Recruitment pipeline costs — filter filter
While writing these posts (starting from here) and when re-reading them I noticed that many of the other article recommendations that touch on recruiting are something like “why recruiters suck”, “how to write a killer CV”, “how to recruit some-profession” or … anyway, from both my personal experience and articles recommended I sort of get the feeling that people looking for a job don’t really understand how recruiting works internally…
While I have already touched a bit on why rejections of CVs can appear entirely random, another thing that seems a major gripe is the lack of personalised contact with applicants. It is not a big stretch then to start thinking that this is caused by the use of keyword filtering. Keyword filtering is the process of keeping only those CVs that contain some keyword though to correlate with the sought skills (think “java”, or “full stack developer”). The implication is that all applications without the “magic keywords” are simply rejected.
While I personally hire only for smaller companies and I absolutely need to check each application carefully — I don’t do keyword matching — keyword filtering is definitely something that happens. Keyword filtering can be fair or unfair, but … complaining that it is an unfair method of filtering applications is really unfair. Unfair for the companies, actually.
So I’ll come back to the recruitment pipeline and take a look on how it appears to work from within the company.
The picture below shows a simplified schematic of a single pipeline phase. For screening this would contain CV evaluation, possible phone or online interviews and tests etc. (I have highlighted in red the two things a recruiter wants to minimise: false negatives and false positives. These are directly related to the “quality” of hires.)
Now here’s the thing: the number of applicants that can be passed to the next phase is limited by time and resources.
Here is where screening and evaluation are distinctly different. A company generally can not control the number of applications. You might get a few or you might get swamped — although usually a reasonable estimate can be inferred from experience. Evaluation — interviewing is usually very resource-intensive (costly!) and often there is also an externally imposed deadline. In practice there is a limit of N people that can be evaluated in detail, for any one successful hire. N thus is the number of people who will get through screening and get called in for actual interviews.
Note that not all people who are considered a positive match get interviewed. Assuming that after screening there are M people who were considered good fits then typically M > N meaning you have to drop the excess. As I said earlier in this series this is done usually by ranking the applicants in some manner and picking N “best.”
Now in turn M is derived from the total number of applications K of which then quite a bit are considered a poor fit (negatives). If K > M > N we get M-N people who are dropped (good fit, but not ranked high enough to pass) and K-M who are considered not a fit (and rejected).
Seems pretty straightforward, no?
Let us consider two cases: a small company looking for a “high-level hire” (management, senior specialist etc.) and a large enterprise hiring for an entry-level job.
Small company, high-level position
This is going to be difficult for the company. An established small company with good coffers is usually better off giving this type of task to a recruiting specialist (“head hunter”) just to save their own time and effort, but assuming this is a not-so-well-funded (*cough* start-up) company they will have to handle the recruitment pipeline themselves.
K is going to be low (for various reasons). Say K = 10. Total of 10 applications. Out of this we might consider M = 6 as good fits and decide to proceed to interviews with N = 3 best candidates. The screening process might take a work-day or two for all applicants — pretty insignificant overall.
In this case the company is not going to be constrained by interviewing resources. You might have two or three interviews each lasting about two hours and two interviewers equaling 5–15 hours of work-time used per applicant. Depending on the position this might be a low or high estimate, but we are still in the general ballpark where a company can interview thoroughly any “good enough” candidate.
This is a good situation to be for the applicant. They will not be screened on keywords and will get personalised and direct communication. Each applicant is a valuable opportunity for the company. If there is even a chance that an applicant would be a fit then the recruiter is going to dig deeper and ask for any information that might be relevant but missing from the application.
(If you do the maths you’ll get rejection rate of 40% and pass rate of 30%.)
Large enterprise, entry-level position
Entry-level jobs are often applied for even by experienced people in the hope their application will be picked out for some unpublished (better) position. Large enterprises also have other things going for them: they are better known, have established recruitment marketing channels and as a large company at least would appear to offer better job security. These and many other factors drive the number of applications up.
K will be large. It can easily be hundreds, so let’s pick K = 500. (You can check the net to see this is not unrealistic.)
If the company is hiring 5 people for the job, they are probably going to interview something in the order of 10–20 people. Let’s pick N = 20. Even for an entry-level job (one interview only) that is going to mean a work-week of interviews, thus there is absolutely no desire to increase N. Eventually the exact value of N in our example does not matter since it is going to be much, much smaller than K. Most of the people applying have to be rejected or dropped and will never get a chance to be interviewed. This is simple, brutal and cold maths. There just is no time, no money, no people to interview any significant percentage of a large number of applications.
But wait, there’s more!
Even screening these people can be very expensive. If the company used 45 minutes to screen one application (like the small company above) this would eat 50 work-days to screen all of them. The time to screen one applicant has to be made much smaller. Let’s turn the time problem around and start with an assumption of how much time is available: say a total of one-work week. This means that to screen all 500 applications you have 4.5 minutes per application. So this is now only screening, not interviewing.
Since a proper screening requires ranking of more applicants than are passed through this means M must be larger than N. So N=20 are picked out of a larger pool of M=50 applicants who get a more thorough screening. Let’s say these M people each get an additional screening of 25 minutes per application (this actually needs to include the time taken to make the ranking, too).
This leaves 1000 minutes of the original time budget to decide who will be the M=50 out of original K=500.
Do the maths.
That comes out as two minutes per application.
I also did describe one work-week for 500 applications as “lavish.” Any less will mean even less time for the first screening step.
Do you now understand why recruiters are said to have only a few tens of seconds to look at your application? Why they use keyword filtering? Because for any position that gets hundreds or thousands of applications there absolutely is no other way to do screening.
(Artificial intelligence might eventually get recruiters out of this rabbit hole as then you can parallelise screening and just throw more CPU cycles at it.)
The alternative way out of this rabbit hole is that entry-level jobs should be only applied by those who have realistic chances to get them. This would result in fewer applications leading to more time to screen each application. Of course this won’t happen. Ever heard of tragedy of the commons? This is it.
(Here the rejection rate is 90% and pass rate is just 4%.)
I’ll shut up soon so bear with me a little bit more. I want to talk about open applications and continuous hiring. Both of the examples above were for recruiting to specific positions (batch processing for you geeks). Some types of companies are in constant process of recruiting and have a continuous, non-stop stream of open applications.
If you’ve followed with me so far you can probably guess that the recruitment mechanics and priorities are going to be different here, again?
One problem with open applications for the company is that often job description is pretty broad (“we are always looking for competent people in software industry!” what does that tell you?). This can mean that a proportionally larger portion of applicants is rejected and worse still, they often pass into evaluation before getting rejected. This is especially acute at consultancies where there are very few ways to mechanically screen applicants for the necessary broad set of skills and attitude.
In the past I once calculated that each successful hire from an open application at a consultancy cost a total in the region of 5000 €. The cost is driven up due to two factors. Firstly the cost needs to account for all applicants who were screened and rejected, interviewed and rejected and those with whom no compromise could be reached at negotiation (not to speak of the occasional bad hire or quick leaver). Secondly in a consultancy interviewing costs do not need only to include wages, but to also account lost invoicing, increased travel and scheduling costs and other overheads (interviewers need to be away from paying customers and need to get there quickly after the interview is over, and for high consultant allocation rates this additionally means that interviews are done on overtime.)
So even a medium-sized consultancy getting “a few applications” each week has a very high incentive to screen aggressively and structure their interviewing to get rejections done as early as possible.
Do not think “continuous hiring” or “open applications” are any easier. You might get more personalised responses, but screening can still be brutal.
(Next one up is more concrete post on screening programmers.)