How many candidates should you interview?
Reading applications and running interviews are the same problem in different clothes: how far into the pile do you have to go before the best one you have seen is good enough? Fewer than you think — and the number barely moves with pool size. To be 90% sure the best of them sits in the top 10%, you need about 20, whether the pile is 50 or 500.
Pick your step below — reviewing applications or interviewing a shortlist. The maths is identical; the pool is not, so the answer changes with it.
Your applicant pool
= the best 20 of 200
Review this many
21
of your 200 applications, for a 90% chance the best of them is in the top 10%
- If you review 10
- 66%
- One more review adds
- +3.6%
- Reviewing all 200
- 100%
Reviewing every one would mean 179 extra reviews to buy the last 10% of certainty.
No benchmarks are used here — this is arithmetic over your own numbers. It does assume that the applications you open are a fair sample of the pile, and that reading one tells you who is stronger. The second is the one that bites.
Ten times the pile, four more looks
The instinct is that a flood of applications means a flood of screening. It doesn't. Here is what it actually takes to be confident the best candidate you screened is in the top 10% of the whole pool:
| Candidates in the pool | Screen this many for 90% confidence | For 95% |
|---|---|---|
| 50 | 18 | 22 |
| 100 | 20 | 25 |
| 200 | 21 | 27 |
| 500 | 22 | 28 |
A ten-fold bigger pile costs four extra looks. What a bigger pool really changes is who is in it to be found — not how deep you have to dig.
How the number is calculated
The whole derivation rests on one observation: "the best of the ones I screened is in the top 10%" is the same statement as "at least one of the ones I screened is in the top 10%". If the group you looked at contains any top-10% candidate, then the best one in that group is at least that good. So the question becomes a straightforward sampling probability:
P = 1 − C(N−m, k) / C(N, k)
- N — how many candidates are in the pool at this stage (everyone who applied, if you are reviewing applications; everyone shortlisted, if you are interviewing).
- m — how many of them clear your bar (the top 10% of a 200-candidate pool is 20 people).
- k — how many of them you actually screen.
No industry benchmark, multiplier, or survey figure appears anywhere in it. That is deliberate: the answer is arithmetic over your own numbers, so there is nothing here you have to take on trust.
Why this isn't the 37% rule
Search this question and you will be told to screen 37% of candidates and then hire the next one who beats them all. That is the secretary problem, and it is a real result — for a situation that is not hiring.
It assumes three things that are almost never true of a shortlist:
- You must accept or reject each person the moment you see them.
- Anyone you pass over is gone forever — no going back to a strong applicant from last week.
- Only the outright best person counts as a win; second-best scores the same as worst.
Real screening works the other way round: you go through a batch, keep everyone in it on the table, and then choose. That is a different problem with a different — and much cheaper — answer, which is what this calculator computes.
The assumption that decides whether this works
Every number on this page assumes that the screening step itself sorts people correctly — that reading the CV, or running the interview, reliably tells you who is stronger. That is the ceiling case, not the normal one.
It is also not a new caveat. Personnel-selection research has treated the accuracy of the screen as its central variable since Taylor and Russell published their tables in 1939, mapping how the quality of your hires depends jointly on how selective you are and on how well your selection method actually predicts performance. This calculator sits at the optimistic end of that second axis.
The practical consequence is worth stating plainly, because it cuts against what most hiring teams do when a role is going badly: if your screen cannot tell strong candidates from weak ones, screening more of them does not help. More volume means you have seen more of the pool. It does nothing for your ability to sort it. Screening depth and screening accuracy are two different problems, and only one of them is fixed by going further down the list.
Frequently asked questions
Does this mean interviewing candidates or reviewing applications?▾
Either — pick the step in the calculator and it relabels itself. The maths is identical at both, because both are the same act: you look at part of a pool and choose the best of what you saw. What differs is the pool. If you are reviewing applications, the pool is everyone who applied, often a few hundred. If you are interviewing, the pool is your shortlist, usually ten or twenty. Enter the pool for the step you are actually about to do, or the answer will be for a different question.
Why does the interview stage tell me to see almost everyone?▾
Because a shortlist is already the good part of the pool, so the top 10% of it is a far higher bar than the top 10% of all applicants — on a shortlist of twenty, it is two people. Finding one of two among twenty genuinely does mean interviewing most of them. That is not the calculator misfiring; it is the reason skipping people late in a process costs you much more than skipping people early. The place to be selective about your own effort is the application pile, not the shortlist.
How many candidates should you interview before deciding?▾
For a 90% chance that the best person you see is in the top 10% of the pool you are choosing from, you need about 20 — and that barely changes with pool size. It is 18 for a pool of 50 and 22 for a pool of 500. The number depends far more on how good a hire you need than on how many people are in the pool. Note that at the interview stage the pool is your shortlist, not your total applicant count.
Does a bigger applicant pool mean I have to screen more?▾
Hardly at all. Going from 50 candidates to 500 — ten times the pile — raises the number you need to screen for 90% confidence from 18 to 22. Four more. Pool size affects who is available to find, not how deep you have to dig to find them.
Is this the 37% rule?▾
No, and the 37% rule does not apply to normal hiring. That result (the secretary problem) assumes you decide on each candidate immediately and irrevocably, can never go back to someone you passed on, and only care about landing the single best person. Real shortlisting lets you review a batch and then choose among all of it, which is a different problem with a different answer.
What if I need the single best candidate, not just a top-10% one?▾
That gets expensive fast. The chance that the best of your sample is the outright best of the pool is simply the fraction you screened — screen 30 of 200 and it is 15%. Being confident of the true number one means going through nearly everything, which is why most hiring targets a quality band instead.
Does screening more candidates improve my hire?▾
Only up to a point, and only if your screening is accurate. Volume decides how much of the pool you have seen; it does nothing for your ability to tell a strong candidate from a weak one. If your screening step cannot rank people reliably, going through more of them samples a pool you still cannot sort.
What numbers does this calculator use?▾
None but yours. There is no industry benchmark, multiplier, or survey figure anywhere in it — the result is exact probability arithmetic over the applicant count, quality bar, and review count you enter.
Methodology & sources
The formula: the probability that a sample of k drawn from N contains at least one of the top m, which is 1 − C(N−m, k) / C(N, k). Exact, not an approximation or a simulation.
Coefficients used: none. Unlike cost calculators, which depend on published multipliers, this result is arithmetic over the numbers you enter.
What it assumes: that the ones you screen are a fair sample of the pool (sorting by relevance or recency will skew it), and that the screening step itself ranks correctly the ones it sees.
What it is not: the secretary problem or the "37% rule", which solve a sequential no-recall problem and do not apply to batch shortlisting.
The accuracy-of-screening axis this tool holds fixed is the subject of the Taylor-Russell tables (1939), still the standard reference in personnel selection.
Share or cite this calculator
Free to use, and free to link to. No sign-up required.
Cite this tool
Kaairo — How Many Candidates Should You Interview? https://www.kaairo.ai/tools/how-many-candidates-to-interviewScreening depth is the easy half. Screening accuracy is the other one.
Kaairo scores candidates on demonstrated competencies rather than résumé keywords — so the ranking you get back reflects what people can actually do, which is the assumption this calculator rests on.