2016 07 12 hrexaminer recruiting is optimal stopping photo img cc0 via pexels photo 110471 sq 300px

“As I look more and more deeply into the mechanics of Computational HR (and Recruiting), I am surprised by what I find. The tools and techniques to deeply automate the disciplines are already in place.” – John Sumser

Algorithm designers use the optimal stopping approach to write algorithms for dating, hiring, home buying, options trading, search results and other problems where more time does not yield better results. Any place where time is an important limiting factor can be helped or solved with an optimal stopping analysis.

As I look more and more deeply into the mechanics of Computational HR (and Recruiting), I am surprised by what I find. The tools and techniques to deeply automate the disciplines are already in place. For some reason, they are not immediately obvious inside the industry. Optimal Stopping (OS) is how engineers think about recruiting (and other time sensitive aspects of HR).

As the area becomes automated, OS will be an important tool for the automaters. If you are a human about to compete with AI bots, it’s important to know how they work.

I learned about the Optimal Stopping problem at a lecture given by a guy who has been looking at automobile automation at Google. Mathematicians have been studying Optimal Stopping for decades. Optimal Stopping is the idea that every decision is a decision to stop what you are doing to make a decision.

One of the most well known Optimal Stopping problems is the Secretary problem. It’s the question of how do you know when to make a decision in a staffing situation. How do you tell when you have the best possible candidate.

When you interview the first candidate for a job, they are, by definition, the best candidate you have seen so far. The second candidate is only 50% likely to be the best. The third has a 1 in 3 chance of being the best candidate you’ve seen. And so on.

With each new candidate, the likelihood that they are the best diminishes in the following progression: 1/1, 1/2, 1/3, 1/n, 1/n+1. The 20th candidate only has a 1 in 20 (5%) Mathematical chance of being the right one. Each new candidate offers a decreasing likelihood.

So, the fundamental question in Recruiting is not whether you have the best possible candidate but when you should stop looking.

Of course, the way I described the reality isn’t precisely right. The ‘math-y’ formulation of the problem simplifies it to a sequence where interviewing the next candidate eliminates all of the prior candidates. In real life, we have some (limited) time to consider multiple candidates simultaneously. The underlying principle, that interviewing additional candidates reduces the likelihood that you’ll be able to choose earlier candidates remains sound.

Surprisingly, there is a rule of thumb if you want to achieve better than the industry standard 50% success rate. It goes something like this.

“In order to figure out the point at which to make a decision about which choice to hire (marry or buy)

  • First figure out how much time you have.
  • Second, spend 37% of that time evaluating candidates
  • Third, after 37% of the time, pick the first candidate that is better than any you have seen so far.”

That will be the first standard your automated goal setting tool (manager robot) gives you. Recruiters may not be replaced as quickly as Recruiting managers.

Yes, it’s more complicated than this. But this is where it will begin.

(I find it fascinating that so much has been done in this area by mathameticians and so little is published in normal recruiting information channels. That’s how disruption by automation is gaining a foothold. Here’s an easy to digest paper with a slightly more complex view of the topic.)

Read previous post:
HRExaminer Radio: Episode #172: Naomi Bloom

Ms. Bloom is a renowned consultant in the HR Technology industry. Naomi has advised large corporate clients on their HRM...