapter 18’s close said the corridor had not confirmed are the resource plan’s deliverable: named, confirmed, and calendared before the cash curve needs them, because the money of chapter 18 and the capacity of this chapter are the same plan in two ledgers.

The priority conflict is the demand-side failure that no instrument fixes, because it is a governance problem wearing a scheduling costume. Two projects need the same specialist in the same fortnight, KijaniPay’s pilot follow-up and the fraud-control volume test both want the two QA specialists in November, and the question “who gets the QA” is not answerable by leveling, because there is no float to smooth into, and it is not answerable by the specialist, because asking the specialist to choose between the pilot and the control test is asking the person with the least authority to make the organization’s priority decision. The answer belongs where chapter 8 put it: the priority conflict escalates to the governance that owns the portfolios, with the options and the consequences priced, the pilot date versus the control test window, the merchant impact versus the fraud exposure, and the decision is made in the light, by the people with authority over both. The chapter’s rule is sharp: a specialist asked to choose between two commitments has been failed by the plan, and the fix is not a better calendar, it is a better escalation.

Build, borrow, buy, partner, or automate

The sixth discipline is the sourcing decision, and it is the one the month-17 review was circling: the gap exists, the question is how to close it, and the five moves are build, borrow, buy, partner, or automate, with a time to capability, a cost, a quality risk, and a context for each, and the craft is matching the move to the window.

Buy is the move everyone reaches for first, and it is the slowest: a hired super-user must be a clinician, must be vetted, must be onboarded, must be trusted by the floor, and the honest time to capability is four to six months, recruitment, notice, onboarding, four weeks of shadowing and four of supervised work before they can run a floor alone. Buying to close a gap two months away is not a plan, it is a hope with a payroll. KijaniPay’s contractors are the same story in another register: hiring engineers for the pilot, the man-month myth of chapter 17, when the constraint is focus and a blocked partner, adds bodies to a system that cannot use them, and the onboarding consumes the team’s attention in the weeks the pilot needs it. The buy decision has a rule: buy when the need outlasts the onboarding, hire for clinic four, month twenty-four, not for clinic two, month eighteen.

Borrow is the fast move, secondment, loan, cross-coverage, and it arrives with a capability discount: a super-user borrowed from the sister network knows the platform but not Meridian’s workflow standard, knows the floor but not the clinic’s roster, and the discount must be priced into the plan, half a person for the first two weeks, not a full one. The Northstar response runs on borrowed capacity, field staff, logistics partners, government vehicles, with the same discipline: the borrowed resource comes with its own owner, its own priority, its own calendar, and the project that treats borrowed capacity as owned will discover the difference on the day the sister network pulls their person home. Borrowing has a rule: borrow for the window, not for the role, and write down what the borrower cannot do.

Build is the investment move, training existing people to the missing capability, and its arithmetic is the trainer’s calendar. Two clinicians from clinic three, trained as super-users, need the six-week training program and four weeks of supervised floor work, ten to twelve weeks to first-line capability: starting now, in month seventeen, lands them in time for clinic three’s go-live in month twenty-one and nowhere near clinic two’s in month eighteen. The build costs trainer capacity, which Meridian has, sixteen person-weeks against a peak of eight, and the trainer slack is exactly the resource the build can spend. The rule: build when the capability will be needed more than once, because the two trained clinicians are the answer to clinics three, four, five, and six, and the one-off hire is the answer to nothing.

Partner is the structural move, contracting the capability from an organization that already has it, the platform vendor’s change team running the go-live support, the commissioning contractor bringing the inspectors, and it moves the capability question from the project’s payroll to the contract’s, which is where chapter 20 begins. Meridian already partners: Marcus Chen’s platform team is partly internal and partly the vendor’s, and the vendor’s change consultants are a standing option for go-live support, with the quality risk that the vendor’s people know the product and not the clinic, and the control question that the vendor’s calendar answers to its own commercial interest. Partnering’s rule: the contract must specify the capability, not the hours, because “two consultants for six weeks” buys attendance, and “a floor that can be run to Meridian’s workflow standard” buys the thing the project actually needs.

Automate is the capability move that most resource plans forget, and it is the move that actually closes the Meridian gap. The e-learning modules, the on-screen workflow guidance, the digital checklists that the trainers can build in their slack weeks reduce the floor support the go-live needs from four super-users for the whole window to four for the first two weeks and three from week three: a demand reduction, not a supply addition, and the only move that costs nothing on the calendar. The automation rule is the clearest of all: automate what is repeatable and safe, the reference material, the checklist, the workflow prompt, and never automate the judgment, the moment the patient’s record does not look right, because the floor’s value is the person who can see that, and the automation gives that person more time rather than replacing them. KijaniPay’s QA question is the same: the fraud-control test’s repeated runs are automatable, a harness the two QA specialists build once and run nightly, and the automation converts the double-booking from a conflict into a schedule, with the human review staying where the judgment is.

The month-17 decision at Meridian, the one the chapter has been building toward, is the five moves priced together. The gap is two person-weeks in the audit week and zero slack in the go-live weeks. The automate move reduces the floor from four to three from week three, which takes the audit-week demand from five to four, against capacity three, short one. The borrow move, one super-user from the sister network for the audit week, two weeks, closes the one. The build move starts now, two clinicians from clinic three, ten to twelve weeks, ready for clinic twenty-one, which is the move that makes the next go-live the last one that runs this tight. The buy move is refused for clinic two, because four to six months to capability against a go-live two months away is arithmetic, not strategy. The rephase move, clinic two to month nineteen, is priced at its cost of delay, the clinic-months of eligibility the grant pays on, about 375,000 units for six weeks, which is more than the borrow and the automation combined, and the grant pays on evidence, so the delay also delays the evidence the grant reimburses on, the compounding cost that chapter 18’s decision room priced in another register. The plan that closes the gap is the plan that used all five moves in the order the window allows: automate, borrow, build, refuse the buy, price the rephase, and never once pretended that headcount was capacity.

The one-person risk

The seventh discipline is the resource risk that hides in plain sight, because it lives in the same sentences that sound like good news. The plan names Amina as the lead on the go-live floor. The plan names the config-capable trainer as the only one who can change the workflow engine. The plan names the integration engineer at BlueLine as the only one certified on the ticketing system. The plan names the settlement engineer at KijaniPay as the only one who knows the banking interface. Each sentence is a risk statement wearing a capability’s clothes, and the risk is the same: knowledge concentration, the capability that exists in exactly one person, which means the project is one resignation, one illness, one week of leave from losing a capability it scheduled. The tell is easy to see in retrospect and easy to miss in the review: the same name appears in every milestone, every risk row, every heat map’s bottleneck cell, and nobody says the sentence aloud, “if this person leaves, this project stops.”

The response is not a lecture about redundancy, it is a plan with three parts. First, name the concentration: the capability matrix from this chapter, the red cells, the capabilities held by exactly one person, read aloud at the review, is the instrument that turns the feeling that someone is indispensable into a list. Second, build the understudy: every named specialist carries a named second, someone who shadows, who is deliberately given the supervised exposure, who can cover the absence at a known quality level, and the specialist’s exit is planned the same way the go-live is planned, with a transition period, a checklist, and a date. Third, price the onboarding so the understudy is real: a new person contributes nothing in the first four weeks, shadowing, learning, asking, and about half capacity in the next four, supervised, so a six-week notice period with a handover is a best case, and the plan must know the difference between the notice and the capability transfer.

The arithmetic of onboarding is what disciplines the whole practice: two clinicians trained as super-users, ten to twelve weeks to first-line capability, six weeks of notice from Amina, which is not enough for the program’s clinic-two window and exactly enough, if the build started in month seventeen, for clinic three. The onboarding curve is the reason the build decision is a now decision: the capability the program will need in month twenty-one must be started in month seventeen, and the resource plan that waits until the need is visible has already missed the window, which is the resource version of the estimate cone from chapter 15, the capability that narrows only when the investment is actually made.

What the machine can do with the capacity files

The assistant has a genuine and bounded place in the capacity work. It can build the demand-versus-capacity heat map from the provided schedule and people data and re-slice it to any period the leader names, weekly, per site, per shift, which is exactly the re-slice that caught Meridian’s audit week. It can flag the overloads, the person loaded above 100 percent, the role at zero slack, the matrix rows with no accountable owner or three, the capability held by exactly one name. It can draft the onboarding plan from the ramp-up curve and generate the sourcing comparison, the five moves priced against the window, from the provided costs and lead times. The source data is the approved, non-confidential capacity and role content, redacted before prompting; the outputs are drafts until a named owner verifies them; and the audit record says what was generated, from what, checked by whom, and decided by whom.

Four boundaries matter, because people are where the machine’s confidence is most dangerous. First, the personnel data is sensitive: medical rosters, leave, performance, and the records of named individuals do not belong in unapproved systems, and the capacity files stay on the project’s owned infrastructure, aggregated and redacted before any prompt. Second, the judgment of who is the right person is human: the machine can rank by hours and grades, and it cannot know that Amina is the only person the clinic three floor will listen to in the first week, which is trust, not data. Third, the release negotiation is human: the machine can compute what the sister network’s loan would cover, and the conversation that gets the sister network to lend is a human one, and the borrowed person’s loyalty to their own program is a human fact the model cannot price. Fourth, the priority decision is human: the machine can present the pilot and the control test side by side, and the decision about which one matters more belongs to the governance of chapter 8, because it is a judgment about value, not a computation about capacity. The verification is the room: the leader walks the heat map with the team leads, tests the machine’s flags against the roster, the calendar, the audit week, and makes the sourcing and priority decisions in the light, while the machine accelerates the slicing and the flagging, and the person, the release, and the priority stay human.

The capacity story can lie in four ways

The failure patterns of resource planning deserve to be named as characters, because each one is produced by competent people doing what the room rewarded, and each one has a signal that reveals it, and each one is a story being told with a table.

The headcount accountant counts bodies and calls them capacity. The plan says six super-users, and the word six carries the whole argument; the review never asks what the six can do, when they are available, or who owns their calendars. The tell is the meeting that opens the table and never opens a roster. The cost is the go-live that has enough people and nobody who can run the floor, clinic two in embryo, caught only because the review asked the second question.

The availability optimist loads everyone at 100 percent and calls the plan full. The person’s month appears as twenty working days of project work; the leave, the meeting, the other program, the audit week, are someone else’s problem. The tell is the capacity plan with no netting-down step, the utilization column that never appears, the resource plan that cannot answer “what does Amina do on the last Thursday of the month.” The cost is the utilization trap of chapter 17 in people: a busy team with no slack that breaks on the first surprise, which is the audit week.

The role smuggler carries responsibility without authority and calls it ownership. The matrix puts the A on Nora’s row and the capacity stays in operations’ hands, the acceptance decision existing without the power to enforce it, or the reverse, the authority without a name, the release date everyone agreed to and nobody owns. The tell is the decision that takes two weeks because the answer requires a meeting that requires a meeting, the blank cell that chapter 8 named and that reappears in a new row every time the matrix is rebuilt. The cost is the escalation that arrives late, the go-live refused after the floor was scheduled, because the person with the A and the person with the capacity never sat in the same review.

The key-person gambler bets the plan on the one name. Amina leads the floor, the integration engineer owns the ticketing, the settlement engineer knows the banking interface, and every milestone carries the same signature. The tell is the name that appears in every risk row and the sentence that nobody says, “if this person leaves, this project stops.” The cost is the resignation that empties the risk register’s top row in one email, the six-week notice that was never an exit plan, the capability that walked out the door with the person who held it, the mastery drill of this chapter.

The four characters share a root, and it is the chapter’s thesis restated: each one replaced an explicit choice with an implicit assumption. The headcount accountant assumed capability, the availability optimist assumed the calendar, the role smuggler assumed authority, and the key-person gambler assumed permanence. The controls that keep each one honest are the chapter’s instruments used as designed: the capability matrix walked for its red cells, the net capacity computed before the plan, the responsibility map that writes each role’s authority and limits in one line, and the understudy rule that names a second for every named specialist, reviewed at the same cadence as the schedule and the money. The capacity story does not need to be comfortable. It needs to be explicit, because the decision the room makes on an explicit capacity story is a decision the project can live with, and the decision it makes on an implicit one is a decision the calendar will make for it.

Practice

One. A quick classification. For each statement, name the resource-planning failure or discipline it reveals, and say what the fix would be. (a) “We have five nurses trained on the platform, so the go-live is covered.” (b) “Two of them are in clinics all day and can join the go-live support only in the evenings.” (c) “The plan loads every trainer at 100 percent for the next three months.” (d) “The settlement engineer’s name appears in every milestone at KijaniPay.” (e) “The RACI row for ‘approve go-live’ lists three A’s.” (f) “The e-learning module replaces one super-user on the floor from week three.”

(a) is headcount presented as capability: five trained nurses says what they can do, not when they are available, where they are based, or how the capability is distributed, and the fix is the capability matrix with the levels defined, which would show that two of the five have never run a floor under pressure. (b) is availability, and the fix is the net-capacity arithmetic, the calendar before the plan, which converts “five nurses” into the person-days that actually exist. (c) is the availability optimist, and the fix is the netting-down step, working days times availability, with the slack named, because a 100 percent load is a plan with no shock absorber. (d) is knowledge concentration, and the fix is the understudy rule, a named second for the named specialist, with a shadowing schedule that is real. (e) is the role smuggler’s matrix, three A’s and no A, and the fix is the one-A-per-row discipline, with the escalation named for the single account-holder. (f) is the sourcing decision, automate, reducing demand rather than adding supply, and the fix is the automation rule: automate the repeatable and never the judgment, with the human review named.

Two. A numbers drill: the capacity you can reproduce. Verify the arithmetic of this chapter, then run the sensitivity. (a) Confirm that twenty-two working days at 80 percent availability is 17.6 person-days, and state the overstatement against the plan’s assumption of twenty. (b) Confirm the audit-week shortfall at Meridian: demand five person-weeks, floor four plus hypercare one, against capacity three, two full-time releasables in audit plus one half-share lead. (c) Sensitivity: suppose the audit window runs two weeks, quarter-end, and the demand is unchanged. Recompute the month-eighteen picture and say what turns red. (d) Confirm the cost of delaying clinic two by six weeks: the grant pays 250,000 units per clinic-month of eligibility, and six weeks is about one and a half clinic-months.

(a) 22 times 0.80 is 17.6 person-days, against the plan’s 20, a shortfall of 2.4 person-days, about 12 percent of the assumed capacity. (b) The audit week: five person-weeks of demand against three of capacity, two full-time releasables bound to the home clinics’ claims review while the two capped leads together contribute one person-week, so the shortfall is two person-weeks in the single week the go-live cannot absorb one; the other three weeks run at exactly five against five, zero slack, so one illness tips them over. (c) With a two-week audit window the shortfall is four person-weeks, and the month-eighteen total becomes 17 against an effective capacity of 16, 20 minus the four lost to audit, so the monthly heat map itself turns red, which is the chapter’s point that the period must be chosen for the constraint. (d) Six weeks is about 1.5 clinic-months, and 1.5 times 250,000 is 375,000 units of lost eligibility, before the compounding cost that the grant pays on evidence; the 375,000 exceeds the combined cost of the borrow and the automation, which is the arithmetic that refuses the rephase.

Three. A field drill: build the picture for your own project. Take the work you lead or know best. (a) Write the inventory: every category of resource the work depends on, people, equipment, facilities, materials, data, environments, and external capacity, with the three facts for each, what it can do, when it is available, who owns its schedule. (b) Build the demand-versus-capacity heat map for the bottleneck role, sliced to the period where the constraint lives, a week or a fortnight, not a month, and mark the cells at or above 90 percent. (c) Walk your responsibility map for the three failures, the row with no accountable owner, the row with three, the column that is overloaded, and write one line for each named role: what they can decide, what they must escalate, who their escalation is. (d) Answer the sourcing question: for each gap the heat map reveals, which of the five moves, build, borrow, buy, partner, or automate, fits the window, and what is the time to capability for each.

The drill succeeds when the heat map’s red cells appear only at the period of the constraint, when the responsibility map can be read by a new person in five minutes, and when every gap has a move and a time to capability attached. The most common failure is the inventory that lists only people, the headcount accountant in drill form, and the repair is to ask of each deliverable “what must physically exist, and whose calendar owns it,” which is how the test environment and the acceptance inspectors appear. The second failure is the heat map at monthly granularity for a weekly constraint, which reproduces the month-17 green plan, and the repair is to choose the period by asking “what is the shortest interval in which this role’s absence stops the chain.” The third is the understudy column empty, the key-person gambler in embryo, and the repair is the rule: every named specialist carries a named second, reviewed at the same cadence as the schedule.

Four. A decision room: close the month-eighteen gap. It is month seventeen at Meridian. Clinic two’s go-live is month eighteen, and the resource review has established the facts: the go-live floor needs four super-users full time for the month, the audit week in month eighteen is short by two person-weeks against the effective capacity, the trainers have slack, sixteen person-weeks against a peak of eight, and the grant pays on access outcomes, 250,000 units per clinic-month, so the calendar is the funding. The options on the table: automate the floor, e-learning and on-screen guidance built by the trainers in their slack weeks, reducing the floor need to three from week three; borrow one super-user from the sister network for the audit week, at the cost of a capability discount, half a person for the first two weeks; swap home-site coverage to release Amina and Tunde fully for the go-live window, a three-week rebalancing of the clinic rosters that the clinic managers resist; buy two super-users, four to six months to capability at about one million units for the hiring; build two clinicians from clinic three as super-users, ten to twelve weeks to first-line capability; or rephase clinic two to month nineteen. Decide what Dana and Nora should do, what they should refuse, and what the review should record.

The defensible answer combines automate, borrow, and build, in that order. Automate first, because it is the only move that reduces demand rather than adding supply, and the trainers have the slack to build it in month seventeen, taking the audit-week demand from five to four against capacity three, short one. Borrow second, one super-user from the sister network for the audit week, closing the one, with the capability discount priced, half a person for the first two weeks, which the floor plan absorbs because the borrowed person is paired with Amina. Build third, the two clinicians from clinic three started now, ten to twelve weeks, ready for clinic twenty-one, the move that makes the next go-live the last one that runs this tight. The refusal is the buy: four to six months to capability against a go-live two months away is arithmetic, not strategy, and the one million units buys nothing for the window it is meant to close; the buy belongs to clinic four’s planning, month twenty-four, where the need outlasts the onboarding. The rephase is refused on its price: 375,000 units of lost eligibility against a borrow and an automation that together cost a fraction of that, before the compounding cost that the grant pays on evidence. The swap is the politically expensive option and the evidence that would change the answer: if the sister network cannot lend, or the automation build slips past the trainers’ slack, the swap becomes the least-bad closing move, and the review should record it as the conditional fallback with its trigger. The unsafe choice is the quiet full-load plan, the go-live run at zero slack with no automate, no borrow, and a hope, because that is the availability optimist’s plan with the audit week already in it. The review should record the decision, the named owners, the borrow agreement’s terms, and the audit-week check as a milestone with a date, so the month-eighteen picture is walked again in the week before the go-live.

Five. The mastery drill: the notice that changes the plan. It is month nineteen at Meridian. Amina, the lead clinical super-user, the only one who has run a full go-live floor, the one who built the floor model this chapter has been using, gives six weeks’ notice. She is leaving for the sister network, the one that lent the audit-week super-user, and she knows the platform, the floor model, and every clinic’s roster. Clinic three goes live in month twenty-one, nine weeks away, and clinic two is in the taper weeks of its go-live, needing two super-users for another fortnight. The facts in the room: two clinicians from clinic three started the super-user build in month seventeen and are in their supervised weeks, five weeks from first-line capability; the automation from the month-seventeen decision is live and the floor runs at three from week three; the trainers have slack; the sister network holds Amina’s notice period and knows the platform; and the grant pays on access outcomes, so clinic three’s date is also the funding. Replan the program: what happens in the six weeks of notice, who covers clinic three’s go-live, what changes for clinics four, five, and six, what you refuse, and what evidence would change the plan.

The replan has three moves and one refusal. First, the six-week transition is planned like a go-live: weeks one through four, Amina runs the floor with the two build clinicians shadowing her on the clinic-three preparation, the rehearsals, the roster walk-throughs, the floor model documented as she runs it, her notes becoming the clinic-three go-live pack; weeks five and six, she hands over, the build clinicians run the rehearsals with her observing, and the document is complete before she leaves. The documentation is the deliverable of the notice period, because the capability transfer is the documentation plus the supervised practice, and six weeks is enough for both if the plan is made the day the notice lands. Second, clinic three’s go-live, nine weeks out: the two build clinicians reach first-line capability in five weeks, the automation holds the floor at three, and the borrow from the sister network, which now has Amina, is the sharpest instrument, because the borrowed super-user for the first two weeks of clinic three’s floor knows the platform and the workflow standard, the capability discount that normally applies is smaller, and the irony is the point: the program that refused to buy her replacement now borrows her replacement from the network that hired her. Third, the program change: the knowledge concentration this chapter named becomes the program’s permanent lesson, the understudy rule becomes real, every named specialist carries a named second with a date, the build pipeline becomes standing capacity, two clinicians trained per go-live cycle, not per crisis, and the capability matrix is walked for its red cells at the monthly review, which is the control that would have caught this earlier. The refusal is the panic hire, the replacement super-user bought for clinic three, because the window is nine weeks and the onboarding is four to six months, and the refusal is the rephase, clinic three to month twenty-two, because the cost of delay is 375,000 units of eligibility and the borrow plus the build plus the automation close the gap for a fraction of that. The evidence that would change the plan: whether Amina’s six weeks are real, whether the build clinicians pass their supervised checkpoint at week five, whether the sister network can lend for clinic three after taking Amina, and whether the automation’s floor-of-three holds under the clinic-three workload; each is an assumption with a test date and an owner, written into the clinic-three go-live pack. A defensible alternative is the rephase with the borrow, if the build clinicians fail their checkpoint, and credit belongs to any replan that schedules the six weeks, names the coverage, prices the refusal, and turns the notice into the program’s first real understudy.

Six. The transfer question. What does your resource plan look like when it is re-sliced to the week, the location, and the shift, and which red cell appears that the monthly totals hid? Which capability in your project exists in exactly one person, and who is their named second, and when did you last check the second’s readiness? Which of your go-live moments runs at zero slack, and what is your shock absorber, the borrow, the automation, the build, the slack itself, and is it named? If the answer to any of those questions is “nobody” or “nothing,” you have just found the chapter’s minimum viable instrument for your project: the capability matrix with its red cells, the net-capacity line under the plan, the responsibility map with one line per role, and the understudy with a date.

The durable principle: a resource plan is a set of explicit choices about capability, availability, focus, and ownership, made at the level where the constraint lives, and the plan that adds up at the monthly level and the headcount level is a plan that has not yet been tested against the calendar, the roster, and the floor. Headcount is not capability, availability is not the calendar, a role is accountabilities plus decision rights plus interfaces, and every named specialist carries a named second. The most common next failure is quieter than the four characters named here: the plan gets built, the heat map gets sliced, the understudy gets named, and then the cadence decays, the monthly review becomes quarterly, the matrix stops being walked for its red cells, the understudy’s checkpoint slips, and the project returns to the headcount table, while the real capacity story runs in the calendars of the people who are too busy to write it down. The capacity must be carried into execution itself, the weekly look-ahead, the audit-week check, the priority meeting, the understudy review, because the resource plan is not finished when it is built; it is finished when the project’s capacity decisions are made on it, every week, in the light. And the resource plan has one more dimension that this chapter has only pointed at: when the gap is closed by buying and partnering, the capability moves off the project’s payroll and onto a contract’s terms, the incentives, the interfaces, and the risk allocation of the external relationship, and the craft of designing that contract, the fifth sourcing move’s full weight, is the next chapter’s subject.

Notes

• The composite cases remain author-created illustrative material. The Meridian month-seventeen resource review, the heat map, the audit week, the go-live floor model, and all named characters are the author’s teaching constructions consistent with the facts established in earlier chapters: the six clinics, the grant of 250,000 units per clinic-month, the month-thirty-six window, the four trainers, the ten clinical super-users of which six are releasable, the two trainers and six-week go-live support per clinic, and the platform team’s one-clinic-per-six-weeks onboarding constraint from chapter 5; the clinic sequencing one per quarter from month twelve, the blank operational-ownership cell and its first-gate consequence, the workflow standard, and Nora’s acceptance authority with Hana and the evidence from Dana and Marcus from chapter 8; and the grant’s evidence-based, in-arrears rhythm from chapter 18. The month-eighteen gap figures are teaching numbers and are reproducible from the text: the floor needs four super-users full time for the go-live month, with the automate decision reducing that to three from week three; the audit week’s effective capacity is three, two full-time releasables bound to audit plus one half-share lead, against demand of five, short two person-weeks; the other three go-live weeks run at exactly five against five, zero slack. The heat map’s monthly totals are internally consistent: trainers peak at 8 against 16, super-users at 17 against 20, platform configuration at 9 against 12, and privacy review at 2 against 3, all under capacity, which is the point. The cost of delay arithmetic is 1.5 clinic-months times 250,000 units, about 375,000 units for six weeks, consistent with chapter 5’s clinic-month valuation. The BlueLine phased-opening and eastern-package references carry forward the month-nineteen schedule decision and the month-twenty cost picture from chapters 17 and 18, including the gangs and specialists the corridor has not confirmed. The KijaniPay QA double-booking and the settlement pilot are from chapters 12, 16, and 17.
• The net-capacity arithmetic, the demand-versus-capacity heat map, resource smoothing and leveling, and the capability-matrix and responsibility-map practices follow the general project-management treatment in ISO 21502:2020, Project, programme and portfolio management, Guidance on project management, which covers resource management including the reconciliation of resource availability with the schedule, and the related vocabulary of the PMBOK Guide, Eighth Edition (Project Management Institute, November 2025), which treats resource management as part of its performance domains; this book describes the ideas in its own words and remains independent of PMI and the standards bodies. The responsibility assignment matrix in its RACI form, the Responsible, Accountable, Consulted, Informed matrix, is a widely used general practice for recording roles per deliverable or decision; this chapter presents the one-accountable-owner-per-row discipline and the matrix’s limits, the inability to show workload, authority, escalation, or flow, as the author’s synthesis of general practice and the governance model of chapter 8.
• The interruption and task-switching evidence follows Gloria Mark, Daniela Gudith, and Ulrich Klocke, “The Cost of Interrupted Work: More Speed and Stress,” Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (ACM, 2008), which reported that interrupted participants took an average of about 23 minutes to return to a primary task after an interruption and reported more stress, frustration, and time pressure; the study was conducted with knowledge workers at a software company, and the chapter applies its finding to project capacity as the author’s extension. The nonlinear growth of waiting time as utilization approaches capacity follows the queueing and production-science treatment in Wallace J. Hopp and Mark L. Spearman, Factory Physics, third edition (Waveland Press, 2011), the same source cited in chapter 17, and the utilization trap follows chapter 17’s flow treatment. The onboarding ramp, nothing in the first four weeks and about half capacity in the next four, is a simplified author-created illustration of the general learning-curve pattern in staffing practice; the man-month argument follows Frederick P. Brooks Jr., The Mythical Man-Month (Addison-Wesley, 1975; anniversary edition 1995), cited in chapter 17, applied here to contractors and the buy decision.
• The build, borrow, buy, partner, or automate framing is the author’s own construction of the general sourcing decision in resource planning, consistent with the make-or-buy treatment previewed for chapter 20; it is presented as a decision aid with time-to-capability, cost, quality risk, and context per move, not as a proprietary framework. The knowledge-concentration and understudy practices, sometimes described in practitioner literature as the bus factor or key-person risk, are described here as general practices without reproducing any commercial manual; the term “bus factor” appears in practitioner usage and is described in the author’s own words. The five failure characters, the headcount accountant, the availability optimist, the role smuggler, and the key-person gambler, are the author’s own constructions, consistent with the failure-aware teaching style established in chapters 17 and 18. No proprietary certification manual, commercial text, or framework guide is reproduced or paraphrased here.