The person who must act cannot always be the person whose interests the action serves. A proprietor sends a buyer to inspect a shipment; a household entrusts a repair to someone who understands the machine; an owner appoints a manager who knows the undertaking better than its shareholders do. Delegation puts someone else’s judgment and effort to work. It also places a choice beyond the principal’s immediate sight. The deputy can see something the principal cannot, and may want something the principal does not. The useful freedom to exercise judgment is therefore also a freedom to depart from the purpose for which the judgment was engaged. The bargain must preserve the first while giving the deputy a reason to restrain the second.
The agency problem begins with a difference in consequences, rather than a finding of dishonesty. An extra hour of inspection costs the buyer his time; much of what it saves belongs to the proprietor. A familiar supplier may make the buyer’s day easier while charging the business more. A manager can prefer a quiet life to a demanding expansion even where both are lawful choices. Jensen and Meckling describe an agency relationship as a delegation of decision-making authority through an agreement to serve another. Its difficulty arises when the agent’s preferred use of that authority differs from the principal’s. The very knowledge that makes the deputy worth employing can make the departure hard to detect. The principal has hired a capacity he cannot cheaply reproduce, and so cannot simply solve the problem by doing all the work again.
The principal often sees the result without being able to separate its authors. A machine fails after a careful repair; a neglected machine happens to last. A buyer who selected well encounters an unforeseen interruption, while an indifferent buyer benefits from a shipment whose quality was unusually good. These possibilities do not make evidence useless. They make its interpretation consequential. A contract promising payment for a satisfactory outcome can influence conduct even when an individual outcome proves little about the conduct that produced it. Conversely, a demand for diligence written into the agreement cannot, by its wording alone, supply an observation that neither the principal nor the appointed adjudicator can make. The difficulty is to connect the intended action to consequences that the parties can actually administer.
A workable payment must secure both acceptance of the job and the action wanted within it. Retain the constructed worker from the earlier Agent, and imagine that the task is a repair. Careful work costs three money-equivalent units of effort and raises the chance of success from one-quarter to three-quarters. The worker is risk-neutral, knows these chances, and can earn two elsewhere. A payment of eight on success and zero on failure gives careful work an expected return of six less three, leaving three. Careless work gives two. He therefore accepts and works carefully. Yet paying six on success would leave him indifferent between the two actions and give only one and a half under either; a generous fixed payment could attract him while leaving careful work privately unrewarded. The two decisions have to be tested separately.
Observation can change the bargain without becoming complete. A service log, a sample inspection or a second opinion may distinguish a neglected repair from an unavoidable failure. In Holmström’s analysis, additional evidence matters when it adds information about action beyond what the outcome already reveals. It can make the reward more discriminating while reducing the need to expose an agent to unrelated chance. The useful observation need not follow him through every minute. But the observation has a price, and someone must be able to understand and use it. A report that duplicates what is already known adds no such information; an informative report can still cost more than the improvement it permits. The appropriate comparison is between feasible systems of delegation, each carrying its own costs.
Jensen and Meckling’s anatomy keeps the cost of restraint beside the loss it restrains. Monitoring is what the principal spends to observe or influence the agent, including rules and compensation arrangements as well as watching. Bonding is what the agent spends to limit particular departures or assure compensation for them. Residual loss is the remaining reduction in the principal’s welfare from the divergence of decisions. Agency cost joins all three. An audit consumes skilled attention even if it finds nothing; a restriction may prevent a valuable response as well as an abusive one; an assurance may require the agent to tie up resources. The account would be false if it counted every unfettered departure as waste while imagining that perfect oversight came free.
The best restraint can leave something uncorrected because correcting it would consume more than it saves. Consider three invented arrangements, all capable of attracting the same deputy. Without a safeguard, the stipulated departure costs the principal twelve. A modest system consumes three in monitoring and two in the agent’s bonding expenses, while leaving a residual loss of four: nine in total. An elaborate system consumes ten and three respectively and eliminates the residual loss: thirteen in total. On this agency-cost comparison, the middle arrangement is preferable to both neglect and complete correction. The example does not price the whole value of delegation or decree that nine is a universal minimum. It shows why an imperfect arrangement can improve upon the alternatives, and why a zero on the residual-loss line need not mean a successful economy of supervision.
To entrust a judgment is to arrange the consequences of a choice one cannot make again.
The person appointed to watch the deputy introduces another decision that must be governed. An inspector can hurry, a supervisor can favor an easy report, a board can accept the account it should have questioned. This does not imply an endless requirement to hire one watcher for every watcher. It means that authority, information and reward must be considered together at each consequential point. Some decisions can be checked against independent records; some can be divided; some can be made visible to people who bear their costs. Each arrangement must also leave room for initiative. If a rule prevents the agent from responding to a difficulty the principal could not foresee, obedience may destroy the very advantage for which discretion was delegated.
Fidelity needs a stated purpose before a departure from it can be judged. A repairer who refuses a dangerous shortcut is not shown to have served himself merely because the owner wanted a cheaper bill. A manager undertaking an authorized long investment is not shown to have betrayed a principal because the next distribution is smaller. The parties’ commitments, the powers legitimately entrusted and the interests affected have to be identified before the theory can explain how to carry out the undertaking. Agency analysis supplies no independent entitlement to every act a principal might demand. Nor does a model of self-interest establish that duty, pride or professional standards are absent; it isolates incentives whose force can then be examined alongside the other reasons people have to perform.
The deputy serves through an arrangement that makes faithful judgment possible and worthwhile. Payment, evidence, restraint and a valued continuing relationship can each supply part of it. Their value lies in what they enable the agent to do and what they discourage him from substituting for the task. Delegation can remain worth buying after all these costs are paid, because the alternative may be ignorance, delay or a job left undone. Yet its price is more than the wage recorded in the contract. Someone must decide how closely the reward follows the result when the result contains both work and luck. The next plate turns that decision into the spur: how much uncertainty should the principal place in the agent’s hands to give those hands a reason to work?