Economics and incentives
The principal-agent problem
When you pay someone to act for you, their interests differ from yours and you cannot see all they do, so expect them to serve themselves at the margin.
Michael Jensen and William Meckling, "Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure", Journal of Financial Economics (1976).
A principal engages an agent to act on their behalf: shareholders hire managers, patients consult doctors, home sellers hire estate agents, you hire a lawyer. The agent knows more about the work than the principal does and has interests of their own. Jensen and Meckling built a theory of the firm around this gap. They defined agency costs as the sum of what the principal spends on monitoring, what the agent spends to signal good behaviour, and the residual loss that remains because no contract can close the gap completely. The paper showed how these costs explain who owns a firm, how much debt it carries and why the separation of ownership from control matters.
It holds because of two things together: the agent sees what the principal cannot, and the agent's pay moves only weakly with the principal's outcome. An estate agent's commission is a small share of the price, so an extra $10,000 on a sale is worth a few hundred dollars to the agent against weeks of extra showings. Contracts cannot specify everything and monitoring costs money, so some gap always remains. In practice, nearly all of corporate governance (boards, audits, stock options, performance pay, reputations, repeat business) is an attempt to shrink agency costs, and the design question is always which costs less, watching the agent or paying them to care.
Treating every agent as a self-server is itself costly. Professional norms, intrinsic motivation and trust do real work, and heavy monitoring can crowd them out. Pay-for-performance invites Goodhart's law, because the agent now optimises the measured outcome. And the fix for one agency problem often creates the next: stock options tied managers to the share price, and then to whatever moved the share price this quarter. Moral hazard is the special case where the agent is insured against the consequences of their choices. The model does not tell you that your agent is cheating; it tells you where to look.
Example
In a 2005 National Bureau of Economic Research working paper, Steven Levitt and Chad Syverson compared houses that real estate agents sold on behalf of clients with houses the agents owned and sold themselves. After controlling for a wide range of housing characteristics, agent-owned homes sold for about 3.7 percent more and stayed on the market about 9.5 days longer. The explanation is the commission: an agent keeps only a small share of any extra price, so when working for a client a quick sale beats waiting, but when selling their own home the agent waits for the better offer.
Where it applies
Whenever you outsource (a contractor, a vendor, a recruiter, a managed AI service), ask how their pay moves with your outcome and what they can see that you cannot. An AI agent acting on your behalf is an agent in this sense too: it optimises its instructions and its reward, not your interests, so log its actions and check a sample.
Think about it
Who acts on your behalf this week, and what would they gain by doing slightly less than their best for you?
Applied it to something real? Log it.
Test yourselfWhy did estate agents' own homes sell for more than their clients' homes in Levitt and Syverson's study?
The agent keeps only a small share of any extra price, so for a client a quick sale beats waiting. Selling their own home they keep all of it and wait: about 3.7 percent more, 9.5 days longer.
Related models
- Incentives drive behaviourPeople do more of what a system rewards and less of what it punishes, so to predict behaviour, look at what gets paid for.
- Moral hazardWhen someone else will bear the cost of your risks, you take more of them.
- Goodhart's lawWhen a measure becomes a target it stops being a good measure, because people optimise the number instead of the thing it stood for.
- Incentive-caused biasAn incentive bends your own judgement towards the conclusion that pays, and you will not feel it happening.
Sources and further reading
- Michael C. Jensen and William H. Meckling, Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure, Journal of Financial Economics 3(4) (1976), record and abstract at IDEAS/RePEc
- Steven D. Levitt and Chad Syverson, Market Distortions when Agents are Better Informed: The Value of Information in Real Estate Transactions, NBER Working Paper 11053 (2005)