Marginal Decision Rule
Allocate the next unit using its next cost and return, not the historical average
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 99%
Marginal thinking asks what the next unit will cost and produce rather than what all previous units cost on average. Historical averages blend cheap early opportunities with increasingly expensive later ones, so they can point new spending toward the wrong option. Define the next tranche, inspect the cost and return of the last comparable units in each alternative, and estimate the next unit's performance. Allocate to the stronger marginal return, then measure again because the ranking can change as each channel saturates. List illustrates this with Lyft driver acquisition: Facebook looked cheaper on a six-month average, but its latest drivers cost twice as much as Google's latest drivers. The next advertising tranche therefore belonged on Google despite the historical average.
Origin
List explains the rule through a Lyft driver-acquisition meeting and earlier White House decisions about hazardous-site cleanup spending.
Core principles
- 01The next unit can have a very different cost from the historical average
- 02Past efficiency does not determine the best destination for new resources
- 03Allocation decisions should compare the last observed unit and the next expected unit
- 04Diminishing returns can reverse a ranking based on averages
How to run it
- 1
Define the increment
Specify the next dollar, hire, customer, site, or batch being allocated. Make the decision small enough that alternatives can be compared.
Pro tip Use the actual upcoming tranche rather than an annual abstraction.
Watch out A vague total budget hides the decision margin.
- 2
Measure the latest units
Calculate the cost and return of the most recent comparable units in each option. Separate them from the full-period average.
Pro tip Inspect the last 10, 20, or 50 acquisitions when volume allows.
Watch out Early cheap units can make a saturated channel look efficient.
- 3
Estimate the next unit
Use recent behavior to estimate what one more unit in each option will cost and return. Include any capacity or saturation effects.
Pro tip State the estimate as a range when uncertainty is material.
Watch out The next unit is not guaranteed to match the last one exactly.
- 4
Allocate and refresh
Send the next tranche to the option with the better marginal result, then remeasure. Continue reallocating as marginal performance changes.
Pro tip Use smaller tranches when the curves move quickly.
Watch out Do not turn one marginal comparison into a permanent channel rule.
In the wild
Lyft's team reported average acquisition costs of $500 per Facebook driver and $600 per Google driver and planned to favor Facebook. List asked about the latest drivers. They cost $700 on Facebook and $350 on Google, reversing the decision for the next tranche.
→ Marginal data pointed new spending toward Google even though Facebook had the lower historical average.
Common mistakes
Allocating from historical averages
An average can describe past spending while saying little about the cost of one more unit now.
Freezing the new ranking
Marginal returns change as resources move, so the comparison must be refreshed after each meaningful tranche.
Is it for you?
Best for
Budgets, acquisition channels, staffing, cleanup programs, and other choices made one tranche or unit at a time.
Not ideal for
Indivisible all-or-nothing choices where no meaningful incremental unit or marginal comparison exists.
From the transcript
“i don't care about the average in the last half year what i care about is how much did we have to spend to get…”
“what was the last driver and what would be the cost for the next driver that's marginal thinking”
“averages in some cases are very misleading when you really should be thinking about the last one and the next one rather than a big…”
From the episode
#566: John List — A Master Economist on Strategic Quitting, How to Practice Theory of Mind, Learnings from Uber, Optimizations to Boost Donations, the Primitives of Decision-Making, and How Field Experiments Reveal Hidden Realities
John List