Information and Decision Rights

A part on the truck, an approval somewhere else

A Harbor Service technician is at a customer’s business, examining a machine that has stopped working. The technician believes a replacement part will finish the repair. The correct part is on the truck and its replacement cost is $120. The customer has already authorized the repair under the agreed service terms. But Harbor requires headquarters approval when combined parts spending exceeds $100 for one repair.

The technician sends a request. The supervisor is dealing with another problem. Meanwhile, the dispatcher wants the technician to leave for the next appointment. If the technician leaves before receiving approval, Harbor may need another visit to install a part that was available all along.

Should Harbor raise the limit, require the supervisor to answer faster, or let technicians approve any part they think necessary?

Each proposal addresses part of the problem. The technician sees the equipment directly, but also benefits from closing another visit. The dispatcher sees the queue of waiting customers, but cannot diagnose the machine from the schedule. The supervisor can compare purchases across jobs, but may not understand this fault without help. Nobody possesses every relevant fact or automatically bears every consequence.

The issue is not simply whether Harbor trusts its employees. It is how to make useful knowledge influence decisions while keeping those decisions consistent with the firm’s objectives. That requires specifying who may do what, which information must move, and who checks the result.

Harbor remains a fictional business. This is a new repair illustration, not another observation from Chapter 6’s technician comparison. All proposed rules and amounts are teaching assumptions. Safety requirements, technical suitability, and customer authorization remain constraints under every arrangement.

Useful knowledge is in more than one place

Start by asking what someone must know to make this particular decision. “The technician has the information” is too broad.

The technician has inspected the machine, observed the symptoms, and performed the relevant diagnostic checks. The customer may have described when the problem appears. Some of that information fits a standard record. Other details may require follow-up questions or another observation before anyone can interpret them confidently.

The dispatcher knows the next appointment, which other technicians are available, and whether a spare part has been reserved for another job. Headquarters knows the purchasing terms, previous failures involving that part, and whether several technicians have been replacing the same component unnecessarily. These facts are not less important because they originate away from the customer.

Hayek’s discussion of dispersed knowledge distinguishes knowledge of general principles from knowledge of particular circumstances. The latter can matter even when it never appears in a central report. His argument also recognizes that someone with local knowledge needs information about conditions elsewhere. Knowing one’s own situation is not the same as knowing how to coordinate with everyone else.

For Harbor, a repair manual and a live appointment schedule perform different jobs. The manual can describe which diagnostic test is appropriate. The schedule can reveal what another hour at this location would displace. Neither tells the whole story.

A shared database could make the part’s price and inventory status available to everyone. That would remove some unnecessary messages. It would not, by itself, establish that the technician interpreted the fault correctly. Someone still has to obtain the evidence, enter it accurately, and understand what it implies. A complete-looking form can contain an incomplete diagnosis.

There is also a difference between information that is unavailable and information that exists but is withheld. A technician may struggle to describe a symptom honestly. Alternatively, a technician may omit evidence that a cheaper repair would work. The first problem concerns communication or expertise. The second involves a conflict of interest. Better reporting tools can help with the first without resolving the second.

Private information means one participant knows something another does not. It does not, by itself, imply deception. Before treating an unanswered question as misconduct, ask whether the person could reasonably know the answer and communicate it in time.

Move the information, move the authority, or do some of each

Harbor has two broad ways to bring knowledge and authority together. It can improve the information reaching the supervisor, or it can give the technician more authority to act on what the technician knows. Often it will combine the two.

Jensen and Meckling’s analysis of knowledge and organizational structure makes this comparison explicit. They distinguish knowledge that is inexpensive to transfer from knowledge that is costly to transfer. Effective transfer means the recipient can use it, not merely that a message arrived. Moving authority toward useful knowledge can reduce communication losses, but the assignment must also account for conflicting interests.

Suppose Harbor keeps central approval but introduces a short request containing the machine identifier, diagnostic result, proposed part, price, and effect of waiting. The supervisor now receives a usable proposal rather than “need a part urgently.” A live call might resolve an unusual question. This arrangement is attractive when a brief exchange lets the supervisor make a substantially better decision before the opportunity disappears.

But filling the record, locating the supervisor, and explaining the fault take time. If every routine request requires the same exchange, the supervisor may become a queue through which work must pass. The relevant cost includes waiting and the supervisor’s displaced work, not just the cost of sending a message.

Now suppose Harbor delegates routine parts approval while sending the technician current information about stock reservations and spending limits. Less diagnostic detail must move upward before each repair. Some coordination information moves outward instead. Delegation does not mean ending communication; it changes which information needs to reach whom before action.

The right choice can differ across decisions within the same job. A technician may be well placed to select a routine part. The dispatcher may be better placed to rearrange three customers’ appointments. A specialist may need to approve an unfamiliar modification. Calling the whole department “centralized” or “decentralized” conceals these separate assignments.

The direction of a technology’s effect is therefore not predetermined. Better remote diagnostics might make central approval less costly. Better local access to purchasing and scheduling information might make delegation more useful. We need to identify the information bottleneck the change actually removes, not assume that more software necessarily means more local authority.

Count the cost of waiting, including what happens elsewhere

Make the repair comparison deliberately simple. Suppose the $120 part is necessary and will work equally well now or on a return visit. Using it requires replenishing that inventory for $120. This is its opportunity cost, not a charge for an unrecoverable past payment. The return visit uses $90 of additional travel and labor resources. Initial-visit and installation costs common to both choices cancel. Customer receipts, repair quality, and other relevant consequences are unchanged. Either timing satisfies the case’s safety and service requirements.

Installing now costs $120 in this comparison. Leaving and returning costs $120 + $90 = $210. Acting now saves $90, not $210: Harbor purchases the part either way. Under these assumptions, an approval delay uses resources without improving the repair.

This calculation does not give the technician permission to ignore the existing rule. It gives management a reason to examine that rule. Nor does the calculation establish that every proposed part is necessary. We temporarily assumed a correct diagnosis to isolate the cost of delay.

Now change one assumption. Staying to finish this repair means losing another job that would have contributed $140 after its avoidable costs. That job cannot be rescheduled. Leaving now preserves it, while the later return still uses the same $90 of extra resources. Hold everything else fixed.

Immediate installation now has a relevant cost of $120 + $140 = $260. Returning still costs $210. Staying costs $50 more. The technician’s ability to finish the current repair does not establish that staying is the best use of Harbor’s capacity.

The $140 is lost contribution, not the other customer’s entire bill. Chapter 2’s opportunity-cost logic applies: count what Harbor gives up after allowing for costs it avoids by not doing the other job. Do not count the same lost work again inside the $90 return cost.

The dispatcher now has information that can reverse the recommendation. Harbor might communicate that information to the technician, give dispatch authority over schedule changes, or require joint approval when another appointment is affected. The arithmetic does not select among these procedures. It shows why a parts decision cannot always be separated from a scheduling decision.

Neither comparison includes uncertain diagnostic mistakes, unnecessary purchases, or the resources used by an approval system. Those matter when choosing the policy. A correct calculation for one repair is not a calculation of the best spending limit for the entire business.

Better information does not eliminate agency costs

Suppose Harbor removes the parts limit while continuing to reward technicians mainly for closed initial visits. Buying a replacement can become attractive if it finishes a visit faster than investigating whether an existing component can be repaired. Harbor pays for the part; the technician receives credit for the completion.

This is a possible incentive problem, not a finding that technicians are wasting parts. A replacement can genuinely be cheaper once diagnostic time, reliability, and future visits are counted. The question is whether the technician’s comparison reflects the consequences that matter to Harbor.

Central approval changes the problem rather than making it disappear. Imagine the supervisor is evaluated primarily on this month’s parts spending. Rejecting a useful replacement protects that measure even if it produces a costly return visit. Alternatively, a supervisor pressed to clear a request queue may approve everything without meaningful examination. Authority at headquarters does not guarantee aligned incentives or adequate attention.

Agency costs help shape the resulting structure. Harbor can leave repair proposals with technicians, retain selected approval rights, and assign later review to someone other than the purchaser. It spends resources on those controls and still faces some losses from imperfect decisions. More review can reduce one kind of loss while creating delay, paperwork, or defensive behavior.

Keep the categories straight. A part used in a worthwhile repair is an operating resource, not automatically an agency cost. An approval review intended to limit unnecessary purchases is a control cost. Waste caused by pursuing a completion target can be a remaining agency loss. An honest mistake caused by missing expertise has a different explanation, even if it produces the same repair bill.

These distinctions matter for the remedy. A reporting change, training, a different reward, and a spending restriction address different mechanisms. Harbor should compare workable combinations rather than choose between imaginary extremes of perfect central control and perfectly informed, perfectly aligned local employees.

Separate the steps in a decision

“Who makes the decision?” can have several answers because a decision contains different tasks. Someone identifies an option, someone authorizes it, someone carries it out, and someone evaluates what happened.

Fama and Jensen (1983) distinguish four functions: initiation, developing a proposal; ratification, approving the proposal to proceed; implementation, carrying it out; and monitoring, evaluating decision-makers and applying the resulting rewards or other consequences. They group initiation and implementation as decision management, and ratification and monitoring as decision control.

For Harbor, a technician can propose buying a part without possessing the authority to approve the purchase. Approval does not itself install the part. Recording an installation does not establish that the choice was justified or the repair successful. These are different activities even when one person performs several of them.

Table 5.1 compares three possible assignments. “Unrestricted” refers only to local parts-purchase approval, not permission to override safety requirements, change customer terms, or rearrange the dispatch schedule. All three arrangements retain later review.

On a small screen, scroll the table sideways to see all columns.

Separating functions does not require four employees for every repair. Nor must every routine choice receive independent approval before action. A supervisor can establish a narrow category of local decisions and examine a sample afterward. The technician then proposes, approves, and implements within that category, but does not have exclusive authority to evaluate the resulting conduct.

The reviewer needs enough information to do more than confirm that a receipt exists. A receipt establishes spending; it does not show whether the purchase was worthwhile. Review might examine the recorded diagnostic result, the available alternatives, and subsequent repair history. It should distinguish evidence available at the time from facts learned only afterward. A failed part does not by itself prove a careless decision, just as a successful repair does not prove every expense was necessary.

Should anyone have discretion?

Before asking who should decide, ask whether the response should be chosen case by case. Discretion is room to choose among permitted actions. A policy may instead prescribe the response once specified conditions are established.

Suppose Harbor adopts a separate protocol for one fully specified fault. After the diagnosis and required safety, scheduling, and customer conditions are verified, the protocol requires a particular replacement. Neither technician nor supervisor may substitute another response merely because they prefer it. This differs from a spending limit, which can leave several eligible choices.

The location of authority and the scope of discretion are separate design choices. A central supervisor might exercise broad judgment, while technicians at several locations apply the same fixed protocol. Rule-based action is not a third location between headquarters and the field.

Alonso and Matouschek frame delegation as a choice over the decisions available to an informed agent whose objectives may differ from the principal’s. Delegating authority need not mean permitting every action. Alonso and Matouschek, 2008

Prescribing a response does not eliminate all decision rights. Someone establishes the facts, interprets uncertain conditions, and decides whether the rule needs revision. Authority to apply a rule is not necessarily authority to waive or amend it. If Harbor gives supervisors no exception power for this protocol, moving a request upward does not make an exception permissible.

This is the useful part of a constitutional analogy: rules can bind managers as well as employees. The organization must specify who can change those rules and through what process. Revising a policy for future cases is different from waiving it for a favored current case. A permitted exception should have its own stated grounds and decision owner.

A fixed rule can save repeated deliberation, reduce inconsistent treatment, and make lobbying less rewarding. It can also make a commitment credible: employees can refuse an unauthorized request when everyone knows that an insistent customer cannot obtain an informal override. Those benefits depend on the restriction actually being respected.

But a rule can be poorly suited to an unusual case. Judgment may also migrate into classification: someone might label a fault differently to obtain the preferred response. Review must therefore examine the facts triggering the rule, not just whether the recorded response matches the recorded category.

The choice is between the gains from adaptation and the costs of inconsistent or misaligned discretion, including the costs of writing, administering, and revising rules. Neither unrestricted judgment nor a fixed response is always best. The next section considers an intermediate arrangement: local choice within explicit limits.

The fictional university-sanctions case applies the same question. Policy might settle the consequence without giving either dean or provost discretion to choose another one. Fact-finding, interpretation, and rule amendment would still need owners.

Make delegation specific enough to use

Consider a proposed bounded rule. Harbor permits a trained technician to approve routine catalogue parts totaling at most $150 per repair, with documented diagnosis, no unresolved safety or technical concern, and no conflict with reserved stock or other appointments. Every condition must hold. This is an illustrative proposal, not an estimate of Harbor’s optimal limit.

The operating sequence could be:

1. Establish the case. Record the diagnosis, proposed parts, combined cost, and reason for the choice. Confirm that the technician is trained for this repair and that the parts are in the routine category.
2. Check the boundary. Confirm the amount and technical conditions. Check current stock reservations and whether finishing would disturb the dispatch schedule. The customer’s existing authorization must still cover the work.
3. Act or escalate. If every condition holds, the technician approves and records the purchase. Otherwise, send the case to the supervisor, with dispatcher input when shared resources are involved. A designated backup handles an unavailable supervisor; silence is not approval.
4. Review and learn. The supervisor reviews a sample of local decisions and flagged escalations. Another reviewer checks supervisor-approved cases. Use the findings to revisit training, the permitted category, and the rule itself.

The unit “per repair” matters. Two related parts costing $90 and $80 total $170. They require escalation even though neither individual purchase exceeds $150. Otherwise a spending limit encourages dividing one decision into several records without reducing its consequences.

A $140 nonroutine part also requires escalation. Conversely, a $150 routine purchase satisfies the amount test, but still needs the other conditions. A cheap action can be technically difficult or affect an important shared resource. Price is one boundary, not a complete measure of risk or coordination needs.

An escalation route needs an actual decision-maker. “Ask management” is incomplete if several people can pass the request along without taking responsibility. Harbor should identify who receives the request, who substitutes when that person is unavailable, and what evidence that person needs. Escalation is not permission to waive a binding requirement. Pending a decision, the technician follows the established safe holding procedure rather than treating urgency as permission to improvise.

The policy also needs a way to revise its own categories. If almost every routine repair requires escalation, the boundary may be poorly drawn. But a high approval rate does not prove review is useless: review might deter weak proposals before submission. Harbor needs evidence about both the cases it sees and the behavior its rules encourage.

Recorded reasons help, but documentation is not free. A few decision-relevant fields may support useful review; a long narrative on every minor repair can consume more time than it saves. The test is whether the record helps someone distinguish sensible action from avoidable waste, not whether the file looks thorough.

Formal authority can differ from actual discretion

Suppose the supervisor formally approves every above-limit purchase but lacks the time or expertise to question a recommendation. The technician’s proposal may determine the result in practice. The signature remains at headquarters while effective influence stays with the person who selects the facts and alternatives.

This arrangement can be worse than explicit delegation if it creates delay without scrutiny. It also makes accountability unclear: the technician points to the approval, while the supervisor points to the technician’s diagnosis. Harbor needs to know which judgment each person was expected and equipped to make.

The opposite problem is possible too. A policy announces local authority, but supervisors criticize every decision that later turns out badly, including reasonable decisions made within the rule. Technicians may respond by requesting permission anyway. The formal limit has changed, but the practical incentive is still to avoid deciding.

This does not mean local decisions should be immune from consequences. It means review must distinguish a violated boundary, a poorly supported judgment, and an unfavorable outcome despite reasonable care. Giving someone discretion while treating uncertainty as disobedience makes the grant difficult to use.

Information acquisition can change as well. A technician who expects every proposal to be ignored may stop investigating alternatives carefully. A supervisor who expects to approve requests may invest in learning the relevant repair categories. These responses are possibilities to investigate, not automatic effects. The assignment of authority can influence which knowledge people acquire, not just how they use knowledge already present.

Choose a workable arrangement, not a slogan

For Harbor, bounded delegation is worth examining because the opening problem combines local diagnosis, routine purchases, and delays. It is not established as the best policy merely because it lies between central control and unrestricted parts approval. Its advantage depends on whether the category is recognizable, the relevant information is available, and review can discourage costly misuse without recreating the original queue.

Central approval becomes more attractive when unusual decisions have important consequences across jobs and timely review adds knowledge the technician lacks. Broader local approval becomes more attractive when useful knowledge is difficult to communicate, decisions are time-sensitive, and their consequences can be evaluated at reasonable cost. Neither conclusion follows from the size of the firm or the wording of a job title alone.

A limited trial could track approval delays, return resources, unnecessary parts, effects on other appointments, and time spent documenting and reviewing. Compare similar repairs over a common follow-up period. Ask whether job difficulty, staffing, or parts availability changed before crediting a new rule with better results. Keep the relevant service and safety constraints fixed.

Evaluation and rewards must change coherently with authority. If Harbor delegates parts purchasing but continues to reward only the number of closed visits, the old incentive to finish quickly remains. If it rewards low parts spending alone, technicians may avoid justified replacements. The next chapter examines how to evaluate performance without confusing a convenient measure with the objective.

Study map: assign a particular decision

Identify the action and its consequences before choosing who controls it. Locate the diagnostic, scheduling, and wider organizational knowledge it requires. Compare the cost of moving that knowledge with the cost of delegating authority, including agency and coordination problems. Separate proposing, approving, implementing, and reviewing. Ask separately which responses policy should prescribe and who may interpret, waive, or amend that policy. State the permitted category, limits, required evidence, escalation route, and later review. Finally, ask whether evaluation and rewards make the intended use of discretion attractive.

Practice and discussion

These are public practice questions. Use the chapter’s assumptions unless a question changes them.

1. Change the scheduling cost. The necessary part still costs $120 and a return visit uses $90 of extra resources. Suppose finishing now displaces only $70 of contribution from another job. Compare the choices. At what displaced contribution do they tie? Why is it incorrect to say immediate installation saves the entire $210?

2. Apply every condition. Under the proposed $150 rule, classify three separate requests: two related routine parts costing $90 and $80 for one repair; one nonroutine part costing $140; and one routine part costing exactly $150. Assume all conditions not mentioned are satisfied. Which can be approved locally, and why? What changes if the last part is reserved for another job?

3. Separate the functions. A technician proposes a purchase, a supervisor clicks approval without examining it, and the technician installs the part and later marks the repair successful. Identify the formal proposal, approval, implementation, and review functions. Which function is missing or weak in practice? Propose a feasible check and identify its cost.

4. Defend a bounded rule. In about 150 words, recommend a trial arrangement for Harbor. Specify the decision being delegated, information needed, one limit, the escalation route, and who reviews it. Explain one reason your proposal might perform worse than central approval and one observation that would make you revise it. Distinguish what your rule prescribes from what it leaves to judgment, and identify who may amend it.

5. Change the information source. A hypothetical AI tool recommends a $130 routine part but has no diagnostic result and cannot see stock reservations. A manager suggests automatic purchase because the price is below $150. Explain why the existing rule does not authorize that conclusion. Distinguish improving the proposal from granting purchasing authority.

Check your arithmetic and boundaries

For question 1, immediate relevant cost is $120 + $70 = $190, compared with $210 later: a $20 saving. The choices tie when displaced contribution is $90. The $120 part is common to both alternatives. For question 2, $90 + $80 = $170 exceeds $150; the $140 nonroutine purchase fails the category condition; exactly $150 passes the amount condition. Only the last request qualifies on the stated facts, and a conflicting stock reservation would require escalation. These answers check the illustrative rule, not whether $150 is the best limit.

Source note

Hayek grounds the dispersed-knowledge distinction, Jensen and Meckling the knowledge-transfer and authority comparison, and Fama and Jensen the decision-function terminology. Alonso and Matouschek supply the bounded attribution about delegating a permitted set of choices. The repair, numbers, proposed rule, separate prescribed-response protocol, table, AI scenario, and exercises are original teaching constructions, not observed findings or recommendations taken from those sources. The chapter offers organizational reasoning, not technical repair instructions, legal advice, or an evaluation of a particular AI product.