Interview Prep

Secretary Problem Answer: Interview 37%, Then Hire [Solved]

Interviewer facing a long line of secretary candidates in the secretary problem answer

The secretary problem answer is to interview the first 37% of candidates, reject them all, and then hire the first candidate who beats everyone you have seen so far. With 100 candidates, that means skipping 37, then hiring the next record-setter. Follow this rule and you land the best candidate about 37% of the time.

Quant trading firms ask versions of this puzzle because it tests the exact skill traders use every day: deciding when to stop looking and commit. The setup is simple, the answer feels wrong, and the number 37% comes from nowhere obvious. This walkthrough gives you the rule, the small cases you can check by hand, and the short proof that 1/e is the best any strategy can do.

The Secretary Problem Answer, Stated Plainly

You need to hire one secretary from 100 candidates, arriving one at a time in random order. After each interview you must decide at once: hire this person, or pass. A passed candidate is gone forever. You can rank candidates against each other, but you only ever see relative ranks, never an absolute score.

Two facts shape the whole problem. First, you have no idea how good the pool is until you have seen some of it, so hiring the first strong candidate is a guess in the dark. Second, every candidate you pass could have been the best, and you never find out. The tension between those two facts is the puzzle.

Solving the Secretary Problem Answer Step by Step

The winning strategy has two phases. Phase one is pure sampling: interview the first 37 candidates and reject every one of them, no matter how good they look. Your only job is to learn the standard, so remember the best candidate you saw, because that person sets the bar. Phase two is selection. Starting with candidate 38, hire the first person who beats everyone you have seen so far. If nobody beats the bar, hire the last candidate.

Why 37, and why not 10 or 50? The table shows the best cutoff and the success chance for smaller pools, each row exact, worked out by counting every possible arrival order.

Candidates (n) Best number to skip Chance of hiring the best
3 1 50%
4 1 46%
5 2 43%
10 3 40%
100 37 37%

As the pool grows, the best cutoff settles at about 37% of the pool, and the success chance settles at the same number, about 37%. That match is not a coincidence: both numbers are 1/e, where e is Euler’s number, about 2.718.

Check the smallest case by hand and the logic clicks. With 3 candidates, skip the first one. If the best candidate is second, you hire them, because they beat the first. If the best candidate is third, you hire them too, because the second candidate cannot beat the first unless the second is better than the first, and in that case you would have hired the second by mistake. Work through all six orders and you win in three of them, a 50% chance. A random pick would win only 1 time in 3.

Why 37%? Where 1/e Comes From

Fix a cutoff r and ask for the chance of winning. You win only if two things happen: the best candidate sits at some position after r, and the best of everyone before them sits inside the skipped first r, so you do not stop early on a weaker candidate. For a best candidate at position k, the second condition holds with probability r/(k-1). Add that up over every possible k and the chance of winning is (r/n) times the sum of 1/(k-1) from k = r+1 to n.

That sum is a piece of the harmonic series, and for large n it is close to the natural logarithm of n/r. So the chance of winning is close to (r/n) times ln(n/r). Call the fraction you skip x, so r/n = x, and the chance is x times ln(1/x). This curve rises from zero, peaks, then falls, and its peak sits exactly at x = 1/e, about 0.368. At that peak the chance of winning is also 1/e, about 37%.

The curve explains both mistakes. Skip too few and your bar is set by a tiny sample, so you stop early on a candidate who only looks good against weak competition. Skip too many and you throw away most of your chances to act. At 37%, the gain from a better bar exactly balances the loss from fewer chances left.

The Traps Most Candidates Fall Into

The first trap is hiring the first great candidate. A candidate who shines in the first few interviews feels like a find, but with no sample behind you, “great” means nothing. You have no scale yet, so you cannot tell a strong candidate from the best one. The rule forces you to build the scale first, and that discipline is the point of the puzzle.

The second trap is waiting for perfection. Candidates who understand sampling sometimes swing the other way and keep rejecting, hoping for someone clearly better than everyone so far. By the end of the pool the bar is so high that nobody clears it, and the rule hands you the last candidate by default. The 37% cutoff is the answer to that fear: it tells you exactly when the sampling phase ends and the committing phase starts.

The third trap is treating the rule as a hiring manual. Real hiring breaks every assumption the puzzle rests on. Candidates do not arrive in random order, recruiters front-load strong CVs. You can usually call a rejected candidate back. And you want a good hire, not necessarily the single best person in the pool. The 37% rule is a benchmark for an idealized problem, not a policy for a real hiring round, and a strong candidate says that out loud.

What the Interviewer Is Testing in the Secretary Problem

First, they test whether you can think in trade-offs instead of hunting for certainty. The best strategy in this puzzle still fails about 63% of the time, and the interviewer wants to see if that bothers you. A candidate who keeps pushing for a safer rule has missed the point: when the information is incomplete, the right answer is the best available odds, not a guarantee.

Second, they test structured decomposition. A strong answer breaks the problem into sampling and selection, works the small cases by hand, and then generalizes. The interviewer watches how you set up the counting, not whether you can recite 1/e from memory.

Third, they test whether you connect the puzzle to real decisions. Optimal stopping shows up everywhere in trading: when to enter a position, when to take profit, when to cut a loss. Each of those is a version of the same question, how much information is enough, and when does waiting cost more than it is worth.

Try Another Puzzle Next

If you liked the slow-burn probability here, try the Morgan Stanley gambler’s ruin puzzle next: two players, three dollars between them, and an answer of 4 in 7. The method is different, equations instead of a cutoff, but the lesson is the same: the obvious answer is usually wrong. Read it here: Morgan Stanley Gamblers Ruin Puzzle: Answer [Solved].

For a second one, the Monty Hall problem asks whether to switch doors after the host opens a goat, and the answer is yes, switch, with a 2 in 3 chance. Read it here: Monty Hall Problem Answer: Switch Doors [Solved].

Frequently Asked Questions

Is the secretary problem a real interview question?

Yes. Quant firms have asked versions of the secretary problem for decades, and it remains a regular feature of probability-heavy interviews. Jane Street, Citadel, HRT, Two Sigma, Optiver, and SIG ask versions of it. The problem itself dates to a Martin Gardner column in Scientific American in February 1960.

What happens if no later candidate beats the bar?

You hire the last candidate. If the best candidate sat inside the skipped group, no later candidate can beat the bar, so you run out the pool and take whoever is left. That loss is already counted in the 37% success chance.

Does the 37% rule work in real hiring?

Only as a benchmark, not as a policy. Real hiring breaks the assumptions: the order is not random, rejected candidates can often be called back, and you want a good hire rather than the single best. The value of the rule is the shape of the thinking, explore first, then commit to the first record-setter, which applies to many search problems even when the exact number does not.

Sources: techinterview.org, “The Secretary Problem and the 1/e Rule”; Martin Gardner, Scientific American, February 1960. The solution and all text are original.

Enjoying this? Add StudyFromNotes as a preferred source on Google to see our latest notes and updates more often:

One response to “Secretary Problem Answer: Interview 37%, Then Hire [Solved]”

Leave a Reply

Your email address will not be published. Required fields are marked *