Why deep learning is the next frontier in talent acquisition

By Alexandra Levit

In current hiring processes, we rely heavily on CVs. However, they are error prone because humans are inherently bad at self assessment. And when candidates tailor their CVs to match job descriptions, they are full of platitudes, exaggerations and omissions as candidates distil years of work down to a few bullet points.

The overall lack of descriptive, meaningful data means employers can be negatively influenced by details with little predictive insight into a candidate’s ability to excel in the job. For instance, typos, font choice and layout may influence the employer’s perception of the candidate’s capabilities.

Historically, workforce data has also been reactive rather than proactive. Companies struggle to determine how many permanent and contingent employees they have, let alone the skills, capabilities and potential of their workforce.

With nearly two million unique skills globally, it is impossible for a human to understand all of them individually, the relationship between them, and the interactive role brand new skillsets will play. A human manager may review the narrow data at their disposal, and then supplement it with their own perceptions and understanding of the underlying skills required to perform in a job. For example, when a company is hiring for a customer service manager, a recruiter may search for candidates who have held that exact title or a related title. We simply have not had the tools to surface candidates who may have the right customer skills based on their work experience.

A customer service representative at a department store may have a skillset that’s more similar to a real estate agent than a customer service representative at a bank. A secondary school teacher may have more of the skills required to be a fantastic product trainer at a technology company than a product operations associate already in the tech industry.

Could deep learning be the answer?

Deep learning is a type of machine learning inspired by the human brain. Deep learning algorithms mimic human conclusions by continuously analysing data with a given logical structure. This structure encompasses multiple layers of algorithms called neural networks.

The main challenge of deep learning is data – specifically, the massive amount required. The talent space alone needs billions of data points about people, career trajectories, skills and experiences. 

But while AI’s reach was once limited by computing power and the availability of data, that’s no longer the case. Today, global neural nets can identify more than two million skills across the world’s 7.8 billion people. Then, using this data, AI engineers can develop deep learning algorithms to determine the best answers to a defined class of questions. In the case of talent, such questions might include: who is the best fit for this specific job requirement? What job is this individual most likely to hold next in their career?

Understanding deep learning’s ability to harness a massive amount of data is only the first step, however. The more important question is how deep learning can be applied to talent acquisition.

The power of skills adjacency

Skills adjacency refers to the inference that a person good at skill A often excels at skill B. For example, we can infer that someone who excels in calculus likely also excels at algebra. When we can understand skills adjacency fully, it helps us complete our picture of a human’s true potential and the transferable skills they possess.

Deep learning can break down human work experience into capabilities and match those capabilities with jobs available now. For example, many restaurant workers have found work in call centres, and a major bank told us that some new employees were formerly bartenders. These aren’t associations one would readily make.

Similarly, if a company is looking for project management skills, CVs and job descriptions often fail to surface the right matches. But AI can tell us that a candidate or employee’s skills are validated or likely based on their work experience, even if the person never had a title of ‘project manager’ or listed ‘project management’ as a skill on their application. On the flip side, AI can determine if a candidate for a project management job is missing the required skills despite a prior project management title.

In other words, deep learning platforms call out potential by looking at people with similar skills and what they were able to accomplish next in their career. Thanks to a global data set in which we can analyse thousands of people who acquired new skills and were subsequently successful in a new role, we can infer that a fresh candidate can also learn these skills and do well in this new role.

Imagine that three people have worked as product managers. One of them worked at Google, one worked at American conglomerate Honeywell and one at Delta Air Lines. An AI-driven system knows, with a high degree of certainty, that the person from Honeywell has skills that the Delta and Google people do not. It can also tell whether the Delta and Google product managers could easily learn the skills the Honeywell person has, and how they could go about doing so.

The use of deep learning means this type of trust isn’t based on blind faith, but rather on a comprehensive analysis, and it takes both the art and science of talent acquisition to a whole new level.