Marginal Analysis

Thinking Tools5.2 Selecting the Right Decision Tools

From the marginal revolution in economics, Hermann Heinrich Gossen (1854), William Stanley Jevons (The Theory of Political Economy, 1871), Carl Menger (1871), and Léon Walras (1874), set in textbook form by Alfred Marshall (Principles of Economics, 1890).

Weighs what one more unit of staff, spending, or output would cost against what it would return, and stops adding where cost overtakes return.

Each added agent costs the same and returns less, so stop where the bars drop below the line

A support desk staffs its evening shift. Each bar is what one more agent returns a week in calls answered before customers give up; the black line is what that agent costs. Set how many agents to staff and what one costs: the net value peaks at the last agent whose bar clears the line, and every agent past it costs more than it brings in. Illustrative.

6
$900
Return from the last agent added
$570
Net value of the shift a week, at this many agents
$2,491
Agents whose return clears their cost
4
Net value a week at that size
$2,962
Value given up a week by staffing this many
$471

Return from each added agent, $ a weekCost of one more agent, $ a week

01,0002,0002,4001,8001,3501,0127595704273202401801234567891001,0002,0002,4001,8001,3501,01275957042732024018012345678910

An example

The support desk of an online furniture retailer was losing customers on weekday evenings: callers waited, gave up, and many did not come back. The head of support asked for six more agents on the evening shift, citing the average value each agent produced across the whole team.

The finance partner asked a narrower question: what would each additional agent add? Using a month of call logs, they estimated the value of calls answered before callers gave up. The first extra agent would recover about $2,400 a week, the second about $1,800, and each one after that about a quarter less than the last, because every added agent answered calls that were a little less likely to be lost. Each agent cost $900 a week. The fourth agent still paid; the fifth returned about $760, less than she cost. The desk hired four, which netted close to $3,000 a week, about $470 more than six would have.

Why it works

Averages answer the wrong question. The average value per agent says how the team is doing; the decision is whether the next agent is worth her cost. When returns fall as more is added, as they usually do, the average stays above the cost long after the next unit has stopped paying, so decisions made on averages overshoot.

Marginal analysis compares the cost of one more unit with what that unit returns, and adds units while the return is larger. The net value of the whole is highest exactly where the two cross. The same logic sets how much to spend on advertising, how many hours to run a machine, and where to set a price. It also explains why the answer shifts when costs change: a cheaper agent moves the crossing to the right without any change in the work.

The thinking behind it

William Stanley Jevons William Stanley Jevons, an economist and logician, was one of three founders of marginal analysis, alongside Carl Menger and Léon Walras, who reached the idea independently in the early 1870s. In The Theory of Political Economy (1871) he argued that value depends on the final degree of utility, the utility of the last increment of a good, and stated economic choice in mathematical terms.

Hermann Heinrich Gossen Hermann Heinrich Gossen, a Prussian civil servant, stated the law of diminishing marginal utility in a book of 1854: the satisfaction from each further unit of a good declines. He drew rules for conduct from it. The book went almost unread until a colleague of Jevons found a copy in 1878. The falling bars in the figure have the same shape, carried over from satisfaction to the returns from added staff.

Alfred Marshall Alfred Marshall, a Cambridge economist and a central figure of neoclassical economics, wrote Principles of Economics (1890). It made marginal reasoning, comparing the cost and the benefit of one more unit, the standard method of the economics textbook, with supply and demand as its frame. In that frame a firm hires, produces, or prices up to the point where one more unit would cost more than it brings in, the rule this page applies to a support desk.

How to run it

  1. Name the unit of the decision. One agent, one thousand dollars of spend, one hour of overtime: the smallest step that can be added or removed.
  2. Estimate what the next unit costs. Count its full cost, including training, management time, and space.
  3. Estimate what the next unit returns. Measure the added result of that one unit, from logs or a small test, never from the team's average.
  4. Add units while the return exceeds the cost. Line them up in the order they would be added and keep going while each pays.
  5. Stop where cost overtakes return. The first unit that returns less than it costs marks the size to hold.
  6. Cut back if past the crossing. Apply the same test in reverse to units already in place.
  7. Recalculate when conditions change. New prices, wages, or demand move the crossing, so redo the numbers at each budget cycle.

Where it goes wrong

The common failure is deciding on averages. A team whose average agent earns twice her cost can still be overstaffed, because the last two agents add little. Ask for the value of the next unit, measured or tested, before approving more.

A second failure is assuming the return curve. Returns usually fall, but not always: a sales team may need a minimum size before it covers a territory at all. Test a small step before assuming the shape. The third is ignoring what a unit enables beyond its direct return, such as the agent who lets the desk stay open an hour later. Count those effects in the return, and say so, so the analysis is not overruled as narrow.

When to use something else

  • Queueing Theory. When the return of each added server comes from waiting times, and the arrivals and service times are known well enough to model.
  • Opportunity Cost Reasoning. When the question is whether to spend the money here at all, against its best other use.
  • Linear Programming. When several resources and products share limits and the best mix must be solved together.

Where it sits in SOLVE eX

Go to the source

  • Entwickelung der Gesetze des menschlichen Verkehrs, 1854
  • The Theory of Political Economy, 1871
  • Grundsätze der Volkswirtschaftslehre, in English as Principles of Economics, 1871
  • Éléments d'économie politique pure, first part 1874, second part 1877, 1874
  • Principles of Economics, 1890

Back to the catalog