MModern Wisdom
← All frameworks
StrategyRory Sutherland

The 37% Optimal Stopping Rule

Sample a third of your options, then grab the first one that beats them all

Difficulty
Easy
Time to result
~days to results
Steps
4
Confidence
72%

The rule solves the classic 'secretary problem': you interview candidates one at a time, must accept or reject each immediately, and cannot recall a rejected one. Spending too little time looking leaves you no benchmark for what 'good' means; spending too long risks exhausting the pool. The mathematically optimal split is to use roughly the first 37% of the field purely to observe — rejecting everyone — while noting the best you have seen. After that threshold, you commit to the very first option that exceeds that observed best. It trades the fantasy of a perfect pick for the highest probability of landing an excellent one, and turns an anxious open-ended search into a two-phase rule of look-then-leap.

Origin

Discussed by Chris Williamson and Rory Sutherland from Brian Christian and Tom Griffiths' book Algorithms to Live By, where it appears as the 'secretary problem' — historically framed as interviewing secretaries one at a time.

Core principles

  • 01You cannot judge a good option until you have seen enough of the field to benchmark it
  • 02Every option you inspect and reject is spent — waiting too long risks running out
  • 03The optimal cutoff balances too-little-information against too-few-candidates-remaining
  • 04A good-enough rule beats endless optimising when you cannot revisit rejected options

How to run it

  1. 1

    Estimate your pool size

    Decide roughly how many options you expect to encounter before you must have chosen — 10 flats, 30 candidates, a season of dates.

    Pro tip A rough estimate is fine; the rule is robust to being slightly off.

  2. 2

    Run a look-only calibration phase

    Deliberately reject the first ~37% of options no matter how good they seem. Their only job is to teach you what the range looks like.

    Pro tip Actively note the single best option you see in this window — it becomes your bar.

    Watch out It will feel wasteful to reject a strong early option; that discomfort is the cost of calibration.

  3. 3

    Set your benchmark

    Lock in the best option seen during the calibration phase as the standard everything after must beat.

  4. 4

    Leap on the first option that clears the bar

    Once past the 37% mark, commit to the first option that is as good as or better than your benchmark — do not keep looking.

    Pro tip Pre-committing removes the paralysis of second-guessing in the moment.

    Watch out Holding out for perfection past your benchmark is exactly the failure mode the rule exists to prevent.

In the wild

Choosing a life partner

Williamson raises the book's mischievous application: if you want to find a partner, don't try to evaluate everyone forever. Roughly sample the first third of the people you'd realistically date, using them to learn what 'good' feels like for you, then commit to the next person who is better than all of them. Sutherland notes the artificiality — you obviously don't cohabit with 33 people and ditch each — but the underlying logic of look-then-commit still guides when to stop searching and choose.

Converts an anxious, open-ended search into a clear stopping point that maximises the odds of an excellent match.

Hiring from a candidate queue

A founder must interview applicants one at a time and give an offer or pass on the spot before the candidate takes another job. Interviewing the first third with no intention of hiring lets them learn the true quality distribution of the applicant pool. After that, they hire the first candidate who beats the strongest person from that calibration set, rather than dragging the process out until the best applicants have accepted offers elsewhere.

A defensible, fast hiring decision that avoids both hiring the first warm body and losing top talent to indecision.

Common mistakes

Committing before you've calibrated

Choosing an early option feels efficient but you have no benchmark yet, so you cannot know if it is actually good.

Turning the look phase into a forever phase

Extending calibration past the threshold in hope of a perfect option risks exhausting the pool and being forced to settle for whoever is left.

Treating it as a guarantee

The rule maximises probability of a great pick, not certainty; occasionally the best option falls inside the reject window and is missed.

Is it for you?

Best for

Sequential one-shot decisions with a rough known pool size — hiring, dating, house-hunting.

Not ideal for

Decisions where you can revisit rejected options later or where the whole set is visible at once.

From the transcript

you interview the first 30, and then you've set your peak, the best of the 30 you've seen, and then the next as soon as…

Rory Sutherland

if you were trying to find a partner, you should 33 people in, and then once you find a person who is better than those…

Chris Williamson

it's partly the explore exploits trade-off, and it's partly the trade-off between having too little information to benchmark what a good secretary might be like…

Rory Sutherland

From the episode

19 Of Human Behaviour's Weirdest Quirks - Rory Sutherland - #587

Rory Sutherland