Is the machine superior to human instincts when it comes to making hiring decisions? According to an article in Bloomberg Business it is: researchers of NBER performed a study across 15 companies with more than 300.000 hires in low-skilled service jobs such as call center or data entry employees. They implemented an algorithm that gave a recommendation who to hire based on technical skills, personality, cognitive skills and fit for the job.
Their conclusion: positive recommended employees stayed on average 12 days longer than neutral recommended employees, which on their turn stayed on average 17 days longer than negative recommended employees. Although this might not sound like much of a deal, in fact it is. The median duration of employees in these jobs is around three months, hence the positive recommended employees remained more than 20% longer in their jobs than the negative recommended employees.
Trust algorithms or gut feelings?
Although the article notes that using algorithms in the talent acquisition process is becoming more popular, still it is in the nature of a recruiter to follow his/her gut feelings. As this example shows, this could lead to significant additional costs. So should we let go of our gut feelings and fully trust what the algorithm is telling us if the empirical results of the algorithm are superior to human insticts? Well, yes and no. Yes, we should be aware of our human biases when making hiring decisions and let the algorithm help to decrease this bias, but also know when it is appropriate to overwrite the decision of the algorithm.
A Google-example: when ethics wins
How this works in a practical situation is illustrated in a TED talk by Andreas Ekström. In this presentation Andreas describes how a Google image search result of Michelle Obama was removed by Google manually. The reason for this manual interference was that in this search result her face was replaced by the face of a monkey. On the other hand, a search result of Anders Behring Breivik (the terrorist behind the attack on Utaya Island) where his photo was replaced by dog poo, was not removed. So why did Google remove the result in one situation and not the other? Simply because of ethical reasons that the algorithm does not take into account.
Man & machine!
Hence it should not be man versus machine, but rather man + machine. The machine can help us reduce our human bias and automate tedious jobs, where HR departments can draw the line which moral standards they value within their company. If the decision of the algorithm is incompatible with the company’s moral standards the algorithm’s outcome should be overwritten.
So, every algorithm has a human part, even Google’s search algorithm. This is not different within the recruitment process: it is the organization that eventually determines which characteristics of an applicant are valued most. However, an algorithm, or analytics in general, can give a crucial insight into the talent acquisition procedure that can help organizations to define which characteristics they value most and to make sure that this is used in the daily practise of recruiters. This can be performed by for example:
- Testing of assumptions: which applicants have most potential? What is the best way to get in contact with your target audience? Most organizations are currently answering these questions based on intuition, and as the NBER study shows this is not always the most effective way.
- Homogeneous decision making: do you make the same decisions during a job interview on Monday morning as on Friday afternoon? Even if you do, the way you make these two decisions can be different. An algorithm does not have this problem, it will give the same advice in both situations. Therefore this can help to make the decision making more homogeneous.
- (Re)defining your organization’s core values: it would be too easy to instruct the algorithm with “find me the best people”. To make the algorithm effective there must be a clear and detailed definition of ‘the best people’. Specifying this definition on a detailed level can already provide much insight into which assumptions are used within the organization and how opinions differ across your employees.
Are you capable to define your ideal applicant on a detailed level? Defining the required knowledge is easy but how about more difficult characteristics such as willpower or creativity? How are you currently estimating these characteristics during a job interview? Write this down and ask yourself: if an algorithm would prove me wrong on the validity of my estimates, would I hang on to it? And if not, then why not use analytics to test whether your assumptions hold in practice?