Despite the desire for equitable hiring practices, humans have biases in every decision they make, and most of the time they don’t realize it. Often, hiring managers recommend a candidate because of a personal connection such as alumni groups or an interest in sports, or qualified candidates are not considered because of a gap in employment.

A diverse workforce is proven to increase an organization’s revenue, but the key challenge is building and retaining that diverse and competitive workforce. It’s met with personal bias from an archaic, bias-prone process that relies on resumes. This is one of the most pressing challenges organizations face, because without the right talent, they cease to compete with their products or services.

A complicating factor is the lack of quality data to make decisions. The sources of data that we use for hiring are diverse, including resumes, interviews, and assessments. Now add to this the engagement data used to track employee performance, potential, and flight risk. These complex datasets are difficult on their own, not to mention problems with data consistency. Most humans, unable to sufficiently comprehend all of these sources, rely on their gut instinct, which is unconscious bias incarnate.

We need data diversity, data integrity, and the ability to consistently process that data at scale. With standard data, one person might base a judgement on 100 hiring outcomes, but a computer can process 100,000+ interviews and hiring outcomes.

One of the main objectives of a phone screen or in-person interview is to evaluate soft competencies. These might include communication, friendliness, empathy, teamwork, motivation, and engagement, or any trait that is difficult to determine from a resume.

In a traditional interview, all of that rich, interactive data is lost once the interview is over. What’s left is a rough yes/no score and someone’s foggy recollection of why they did or didn’t think the candidate was going to be the best fit. In light of the growing shift toward digital interviewing, that data is no longer lost — it’s captured and can be analyzed. We are sitting on a data goldmine that previously didn’t exist.

AI identifies qualified candidates better than humans

HireVue developed our interview model to extract raw audio features, text from speech, and micro-expressions. Then, by including all of these features into a holistic model, we found that not only was the model predictive on new data using cross-validation, but that it was competitive with assessment alternatives. Some of the more remarkable test cases have included predicting post-hire safety infractions from a three-question interview for a transportation company, and reducing the time to hire from six weeks to five days for a global hospitality company.

The standard measure of fit/accuracy in Industrial/Organizational Psychology (IO) in talent acquisition is the R value. The deep machine learning based models are producing R values much higher than the human hiring control groups.

Another side effect of doing interview analytics is the discovery that question quality can finally have an objective measure. Most companies, large and small, are insecure in the questions that they ask during the interview process. These questions are crucial for the interview modeling to work well, and the data science team discovered that questions with dramatically different responses between low and top performers are the most predictive. With the feature set that is already harvested from the interview, that difference can be measured objectively. Using this type of interview modeling helps employers focus on the questions that really matter, and avoid bias.

How AI could actually improve employment

The goalpost for jobs a human needs to do vs. AI keeps moving. What we thought had to be done by humans is being challenged each year. Surely a human has to drive a car, screen job applicants, or interpret medical images, or be competitive at Jeopardy … you get the idea.

So, how might you look for opportunities to explore impactful machine learning and AI solutions in your business?

  • Look for the weak links in your processes that are manual and time-consuming.
  • Where is the human inefficiency in your processes? What positions in a process are always causing bottlenecks?
  • What processes do you have that have never had meaningful data capture? Brainstorm ways to collect that data.

The use case for AI in hiring is already having an impact at many organizations because getting the best candidates onboard faster saves real money and provides competitive advantage. It’s also having a positive impact on job seekers because they are getting the same shot regardless of gender, ethnicity, age, employment gaps, or college attended. As this practice evolves, humans may just have machines to thank for jobs.