Let's begin with the assumption that, at any given time, there are a finite number of people seeking a job. Each person only spends a finite amount of time seeking a job. Let's call this set of people S (for Seeking), where S is a subset of all living humans.
Let's say you're hiring for a specific position, e.g. Font End Developer. You don't want to accept resumes from short order cooks or airline pilots or geneticists, you only want resumes from people who are (or can quickly become) front end developers. Let's call this set P, again, a subset of all living humans.
Finally, you only want people who live in, or are willing to relocate to, a certain location e.g. Tampa. We'll call this L, again, a subset of all living humans.
Let's also accept that there is a cost associated with not hiring, i.e. the longer you wait, the more your business suffers for lack of a front end developer.
You can collect a bunch of resumes, and order them for fitness in a relatively short period of time (here we're assuming that a good resume is correllated with a hire decision, but imperfectly so). Interviewing takes much more time and is more resource intensive, so that is your bottleneck. You call in the best n for interviews. Assuming you can only interview at a fixed rate, you have a time window in which to interview as many candidates as possible (n) before making the decision to hire 0 or 1 candidates. In that time window, you will be able to sort candidates into two groups: well qualified and not well qualified candidates.
The smaller your time window, the smaller the likelihood that your interviews will produce at least one well qualified candidate. The the greater your time window, however, the greater the likelihood that a well qualified candidate will leave the market. It is in your best interest to interview in such a manner that many well qualified candidates are found, so that you can hire at least one.
Duh. Obviously. The only real control we have here is the set of people from which we collect resumes, R. In the above-defined example, R = S ⋂ P ⋂ L . We can change R, however. Because cardinality (S ⋂ P ⋂ L) ≤ cardinality(S ⋂ P), we can (probably) collect more resumes by removing the location constraint.
By collecting more resumes, the top n resumes in our ordered resume list will (probably) contain more well-qualified candidates. That increases the probability that we will be able to hire one well qualified candidate in the first allotted time window, thereby reducing the total amount of pain and resources spent interviewing and not hiring.