Chapter 5: Select the Right Work

Twelve asks, one training team

Three months after the board approved the clinic program in the language of chapter 1, Elena Marchetti does something unusual: she asks every director to submit everything their department wants funded in the next two years, with no filtering. The steering committee now faces twelve one-page asks: the six clinics, the shared patient-record platform, a telehealth pilot, a lab modernization, a patient-facing app, an urgent-care consolidation study, a state-mandated privacy and security upgrade, a community health-worker program, a call-center refresh, a school-based clinic pilot, and two research proposals. The finance director has totaled the asks: 2.3 times the budget available in the grant window.

Dana Okafor has been counting something else. “We have four trainers, and each clinic go-live needs two of them for six weeks. We can release six of our ten clinical super-users for go-live support without pulling clinicians out of existing sites. The platform team can onboard one clinic every six weeks, and not two.” She pauses. “I do not know how we fund all of this. I do know how we cannot deliver all of it. The question is not what we want to build. It is which outcomes we can actually buy with the capacity we have.”

Elena nods. “Then bring us a recommendation that says no to some of these. Not a wish list. A recommendation.”

That request is the subject of this chapter.

Selection is strategy with a budget

Selection is the decision about which work gets funded, staffed, and started, and it happens before authorization: it is the first act of project leadership, not a portfolio office ritual. Every authorized project was once an initiative competing with others for money and people. The projects that should have been funded lost to the ones that were, sometimes on bad arithmetic, more often because the decision was never deliberately made.

The discipline begins with strategy, not with lists. Every organization that funds work has strategic themes: what it exists to do and the directions it has chosen to move. Meridian’s are equitable access to primary care, quality and continuity through shared records, financial sustainability, and restored community trust. An initiative earns a seat at the table only if it can say, in one sentence, which theme it serves and what it measurably contributes: not what it will build, but what will change. “Patients complete check-in, book follow-ups, and see results without a phone call, cutting reception time per visit by 15 percent” is a contribution; “we will build a patient portal” is an activity. An initiative that cannot produce that sentence is an idea, and ideas are cheap.

The funnel below is the essential picture: ideas enter at the top, funded initiatives leave at the bottom, and rejection and deferral are visible outputs, not silent disappearances.

Figure 5.1: The selection funnel. Ideas enter at the top; funded
initiatives leave at the bottom. Rejection and deferral are
explicit outputs, not silent disappearances.

  MANY IDEAS ──────────────────────────────────────────────┐
        │                                                   │
        ▼                                                   │
  one sentence: theme + measurable contribution ──► reject  │ still an idea,
        │                                                   │ not yet an ask
        ▼                                                   │
  opportunity cost + capacity check ─────────────► defer    │ return date
        │                                                   │ + trigger
        ▼                                                   │
  score and argue: fit, contribution, capacity, ──► kill    │ capacity better
  urgency, option value                                     │ spent elsewhere
        │
        ▼
  FUNDED (few): authorize, combine, sequence

The funnel is not a once-a-year ritual. Selection happens every time money or people are committed, and a project leader who inherits an authorized project still exercises selection judgment by asking one question at every review: if we knew then what we know now, would we fund this again? That question is a selection decision wearing the costume of a status update.

Five kinds of work

Not all work should be judged by the same question, and the fastest way to corrupt a selection decision is to apply growth economics to mandatory work or compliance discipline to experiments. Five kinds cover most of what organizations fund, and each has its own test.

Mandatory work is imposed from outside: a legal obligation, a license condition, a safety requirement, an expiring certification. It is not selected; it is sequenced. The question is never whether, only how to do it at the lowest cost and least disruption, and what other work it can be combined with. Sustaining work keeps value you already have from eroding: replacing a lab analyzer the supplier no longer supports, renewing an expiring platform license, refurbishing a clinic roof. Its test is the cost of failure: what breaks, for whom, and how soon, if the work does not happen. It is easy to underfund, because nothing dramatic happens when it is delayed, until the analyzer fails at the worst possible moment.

Growth work produces new value in the current operating model: a new clinic, a new market, a new product line. Its test is contribution relative to alternatives, where opportunity cost bites hardest, because growth work almost always competes against other growth work. Transformation work changes how the organization operates: the shared patient-record platform, a new operating model, a migration. Its test is the hardest to score, because transformation pays off through other work: the platform earns nothing by itself, yet every grant incentive and every future digital service runs through it. Underfunding the enabler is the classic portfolio error, because it has no dashboard of its own.

Experimental work buys information and options: the telehealth pilot, the AI support trial, the feasibility study. Its test is not return on investment but learning per unit of spend, and every experiment needs an explicit kill criterion before it starts, so that a pilot that fails costs its budget and nothing more.

The labels are judgment calls, not objective facts. The telehealth pilot is experimental if it tests whether telehealth should exist, but growth if the network has already decided to scale it. The label matters, because the label silently chooses the test.

The cost of every choice

The word that makes selection honest is opportunity cost: the value of the best alternative foregone, the thing selection meetings so often refuse to name. When the network funds the urgent-care consolidation study, it is not just spending 400,000 units; it is choosing not to spend them on telehealth, or on a second trainer who would let two clinics go live in the same month. Budget is the visible currency. Capacity is the binding one: the trainers, the super-users, the platform team, the specialists, and the organization’s adoption bandwidth, which can absorb only so much change in a year.

The minimum viable practice is an opportunity-cost statement, one sentence written at authorization and re-read at every gate: “Funding this initiative means [X] will not start, will slip, or will not be staffed, because [the shared constraint].” If the sentence cannot be written, the selection was not a decision; it was an approval.

Cost of delay is the companion concept: the value lost per unit of time a start or delivery is postponed, and it converts urgency from a feeling into a number. The idea was popularized by Don Reinertsen in his 2009 book on product-development flow. It is simple: an initiative that earns 400,000 units per month of earlier delivery should generally beat one earning 100,000, unless the slower one carries a deadline with a penalty. Urgency is a deadline and a penalty, or a window that closes; urgency with neither is enthusiasm, and enthusiasm has a cost of delay of zero.

Meridian’s numbers show the shape. The state grant pays 250,000 units per clinic-month: a clinic earns that for every month it is live on the platform and meeting the access targets, until the window closes at month 36. Delay one clinic by a month and you lose 250,000. Delay all six, and the number is 1.5 million, and nobody reports it, because the grant line does not appear in any construction schedule.

Dana works the comparison between the two plans the committee keeps discussing. Open all six clinics at month 18, compressed: that means hiring extra super-users, trainers, and a platform specialist, about 1.5 million. If it holds: six clinics times eighteen eligible months, 108 clinic-months at 250,000, is 27 million, minus 1.5 million of staff, 25.5 million net. Sequence one clinic per quarter from month 12: eligibility runs 24, 21, 18, 15, 12, and 9 months, 99 clinic-months, 24.75 million, no extra hiring. The compressed plan is ahead by 750,000 on a point estimate, and that is the moment the discussion should change character, because the compressed plan has zero slack. If a construction delay, a data-migration snag, or a super-user pulled back to her home site slips the batch to month 22, the numbers change: 84 clinic-months, 21 million gross, 19.5 million net, and the sequenced plan is ahead by 5.25 million. The decision is not a scheduling preference. It is a bet: 750,000 upside if a fragile plan holds perfectly, against 5.25 million downside if any of six dependencies slips. The honest question is what chance the compressed plan really has, and it belongs to the people with evidence on construction, platform, and staffing, not to the finance director alone. These figures are illustrative; the structure is the lesson: cost of delay turns “which plan” into “what are we betting, and who answers for the assumption.”

What makes a scorecard honest

A selection scorecard structures the argument once the one-pagers and the cost-of-delay statements are on the table. Five criteria do most of the work: strategic fit, outcome contribution, capacity demand, urgency, and option value. Score each from 1 to 5, weight the criteria, and the arithmetic produces a number that orders the discussion. The weights are the strategy; they are the first thing to argue about, and changing a weight should change the ranking, or the scorecard is decoration.

Option value is the least intuitive criterion, so it deserves a sentence of its own. It is the value of keeping a future choice open. Funding an experiment buys the right, not the obligation, to act later, at a cost small relative to the decision it keeps available. The telehealth pilot’s option value is not its expected revenue; it is the evidence that lets the network decide, two years from now, whether a full telehealth program should exist. The platform’s option value is larger still: once it is live, the network can add digital services it has not yet imagined. Work whose entire value is in the future always loses to work with immediate returns unless option value is scored, which is why it must be.

The scorecard has three failure modes. Selection theater: weights tuned after the list is known so a favored initiative wins, and the meeting scores a decision already made; the tell is the same initiative winning under every plausible set of weights. Scoring presentation instead of value: criteria like “executive support” that measure who asked, not what changes. False precision: a weighted average of 3.62 versus 3.58 presented as a verdict when the inputs are judgment calls. The remedy is the same for all three: agree criteria and weights before the candidates are known, score in ranges, and treat the number as a starting point for the argument, never as the verdict. The scorecard’s real product is not the ranking. It is the public act of naming what the room values, before it spends money.

Balance, contention, and the dependency map

A portfolio, even a small one, needs balance across horizons: work that keeps the lights on, work that grows the current model, and work that transforms it. The failure is funding only the horizon with the clearest dashboards, usually growth, while the mandatory work backs up and the enabler starves. Balance across risk matters too: five experimental bets are five possible cancellations, and five certain low-return projects are a plan to stand still.

The map that makes contention visible is the dependency map: each initiative lists the people it needs, and the map shows who appears on more than one list. That overlap is the true budget.

Figure 5.3: What the clinics depend on. The platform is the
linchpin: clinics earn grant incentives only once they are live
on the platform and meeting the access targets.

  STATE GRANT (access incentives, window closes month 36)
        ▲
        │ paid once outcomes are measured
  CLINIC OUTCOMES (wait times, screening uptake)
        ▲
        │ clinic live on platform and meeting targets
  SIX CLINIC OPENINGS ──── each go-live needs: 6 releasable
        │                 super-users for 6 weeks, 2 trainers,
        ▼                 platform onboarding (one at a time)
  PATIENT-RECORD PLATFORM ◄─── shares the platform team with:
        │
        ├── compliance upgrade (same controls, one audit)
        ├── telehealth pilot (deferred until two clinics stable)
        └── patient-facing app (killed: duplicates the portal)

At Meridian the map collapses twelve initiatives into one binding truth: the platform team, the trainers, and the super-users appear on almost every list. Fund everything and the platform team becomes a four-way bottleneck while three projects wait on it. The map also shows the combinations that look like cuts but are actually wins: the compliance upgrade rides the platform’s data-controls work, one audit instead of two, and the patient-facing app is mostly a duplicate of the portal the platform will already provide.

The biases that walk into the room

Selection decisions are made by people, and the biases are dependable enough to be named. The costliest is escalation of commitment, studied by Barry Staw in a 1976 experiment. People who had already invested in a failing course of action kept investing, more than people who inherited the same situation fresh. In selection terms: an initiative with 2 million already spent gets its next million funded because of the first 2 million, which are gone either way. The sunk cost is not an argument; only the future value is, and it has to stand on its own.

Availability is the second, documented by Amos Tversky and Daniel Kahneman in 1973: people judge likelihood by how easily examples come to mind. The urgent-care complaint that made the evening news last month shapes the urgency score of the urgent-care study this month. The analyzer that quietly fails every quarter scores nothing, because no one thinks of it when the criteria are filled in. The correction is structural: criteria and weights agreed before the candidates are announced, a written basis for each score, and a named devil’s advocate.

Power walks into the room in a third form. A senior sponsor’s initiative is not scored worse for its thin evidence; it is scored with confidence borrowed from its advocate. The counter is to make the criteria public before the sponsor speaks, so influence argues through the same numbers as everyone else. Start bias is the fourth: organizations measure starts and celebrate approvals while retirements go unrecorded, so the portfolio grows by addition and the ratio of started to retired climbs year over year. And optimism bias enters through the estimates: marginal initiatives look better when costs are understated, so selection inherits every estimation sin of chapter 15 before that chapter is even reached. The defenses are familiar: ranges instead of point estimates, a pre-committed kill criterion written at funding, and a decision journal recording what the committee expected, so later reviews can test the record against the event.

None of this is about being immune to bias, which no one is. It is about designing the meeting so that bias has to work harder than the evidence.

Kill, defer, combine, sequence, authorize

The output of selection is a decision set with five members, and only one is positive. Authorize means the work proceeds with a named budget and, critically, a named capacity commitment, not just a number. The other four are the discipline that makes authorize honest.

Kill is the decision that value will not appear or that the capacity is better spent elsewhere. Killing is not a judgment failure; it is the exercise of judgment, and the organization that cannot kill learns to fund its past instead of its future. Defer is the same work at the wrong time; a deferral is only real if it carries a return date and a trigger, and a deferral without a trigger is a polite kill. Combine folds related work onto shared enablers: the compliance upgrade into the platform build, the research proposal into the consolidation study. Sequence is the answer selection most often needs and the one that feels least like a decision, because it looks like delay. It orders the work to the capacity that actually exists, converting an impossible list into a possible one. It turns the grant arithmetic of this chapter into real money: a plan that opens on achievable dates earns more than a plan that promises the impossible and slips.

Sequencing also carries an ethical load: deferral is never neutral, because it lands on someone. When Meridian sequences clinic openings, the neighborhood with the last clinic waits longest, and the community liaison will explain why, to people who remember the clinic that closed three years ago. Those neighborhoods are owed an honest reason and a visible commitment, not a portfolio memo.

The Thursday recommendation

Dana returns one week later with four artifacts and a recommendation.

The one-pagers reduce all twelve asks to the selection sentence: theme, measurable contribution, capacity demand, and the shared constraint. Eleven can produce a sentence; one research proposal cannot say what would change, and Dana marks it as an idea, not an ask.

The scorecard orders the five that matter.

Figure 5.2: The selection scorecard, Meridian. Weights: strategic fit 30 percent, outcome contribution 25 percent, capacity demand 15 percent (higher scores mean lighter demand), urgency 15 percent, option value 15 percent.

The platform row shows the arithmetic: 0.30 × 5 plus 0.25 × 5 plus 0.15 × 2 plus 0.15 × 4 plus 0.15 × 5 equals 4.40. The room re-argues the weights: one director wants capacity higher, because the platform team is the binding constraint everywhere; another wants urgency higher, because the grant window does not move. The argument is the point: the scorecard makes the room decide what it values, in public.

The opportunity-cost statement is one sentence: funding all six clinics at once means hiring staff the network does not have, or starving the platform team that every other initiative depends on, or both.

The dependency map is the sheet that ends the meeting, and the recommendation writes itself. Authorize the platform first, and the six clinics as one sequenced program: one go-live every three months from month 12, synchronized to the platform onboarding cadence. Combine the compliance upgrade into the platform build, one audit instead of two. Defer the telehealth pilot with a trigger: it returns when two clinics have held 90 percent platform adoption for eight weeks. Kill the patient-facing app as an initiative; its one distinctive feature becomes an experiment inside the platform backlog. The lab modernization, the call-center refresh, and the school-based clinic pilot defer with return triggers. The community health-worker program folds into the clinic program’s staffing plan, and the research proposal joins the urgent-care consolidation study.

One committee member objects that sequencing sounds like giving up on December. Elena answers: “Sequencing is the difference between promising six openings and delivering six. The grant pays on outcomes, and outcomes arrive in the order capacity can carry them.”

What the delivery style changes

Selection is method-neutral; the life cycle changes its cadence. In a predictive setting, selection is a formal gate: business case, stage review, signed authorization, criteria fixed in advance, quarterly portfolio reviews, change control protecting the committed plan; the discipline is that the gate must actually reject, or it is a rubber stamp. In an adaptive setting, selection is continuous: work is funded in options and increments, experiments are authorized deliberately with kill criteria, and portfolio rebalancing on evidence is a normal act, not an admission of error. In a hybrid setting, the most common real-world case, selection is staged: authorize a tranche, review the evidence, authorize the next, sequencing across the seams so the streams share capacity without tearing it. What does not change across styles is the decision set, kill, defer, combine, sequence, authorize, and the sentence every leader must be able to write: what will not get done.

The everything portfolio

The failure pattern that follows selection is not a bad ranking; it is the refusal to rank at all. The everything portfolio looks like this: at every planning cycle, each department’s ask is funded, trimmed by 10 percent so that everyone can claim partial success. Every team reports 90 percent utilization. Every project slips, because none has the capacity it was promised, and each leader privately believes their project would have made it with the specialist everyone agrees is spread across four others. Nothing has been killed in two years; the portfolio has grown by addition only: more starts, fewer finishes.

Competent people produce this every year, and the reasons are human and respectable: saying no costs a relationship, approving feels like progress, and funding everything looks fair. The tell is measurable: the ratio of starts to retirements climbs every year, and the opportunity-cost sentence cannot be written, because nothing was refused. The mirror failure is selection theater, the ranking that never changes under any weights; between them, the two kill more value than any bad ranking, because they make selection look like work while it performs nothing.

A language model can draft the one-pagers, propose scorecard criteria, and stress-test cost-of-delay arithmetic in minutes, and the drafts are worth having. But selection is where the automation caution matters most, because the output looks like judgment. Every number must be traced to a source the room trusts. Every criterion must be argued by the people who live with it, and the kill and defer decisions made by accountable humans. Fluency is not evidence, and in selection it is an expensive substitute, because the decision commits other people’s work.

Practice

1. A quick classification. Label each initiative by kind of work, and name the test that should judge it: (a) the state-mandated privacy-controls upgrade; (b) replacing lab analyzers the supplier no longer supports; (c) the six community clinics; (d) the shared patient-record platform; (e) the telehealth pilot in two neighborhoods.

(a) is mandatory: imposed by obligation, so the question is sequencing and cost, not whether. (b) is sustaining: its test is the cost of failure if the analyzers break. (c) is growth: new value in the current model, judged on contribution against alternatives. (d) is transformation: an enabler whose payoff runs through other work, which is why it gets underfunded and must be protected. (e) is experimental if it tests whether telehealth should exist, growth if the network has already decided to scale it; the label decides the test.

2. The arithmetic of sequencing. The grant pays 250,000 units per clinic-month once a clinic is live on the platform and meeting targets, until the window closes at month 36. Plan A opens all six clinics at month 18 and needs 1.5 million of extra hired staff. Plan B sequences one clinic per quarter from month 12 with no extra hiring. (a) Compute the expected net grant incentive for each plan. (b) If Plan A slips to month 22 because its zero-slack schedule hits one dependency, what does the gap become? (c) What evidence would change your recommendation?

(a) Plan A: 6 × 18 = 108 clinic-months at 250,000 is 27 million gross, minus 1.5 million of staff, 25.5 million net. Plan B: eligibility runs 24, 21, 18, 15, 12, and 9 months, 99 clinic-months, 24.75 million. Plan A is ahead by 750,000 on the point estimate. (b) Plan A at month 22: 6 × 14 = 84 clinic-months, 21 million gross, 19.5 million net, and Plan B is now ahead by 5.25 million. The structure is the lesson: a fragile plan wins a little if it holds perfectly and loses a lot if anything slips, so the decision belongs to the people with evidence that the schedule can hold. (c) Strong evidence that the compressed schedule can hold, cheaper hiring, a longer window, or a smaller incentive would each shift the arithmetic; revisit the decision when evidence on any dependency changes.

3. Your opportunity-cost statement. Choose an initiative you know from work, community, or study. Write the one-sentence opportunity-cost statement for funding it, naming the initiative that will not start, slip, or go unstaffed, and the shared constraint that forces the trade. Then name the single initiative in your own environment that should be killed or deferred, and the evidence that would revive it.

The exercise succeeds when the sentence names a real constraint, a person-visible capacity (a specialist, a team, a season) rather than “the budget” in the abstract, because budget trades are easy to evade and capacity trades are not. A defensible kill or deferral names a return trigger: what would have to be true, with what evidence, for the work to return. The red flag is a statement with no second term: “funding this means nothing else changes” is an unexamined assumption, not an opportunity cost.

4. The mastery drill. Rank these five initiatives for the funding cycle, then defend what should not be funded: (a) a regulatory compliance upgrade with a 9-month deadline and a 2 million penalty, needing the scarce integration specialist for 2 months; (b) a flagship growth launch with the year-one revenue potential, needing the same specialist for 6 months; (c) a churn-reduction program with strong evidence of 300,000 per quarter after month 12; (d) an AI support experiment, small budget, 3 months, uncertain; (e) internal tooling modernization that saves 100,000 per quarter with no deadline.

Ranking without capacity is arithmetic fiction, so the first move is to sequence around the shared specialist: (a) and (b) go first, the specialist on the compliance critical path for 2 months and then on the launch, because the penalty makes (a) a deadline and the launch makes (b) the largest future value; (c) next on its evidence; (d) after, as a bounded option; (e) last. What should not be funded this cycle is (e): its benefit is real but it has no deadline, no penalty, and no window, so its cost of delay is near zero, and its natural window opens once the launch releases the engineers. A defensible alternative defers (d) rather than killing it, with a trigger, because its option value is its point. The unsafe answers are funding all five, which recreates the everything portfolio, and killing (a), which converts a 2 million penalty into a decision. The strongest answer also names the trade openly: funding the launch is what pays for not funding the tooling.

5. Transfer question. What is the ratio of initiatives started to initiatives retired in your organization over the last year? If nothing has been retired, what does that tell you about how selection decisions are made, and what is the smallest change that would make the next decision honest?

Selection is strategy with a budget, and the scarcest resource is almost never money: it is the trainers, the specialists, the teams, and the organization’s capacity to adopt change. Strong delivery begins before authorization, with the courage to write what will not be done, and it continues after authorization, with the question of whether this work would still be funded if it were proposed today. The most common next failure is subtler than a bad ranking: once the right work is selected, the instinct is to lock the solution and start building, which converts a good selection into a premature commitment to a favored answer. That is the subject of the next chapter, “Discover the Real Problem.”

Notes

• Meridian Community Health Network is a composite case created for this book; no real organization, grant, or region is depicted. All figures in this chapter are author-created illustrations for teaching; the arithmetic is provided so the reader can reproduce it, not as a claim about any real grant.
• The opportunity-cost and cost-of-delay vocabulary: opportunity cost is a standard concept in economics; the explicit framing of cost of delay as a decision input in product and project work was popularized by Don Reinertsen in Principles of Product Development Flow (Celeritas Publishing, 2009). The five-kinds taxonomy, the selection scorecard, and the kill, defer, combine, sequence, authorize decision set are the author’s synthesis.
• Escalation of commitment: Barry M. Staw, “Knee-deep in the big muddy: A study of escalating commitment to a chosen course of action,” Organizational Behavior and Human Performance 16, no. 1 (1976): 27-44.
• The availability heuristic: Amos Tversky and Daniel Kahneman, “Availability: A heuristic for judging frequency and probability,” Cognitive Psychology 5, no. 2 (1973): 207-232.
• Portfolio selection sits within the guidance of ISO 21502:2020 (project management) and ISO 21504 (portfolio management); the PMBOK® Guide, Eighth Edition (Project Management Institute, November 2025) and PMI’s portfolio standard address related selection concepts. Readers should consult the official publishers for current editions; this book is independent of all of them.