Return to Harbor
Harbor Service wants technicians to solve customer problems promptly. Yet it rewards completed first visits, requires central approval for above-limit parts purchases, and compares regions without adequately considering the jobs they receive. A technician who waits for approval can fall behind on the completion count. Another who closes a visit quickly can look productive even when the customer needs a return visit.
The operations director proposes giving technicians more authority. The finance manager proposes stricter spending controls. A regional supervisor wants to replace the completion bonus with a quality bonus. Each proposal addresses a possible problem. None yet explains how the rest of the organization would work with the change.
This final chapter treats Harbor as an integrated design problem. The facts remain fictional; the separate calculations in earlier chapters are not observations from a continuing study of the business. Nor should we assume Harbor adopted the policies those chapters considered.
Keep the objective from Chapter 1: reliable service that earns more than its relevant costs over time, within safety requirements and customer commitments. A lower repeat rate is useful only in that wider comparison. Harbor could reduce repeats by refusing difficult work or replacing too many serviceable parts. Neither is necessarily an improvement.
The task is to diagnose the mismatch, compare feasible packages, assign authority over the changes, and decide how to learn whether they helped.
Why the pieces have to fit
Suppose Harbor delegates more parts decisions while leaving the first-visit reward unchanged. Buying a replacement might become the easiest way to close a visit, even when further diagnosis would reveal a cheaper solution. Faster decisions could then come with unnecessary spending.
Now add credible review of diagnosis, parts use, and lasting repair. That review could make expanded local authority more valuable by discouraging waste while preserving the benefit of prompt decisions. Conversely, review has more useful decisions to support when technicians can act on what they know.
Two practices are complements when adopting more of one increases the incremental return to the other. Milgrom and Roberts (1995) develop this idea to analyze organizational fit and change. Our delegation-and-review example illustrates a possible complementarity; it does not establish one from Harbor data.
Fit does not mean maximizing every practice. Reviewing every purchase could destroy the speed that delegation was meant to achieve. Some controls can substitute for others: reliable checks after a repair may reduce the need for prior approval. The question is which combination creates more value after counting its costs.
Nor is a familiar bundle necessarily a good bundle. A completion bonus, rapid job closure, and a habit of hiding returns might reinforce one another while harming the business. Internal consistency is useful only when the resulting behavior serves the objective.
This explains why copying one visible practice from a successful firm can disappoint. A delegation rule may depend on training, records, and credible evaluation that the copy leaves behind. We need to identify the mechanism, not borrow the label.
The manager is also an agent
The operations director evaluates supervisors, but who evaluates the director? A reform that exposes costly returns might make the director’s previous results look worse. The director might favor a policy that enlarges the department, protects a familiar arrangement, or produces a quick improvement in the headline score. Those are possible incentives, not findings about this fictional person.
Governance concerns how important decisions are authorized, constrained, and reviewed, including decisions made by managers. It is broader than choosing a board or an executive bonus. For Harbor, it includes who can approve the service policy, obtain records, challenge classifications, and stop a poorly performing change.
Fama and Jensen (1983) distinguish initiating and implementing decisions from ratifying and monitoring them. The practical point is that the person proposing and carrying out a decision need not also have exclusive authority to approve it and judge its success. Separation can limit conflicts, but requires information and consumes resources.
For the proposals below, suppose Harbor’s owners appoint a service reviewer with access to original job records and a direct reporting route to them. The reviewer examines both technicians’ and managers’ decisions, including disputed classifications. The director cannot remove an inconvenient finding from the report.
That does not make the reviewer perfectly informed or disinterested. Someone rewarded for finding violations may overstate minor faults; someone judged only on review expense may inspect too little. The owners must consider the reviewer’s competence, evidence, costs, and incentives too. A small firm might assign this role part-time, using qualified assistance when needed. Independence of a particular judgment need not require a large new department, but nobody should review their own disputed decision.
Package A: make central approval work better
Package A retains supervisor approval above the existing local spending limit. It puts a trained approver and a designated backup behind that requirement, supplies concise diagnostic information, and establishes a route for urgent requests. Dispatchers continue coordinating appointments and shared stock.
Approval speed becomes part of the supervisor’s evaluation, alongside the quality and cost of approved decisions. A supervisor cannot appear successful simply by preventing technicians from spending money while customers wait for necessary repairs. At the same time, an approval desk rewarded only for speed could approve weak requests without scrutiny. The evaluation must consider both sides.
Technician evaluation changes too. Harbor considers lasting repairs, relevant resource use, and job difficulty rather than paying mainly for closed first visits. Documented approval delays help interpret the technician’s results. The delay’s cost still belongs in Harbor’s accounts even if it should not be blamed on the technician.
Rewards follow this broader evaluation, with an opportunity to correct factual errors. Harbor does not mechanically penalize every return visit: some follow-ups are necessary, and some faults are difficult to anticipate. Neither does it excuse every return simply because an employee supplies an explanation.
This package is attractive when errors are costly to reverse, specialized judgment is concentrated centrally, and the approval desk can respond while the decision still matters. It pays for that protection with staffing, information transfer, and possible delay. Central reviewers may remain too cautious or miss details that the technician cannot easily convey. Faster communication helps only if someone has the time and knowledge to use it.
Package B: make bounded local authority workable
Package B lets trained technicians approve a defined category of routine repairs within a total-parts limit per repair. They must document the diagnosis, resolve technical and safety concerns, respect reserved stock and other appointments, and remain within the customer’s authorization. The dispatcher retains authority over shared scheduling and stock commitments. Being near the equipment does not give a technician authority over everyone else’s resources.
The category and monetary limit must be specified before this arrangement operates. Chapter 5’s worked rule shows how to draw such a boundary; its illustrative $150 limit is not an established optimum or an automatically adopted Harbor policy.
Cases outside the technician’s authority go to the supervisor. Silence is not approval. Package B removes selected prior approvals, not every approval or every obligation to coordinate.
Evaluation and rewards change along with authority. Technicians cannot keep the old completion incentive while acquiring an open-ended parts budget. Harbor uses evidence on lasting repairs, resource use, job conditions, and cooperation. This does not require turning every item into a bonus formula; some information may support informed review rather than a separate numerical target.
The service reviewer examines flagged decisions and samples ordinary ones. Looking only at reported failures would miss questionable purchases that happened to work and returns disguised as unrelated jobs. Review also covers supervisor-approved cases. A technician cannot avoid scrutiny merely by obtaining a signature.
Package B is attractive when local knowledge is valuable, waiting is costly, and eligible errors can be detected or limited without recreating approval on every job. Its costs include training, recordkeeping, sampled review, and remaining waste. Employees may split purchases or misclassify unusual repairs as routine. Supervisors may undermine delegation by demanding permission unofficially. The proposed boundary must work in practice as well as on paper.
| Design choice | Current arrangement | A: supported central approval | B: bounded local authority |
|---|---|---|---|
| Repair authority | Above-limit parts require central approval | Retain approval with a staffed escalation route | Delegate eligible repairs; escalate outside the boundary |
| Evaluation | Initial completions; weak job-mix adjustment | Lasting repair, costs, conditions, and approval delay | Lasting repair, costs, conditions, and use of local authority |
| Rewards | Mainly completed first visits | Revised technician and supervisor evaluation | Revised technician and supervisor evaluation |
| Coordination | Approval and dispatch handoffs | Better diagnostic records and central response | Local decisions with dispatch and stock constraints |
| Review cost | Existing approvals, delays, and unresolved errors | Approval staffing, information transfer, and review | Training, records, sampled review, and remaining errors |
On a small screen, scroll the table sideways to see all columns.
Neither proposal is a costless ideal. A poorly staffed approval desk is not a fair test of Package A; uncontrolled local purchasing is not Package B. Comparison requires enough detail to know what each arrangement would actually demand.
The boundary of the firm remains a separate choice. Either package could buy specialist diagnostic advice when maintaining that expertise internally is too expensive. Purchasing advice does not automatically transfer final repair authority to the adviser. As Chapter 11 emphasized, Harbor must assign responsibilities at the handoff and count the cost of managing the outside relationship.
Who may change the rule—or make an exception?
In both packages, suppose the operations director may propose a permanent change in the permitted repair category or spending limit. The owners must ratify it, with a recorded reason and prospective effective date. The director administers the policy but cannot silently rewrite it when a difficult case arises.
An ordinary escalation is not necessarily an exception. If policy assigns expensive purchases to a supervisor, that supervisor’s approval applies the rule. It does not waive the technician’s limit or raise it for future cases.
An exception permits a departure on stated grounds without rewriting the general policy. Suppose the policy allows the director or designated backup to authorize a technically verified substitute when the approved catalogue part is unavailable. The authorization must remain within the overall approved budget and satisfy customer and dispatch requirements. The technician records the reason, approving person, and outcome; the service reviewer checks these decisions too.
No role receives an informal waiver of safety or customer authorization. Nor does the owners’ power to amend policy give them an unrecorded current-case override. These are proposed internal commitments, not claims that an organization can write away obligations imposed from outside it.
Binding limits help employees resist pressure from an insistent customer or a powerful manager. But rigid rules can also delay a sensible response. Harbor therefore needs both a usable permitted-exception route and a timely amendment process. If the same exception recurs, the director should investigate changing the category. Repetition is evidence for review, not permission to ignore the rule.
This is the rules-versus-discretion distinction from Chapter 5. Moving authority upward does not settle how much discretion anyone should have. A sensible governance proposal specifies who establishes the facts, applies the rule, authorizes a permitted departure, changes future policy, and reviews those actions.
Informal understandings still matter
A written policy may protect technicians who report an honest mistake. If supervisors then give those technicians the least desirable assignments, employees learn a different practical rule. Conversely, an experienced technician who helps colleagues diagnose difficult repairs can support performance even when that assistance is hard to reward precisely.
Informal norms work through expectations about other people’s responses. A norm of reporting trouble early can improve information; a norm of protecting colleagues from scrutiny can conceal it. “Improve the culture” is not a sufficient recommendation until we specify which behavior should change and why people would expect the new response.
Baker, Gibbons, and Murphy (2002) analyze informal agreements supported by the value of continuing relationships. The future matters because taking an immediate advantage can damage later cooperation. It does not guarantee that a promise will be kept.
For Harbor, management’s handling of the first difficult case will affect the credibility of its announced rules. A fair review of a reasonable but unsuccessful decision can support genuine delegation. Quietly restoring a first-visit ranking when the new measures become inconvenient can undermine it. Formal controls and informal expectations therefore need to be examined together.
Changing the arrangement is itself a costly decision
Even a better eventual system can be a poor investment if transition costs exceed its likely gains. Harbor may need training, new record fields, reviewer time, and temporary coverage while technicians learn the procedure. Some early delays are learning costs; others may reveal a persistent defect. Calling every problem “temporary” prevents useful reconsideration.
Prepare the supporting parts before activating the package. Define eligible repairs, test whether the necessary records can be collected, train the people who approve or review decisions, and announce how evaluation will change. Delegating first and promising to devise oversight later exposes the firm to a different arrangement from the one being recommended.
A bounded pilot can test linked changes together in one service category or set of teams. Small scope does not require changing only one rule. If authority and evaluation need one another, testing authority alone may teach little about the proposed package. This is the practical difficulty of coordinated change highlighted by Milgrom and Roberts, not an argument for redesigning the entire firm overnight.
Resistance deserves examination. A supervisor may fear losing status, but may also know that the proposed records miss an important fault. Technicians may dislike additional effort, but may reasonably worry that management will use trial results to withdraw promised rewards after the work is done. Ask what information or commitment would distinguish these concerns; disagreement does not identify its motive.
Harbor should honor existing commitments and state prospective trial terms, including what may change after review. It need not promise that a flawed procedure will last forever. It should distinguish revising future terms through the announced process from denying rewards already earned under agreed terms. A cheap short-run saving can make future cooperation harder to obtain.
What would count as evidence of improvement?
Before starting, specify the eligible jobs, comparison period, and follow-up horizon. Count repairs requiring another visit relative to comparable initial jobs, allowing the same time for failures to appear. A newer group of repairs has had less opportunity to generate returns.
Track service and resources, not just the favored score. Relevant information includes lasting repair, customer waiting, parts and labor use, displaced appointments, and approval or review time. Keep safety and customer requirements as constraints. Do not erase a cost because its cause is disputed, double-count return labor already included elsewhere, or add a customer’s entire loss to Harbor’s expenses. Customer consequences still matter through commitments, willingness to buy, and other relevant effects.
Watch the ways the score can improve without the service improving. Are difficult jobs being refused or shifted to other teams? Are return visits relabeled? Are technicians replacing more parts to avoid reported failures? Has review become so burdensome that appointments take longer? Check a sample of underlying records rather than relying entirely on the new summary.
For comparison, Harbor might keep similar teams on the existing arrangement temporarily or introduce the package at different times. Compare changes, not just final levels, and examine job mix, parts availability, and seasonal demand. A before-and-after improvement alone does not establish that the package caused it.
Random assignment can strengthen a comparison when practical, but is not a condition for every managerial decision. A small pilot may provide limited evidence; shared staff can carry new practices into comparison teams, and selecting the most enthusiastic team can make expansion look easier than it will be. State these limits rather than dressing a rough comparison as certainty.
Testing a package also does not identify the separate contribution of every component. If results improve, Harbor may still need to learn which records or reviews are worth retaining. Removing a component without considering its links could, however, undo the improvement.
Agree in advance on review dates, resource limits, and conditions for pausing unsafe or unmanageable work. The director manages the trial; the reviewer reports findings directly to the owners; the owners decide whether to expand, revise, or stop it. A favorable headline result is not enough to bypass that review.
Change one assumption before defending the recommendation
Suppose reliable diagnostic software makes some local information cheaper to communicate. Headquarters can now see details that previously required a lengthy explanation. This is an assumed change in information costs, not a claim about what a current AI product can do.
The change could strengthen Package A by making central approval better informed and faster. It could also strengthen Package B by making sampled review more useful. Which gain is larger depends on the remaining approval bottleneck, the importance of facts the software misses, and the cost of errors.
Neither conclusion removes the need to evaluate rewards or assign authority. A better diagnostic report does not decide who may waive a rule, resolve a stock conflict, or review the director. Technology changes some costs in the comparison; it does not supply the whole organization.
For a conditional recommendation, suppose review suggests that routine repairs are well understood locally, approval delays are a major source of avoidable returns, and sampled checking can detect waste at reasonable cost. A bounded Package B pilot would then be a sensible next step. If errors instead require scarce central expertise and can cause serious irreversible damage, Package A becomes more attractive. These are reasons to choose and reconsider, not proof that either package is best for all of Harbor.
A public organizational audit
The following sequence is a reusable way to organize a recommendation. It is not a checklist that produces the answer automatically.
- State the decision and objective. Name the activity, whose interests the arrangement should serve, the relevant horizon, and binding constraints. Distinguish a useful outcome from a convenient count.
- Describe the current arrangement. Identify who knows what, who may decide, how results are interpreted, and what consequences follow. Include handoffs across teams or outside the firm.
- Identify competing mechanisms. Explain how particular choices could produce the problem. Separate conflicting incentives from missing skills, unavailable inputs, and changes in the work. Name information that would change the diagnosis.
- Compare feasible packages. Show which changes support one another. Count production, coordination, monitoring, transition, and remaining error costs without assuming away uncertainty. Explain why a narrower change would or would not suffice.
- Assign governance and implementation. Identify the proposer, approver, operator, and reviewer. Separate ordinary authority, permitted exceptions, and policy amendment. State how people will learn the new arrangement and which commitments it must respect.
- Recommend, test, and reconsider. Defend a bounded choice, name the assumption most likely to reverse it, and specify an affordable review. Say what the available evidence cannot establish.
The useful result is an argument another person can inspect: this problem appears to arise through these choices; this package changes them in these ways; it introduces these costs; and these findings would make us revise it. That is more demanding than choosing “centralization” or “stronger incentives,” but it gives a manager something to act on.
Study map
Organizational fit links authority, evaluation, rewards, coordination, and boundaries. Complementary practices can make coordinated change worthwhile, while some controls substitute for others. Governance applies the agency problem to managers and monitors as well as employees. Rules need owners for application, exceptions, and amendment. Implementation consumes resources and depends on credible commitments. Evaluate a change against feasible alternatives, watch behavior outside the favored measure, and distinguish an encouraging outcome from evidence about its cause.
Practice and discussion
These are public practice questions. The Harbor audit is a learning tool, not a solution to a separate graded capstone. Use the stipulated facts and state any additional assumptions.
Spot the incomplete reform. Harbor delegates routine purchases but continues rewarding mainly completed first visits. Explain a useful and a harmful response this could encourage. Which companion change would you investigate, and what would it cost?
Compare packages. Choose A or B for a service category you define. Explain the importance of local information, error costs, delay, and review feasibility. Name one condition under which you would prefer the other package. Do not assume one proposal has perfect employees or costless oversight.
Separate three decisions. A technician requests an above-limit purchase; a catalogue part is unavailable but a verified substitute exists; repeated substitution requests suggest the catalogue should change. For each, distinguish ordinary approval, a permitted exception, and policy amendment. Who acts, who reviews, and which constraints still bind?
Evaluate an encouraging report. Repeat visits fall after Harbor introduces Package B, but the pilot team receives fewer difficult repairs and has a shorter follow-up period. What can the report establish? Propose a feasible comparison and one check for unintended behavior. Why would a credible improvement in the package still not identify the effect of every component?
Audit a changing organization. Use the public audit to outline a recommendation for Harbor under the diagnostic-software assumption. Explain what information becomes cheaper to communicate, which decisions might move, which should not move merely for that reason, and who may revise the policy. Include a review cost and evidence that would make you reconsider.
Source note
The linked sources ground the distinctions between decision management and control, organizational complementarity, and informal agreements supported by continuing relationships. Harbor’s packages, role assignments, comparison table, pilot, audit, and exercises are original teaching constructions, not findings or examples taken from those papers. No real policy, implementation result, or current software capability is asserted.