Principal-Agent Problem

A report can look good while the building gets worse

A landlord lives far from a small apartment building and hires a property manager. The manager receives a percentage of rent collected plus a bonus for keeping annual maintenance spending low.

For a while, the reports look excellent. Rent arrives, costs fall and the outside of the building remains tidy.

Inside one flat, a pipe leaks behind a wall. The manager sends someone to patch the visible stain instead of replacing the failing section. The proper repair would reduce this year's bonus, while most of the future damage will fall on the owner and the tenant.

The tenant keeps a bucket under the leak and sends another message that still doesn't reach the owner. The owner sees a clean ledger and assumes the building is being cared for. By the time damp appears in the next room, the cheap decision has become an expensive repair and a broken relationship.

Nobody had to lie for this to happen. The contract rewarded low visible maintenance spending, the manager knew more than the owner about the building, and the tenant carried the immediate cost.

The manager knows things the owner cannot easily see

The Principal-Agent Problem appears when one person, the principal, gives work or decisions to another person, the agent, and they have different information or interests.

Four conditions create the familiar shape:

1. The principal wants an outcome but cannot do or supervise every part of the work.
2. The agent controls important choices.
3. The agent knows more about effort, quality or local conditions.
4. The agreement can reward only what people can see, while the real outcome is more complicated.

The principal can see maintenance spending. The manager can see which repairs are urgent, which shortcuts are hidden and how much effort each option requires. A complete contract would need to describe every future fault and verify the quality of every repair. Real contracts can't do that.

Good intentions do not close the information gap. Decide what the owner can verify, which choices the manager can make alone, how quality is checked and how the tenant can escalate a bad repair.

More money is only one part of the answer

Money is only one way to narrow the gap, and stronger incentives can make one narrow number crowd out everything else. Use several levers together.

Improve information. Give the principal direct access to customer complaints, maintenance history, audit logs or random samples. Independent information reduces the agent's ability to define reality through a single report.

Shape incentives across time. Combine immediate measures with later outcomes such as whether customers stay, how often work fails or the long-term condition of the asset. Delaying part of the reward, requiring a warranty or giving the agent some ownership can make future consequences matter now. Too much delayed or uncontrollable risk can make good agents leave, so the balance matters.

Match approval to the cost of a bad decision. Let the agent move quickly on cheap, reversible decisions. Require a second view for irreversible or high-impact choices.

Let affected people reach the principal or an independent reviewer. Complaints, appeals, second opinions and the ability to change providers give the principal information a detailed contract can't provide.

The owner needs independent evidence of repair quality, rewards tied to long-term building condition and a direct route for tenant complaints.

The same problem shows up in different places

Economics and organisational design. A shareholder gives authority to executives, an executive gives delivery to managers, and a manager brings in a specialist vendor. Each delegation layer gets more local information, while some costs remain with the shareholder, customer or asset owner. Another dashboard changes little when the same layer being judged produces every dashboard, so independent signals and clear decision rights matter more than more reporting.

Law and governance. Fiduciary duties require someone acting for another person to put that person's interests first. Disclosure rules, independent directors, warranties and professional licences add other ways to reveal information or hold agents responsible when a contract cannot spell out every future decision. They still need evidence and enforcement before they change behaviour.

Software and automated agents. Automated software creates a related risk. It may have no private interest, but incomplete instructions, a narrow measure of success and permission to do too much can still produce an outcome the user would reject. Limits on what it can do, records of what it did, permission added in stages and actions that can be reversed give the user ways to inspect and stop it. This still differs from the usual Principal-Agent Problem because the software has no competing interest of its own.

Use it when the work is hard to see

Use it whenever someone acts on behalf of someone else and the work is difficult to observe directly: hiring vendors, designing executive compensation, giving teams autonomy, using professional advice, managing remote assets or granting software permission to act.

Ask:

• Who chooses, who benefits and who carries the downside?
• What does the agent know that the principal can't easily verify?
• Which valuable outcomes are absent from the contract or dashboard?
• Can the agent improve the visible result by shifting cost across time or onto someone else?
• Which independent signal would reveal the difference between surface performance and durable quality?
• How cheaply can the principal reverse, appeal or exit the arrangement?

Low cost, high speed and no complaints are useful signals only when the principal also has an independent way to observe quality and suppressed complaints.

Where the model can mislead

The agent may have better local information and a broader understanding of the work than the principal. An executive far from the work may have worse information and a narrower goal than the agent. Heavy monitoring can destroy trust, slow judgment and encourage people to optimise for surveillance. A contract that transfers every risk to the agent may become unfair or attract only agents willing to hide risk.

Use the model to improve information, incentives, decision rights and appeal paths. The problem can exist even when the agent is honest. Local knowledge may justify more discretion, while decisions that are hard to reverse still need better feedback or an independent review.

A target can make an existing information gap worth exploiting, but Goodhart's Law can also happen without delegation. A measure can become a poor guide after it's turned into a target. The Principal-Agent Problem requires delegated action plus a gap in interests or information. A badly chosen target often becomes the mechanism through which an agency problem operates.

Moral hazard is one narrower agency pattern: protection from downside changes how much risk the agent takes. Moral hazard can produce a Principal-Agent Problem, but agency problems also arise through hidden effort, concealed quality and different time horizons without unusual risk-taking.

If the agent both earns the reward and produces the only performance report, add an independent signal of quality or harm.