Two of your best engineers want to add a feature. It will take three weeks. Your gut runs the maths on the whole product, months of work, a roadmap, a strategy deck, and freezes. The useful question is much smaller: is the value of this one feature worth the next three weeks? That swap, from the whole to the next slice, is most of what microeconomics has to teach a manager.

The quick version

  • Think at the margin. Good decisions compare the next unit's benefit to its next unit's cost, not the total, and not the average. Keep going while the next step is worth more than it costs; stop when it isn't.
  • Ignore sunk costs. Money and time already spent are gone whatever you choose, so they shouldn't sway the next decision. Only future costs and benefits count.
  • People respond to incentives. Change the rewards and you change the behaviour, including in ways you didn't intend. What you measure and pay for is what you'll get.
  • Watch the perverse case. A measure aimed at one thing often quietly rewards another. The cheapest leadership win is removing an incentive that's pushing the wrong way.

The idea in depth

Decide at the margin, not the average

The word that does the work here is marginal, economist-speak for "one more." Marginal benefit is the extra value of the next unit; marginal cost is the extra cost of producing or consuming it. In his standard textbook Principles of Economics, Greg Mankiw lists "rational people think at the margin" as one of his ten foundational principles: good decisions are made by weighing marginal benefits against marginal costs, not by staring at the grand total.

The reason this matters is that averages lie to you. A team that ships ten features a quarter has an average value per feature, but the eleventh feature isn't average, it's the marginal one, and it's almost always worth less than the ones you already picked (you did the obvious wins first). The same logic runs the other way on cost: the marginal cost of the eleventh feature might be higher, because it competes for the same tired people. So when you're sizing any "should we do more of this?" call, one more hire, one more report, one more nine of reliability, don't divide the total by the count. Ask what the next one adds and what the next one costs, and act only while the first number beats the second.

flowchart TD
    A(["Considering one more unit
(hire, feature, hour, report)"]) --> B{"Marginal benefit >
marginal cost?"}
    B -- "Yes" --> C("Do it, then ask again
about the next one")
    C --> A
    B -- "No" --> D("Stop here.
You've hit the right amount")
Marginal thinking is a loop, not a one-off: keep adding while the next step pays, stop when it doesn't. Leaders Loop

A close cousin of marginal thinking is the discipline of sunk costs. Because only the next cost counts, money and effort already spent are irrelevant to what you do now, they're gone in every version of the future. The hard part is emotional, not analytical: we hate to "waste" what we've put in, so we keep funding doomed projects to justify the spend. The economically correct stance is colder and kinder: the £200k already burned on the failing platform is not a reason to spend £200k more. (This is one place theory meets behavioural economics, the sunk-cost fallacy is exactly the gap between what the model says and what humans actually do.)

People respond to incentives, including badly

Mankiw's fourth principle is the one every leader has lived: people respond to incentives. Because rational people decide by comparing costs and benefits, anything that shifts those, a bonus, a target, a public ranking, a fee, shifts behaviour. Steven Levitt and Stephen Dubner put it more bluntly in Freakonomics (2005): economics is, at root, "the study of incentives, how people get what they want, or need, especially when other people want or need the same thing."

The catch is that incentives are literal-minded. They reward the measured behaviour, not the intended one, and when those two come apart, you get what the economist Horst Siebert named the cobra effect, after a colonial-era bounty on dead cobras in India that ended with locals breeding cobras to cash in. The British got more cobras, not fewer. The same trap has a more academic name: Goodhart's law, after economist Charles Goodhart (1975), popularised in the crisp phrasing anthropologist Marilyn Strathern gave it in 1997, "when a measure becomes a target, it ceases to be a good measure."

"When a measure becomes a target, it ceases to be a good measure.", Marilyn Strathern (1997), on Goodhart's law

This is not a fringe curiosity. Wells Fargo's cross-selling scandal, surfaced in 2016, with the bank admitting to as many as two million accounts opened without customers' consent and a US$185m settlement with regulators, grew straight out of a sales incentive. Staff were pushed to hit aggressive product-per-customer targets ("eight is great"); when the target became the goal, employees met it by faking accounts. The metric was hit. The intent was destroyed. So the move is: before you launch any target, bonus, or league table, ask the cobra question, "if someone wanted to game this, what's the laziest way to win without doing the real job?" If a cheap fake exists, your honest people will resent it and your cynical ones will find it.

flowchart LR
    A(["You set a target
or reward"]) --> B(["People optimise for
exactly what's measured"])
    B --> C{"Does the measure
fully capture the
real goal?"}
    C -- "Yes" --> D(["Behaviour and intent
line up, keep it"])
    C -- "No (a gap exists)" --> E(["Cobra effect:
the gap gets gamed"])
    E --> F(["Measure looks great,
real goal suffers"])
Goodhart's law in motion: incentives reward the measure, so any gap between the measure and the real goal is where things go wrong. Leaders Loop

An honest limitation: people aren't pure calculators

The machinery above assumes people weigh costs and benefits like a spreadsheet. They don't, fully. In his TED talk drawn from Drive (2009), Dan Pink walks through decades of experiments showing that for work needing even rudimentary creative thinking, bigger cash rewards can worsen performance, they narrow focus and crowd out the intrinsic motivation (autonomy, mastery, purpose) that complex work runs on. Marginal analysis has its own blind spot: it assumes you can measure the next benefit and cost cleanly, and for fuzzy goods, trust, morale, brand, you often can't. So the move is: use these tools as lenses, not laws. Reach for explicit incentives on simple, measurable tasks; lean on purpose and autonomy for creative work; and treat any margin you can't actually measure with suspicion rather than false precision.

A worked example

Say you run a 20-person support team. Tickets are piling up, so you do two intuitive things: you set a target of 40 closed tickets per agent per day, with a small bonus for the top closers, and you weigh up hiring two more agents. (Figures here are illustrative.)

Run both through the toolkit. First, the marginal hire. Your team closes, say, 600 tickets a day between them. Two more won't add the average (30 each), the new agents need ramp-up and pull a senior off the queue to train them, so the next two might net you 35 closed tickets a day for three months, at a cost of two salaries plus that lost senior time. Compare the marginal benefit (35) to the marginal cost, not the comforting average, and the hire may or may not clear the bar. The average flatters it.

Now the incentive. A "tickets closed per day" target with a bonus is a cobra waiting to happen. The laziest way to win is to close easy tickets fast, or mark hard ones "resolved" prematurely so the customer returns tomorrow as a fresh ticket, inflating the count while satisfaction quietly falls. The measure (tickets closed) has drifted from the goal (customers helped). The fix isn't a bigger bonus; it's a better-aimed measure, say, resolved-on-first-contact and still-resolved-after-7-days, paired, per Pink, with genuine autonomy over how agents solve problems rather than paying piece-rate for volume. Same instinct as a reversible decision: pilot the new measure on one squad for a fortnight before you wire a bonus to it.

Frequently asked questions

Isn't "think at the margin" just common sense?

The principle is obvious; the practice isn't. The default human move is to judge a decision by the whole project or the average outcome, and to let sunk costs drag the next call. Marginal thinking is the deliberate habit of stripping a decision down to only the next unit's future benefit and cost. It feels obvious right up until you catch yourself defending a failing project because of what you've already spent.

What's the difference between marginal cost and average cost?

Average cost is total cost divided by total output, a backward-looking summary. Marginal cost is the cost of producing one more unit, right now. They can point in opposite directions: your average cost per feature can be falling while the marginal cost of the next feature is rising (because the easy work is done). Decisions about "more or less" should run on the marginal number; the average is for reporting, not deciding.

If incentives backfire so often, should I just stop using them?

No, incentives work, which is exactly why they're dangerous. The lesson from Goodhart and the cobra effect isn't "never incentivise," it's "incentivise the real goal, and assume people will optimise for precisely what you wrote down." Test every target with the cobra question before you launch it, prefer measures that are hard to fake, and keep explicit cash rewards for simple, measurable tasks (Pink's research suggests they can hurt on creative ones).

How do I spot a perverse incentive before it bites?

Ask three things. One: what's the laziest way to hit this target without doing the real work? Two: who has both the motive and the means to game it? Three: what behaviour am I implicitly punishing, for example, does rewarding individual output quietly penalise helping a colleague? If any answer is uncomfortable, redesign the measure before, not after, you roll it out.

Where does this sit relative to behavioural economics?

Classical microeconomics assumes rational actors who weigh costs and benefits cleanly; behavioural economics studies where real humans systematically deviate, loss aversion, the sunk-cost fallacy, motivation crowding. Treat them as a pair: the marginal/incentive model tells you what should happen, and behavioural economics tells you where it predictably won't. Good leaders use both lenses at once.

Related in the Toolkit

Where to go next