If you live in the US and you're applying for work with one of the more innovative companies, chances are you'll encounter at least one of these two interview methods.
Even though worldwide unemployment is at its lowest level of the decade, the pace in which we exchange jobs is increasing every year. Companies are hiring like never before, and people are sending out applications in vast numbers. As a result, the pressure on recruiters becomes twofold. On the one hand, they're expected to source valuable talent for hard-to-fill-positions. On the other, they have to sift through endless heaps of applications for low to medium qualified openings. When every position attracts the attention of hundreds, or even thousands, of hungry candidates, there’s no realistic way for the recruiter to look through every single application in a fair and objective way.
With AI, the initial screening steps can be automated to a large extent. By taking into account a number of factors with a high impact on the hire, this tech can help recruiters everywhere to streamline the process of acquiring talent.
A challenge in modern hiring is that a large portion of the applicants are either completely unqualified, have unreasonable demands to make a job switch, or simply haven’t given their application enough thought. An AI can get down to details as small as calculating commuting times for each candidate, and weigh this factor into the ranking in a matter of microseconds. At scale, the potential for time savings in any given HR department becomes unfathomable.
When it comes to recruiting and screening, AI system service providers take very different approaches to the problems at hand. In this post, we're going to focus on AI interview screening, as well as forecasting future performance of job candidates as challenges that AI solutions can solve.
The ability to predict future performance of a candidate is a task an AI could be trained to do in a data-driven way. For the human recruiter, this approach would be incredibly time-demanding simply because humans cannot process huge amounts of information in the same way a sophisticated piece of software can. Adding to this, an AI can also feed back the end result (actually tracking whether the chosen candidate is a good fit or not), and so facilitate a complete loop and an ever-improving recruitment system.
Exactly how many factors and signals a human recruiter takes into the overall judgment of a candidate is hard to say, but according to various AI assessment vendors on the market, AI systems take in signals to a much greater extent.
AI assessments are hosted by a video format in which each candidate is asked to answer a number of preset questions. As the applicant answers, the AI looks at factors like facial expression and language proficiency. Combined with the actual content from the interview, it's then able to create a reasonable forecast on future performance for each interviewee.
The main purpose of AI assessment is to save the employer time in the interview process without losing too much quality. Then again, it also makes life a lot more comfortable for the candidate. The applicant can choose to have the interview wherever they like, whether that's at home or somewhere else. They don't have to rush to numerous locations around town in a single day - and they can wear what they like for the interview. The only thing they need is a computer or cell phone with a camera function.
From an internal recruiting point of view, AI assessments can also be helpful. When attractive employers announce open positions, the sheer volume of applications is often enormous - and filled with ineligible inquiries. AI assessment filters the applications and uses machine learning to help recruiters with the initial screening - before an actual human-to-human interview takes place.
AI assessment is primarily used to differentiate candidates in terms of behavior and team fit.
AI assessment questions are based on traditional job interviews and so will vary from company to company. The common denominator is that the questions primarily focus on soft values such as how you would act in a given situation, how your values compare to the company culture, and how you'd complete the existing team being hired for.
When a question is asked, the first 20–30 seconds don't record and are meant as a chance for you to think through your answer. After that, you're given a slot to give your final response. The length can vary depending on the question, but is usually somewhere between 30 seconds and a couple of minutes.
Here’s a list of the most commonly used questions from the vendor HireVue.
Candidates should be sure to practice a confident and well-formulated response to these questions. And remember, all applicants will be asked the exact same questions - so try to use a response that will separate you from the crowd.
According to the vendors themselves, they use two main AI-based technologies:
In the case of artificial intelligence for facial expressions (which, according to HireVue, is not facial recognition), the idea is to identify emotions and to predict future behavior (and before you ask: yes, it's been scientifically proven to work). HireVue calls them Facial Action Units and says they represent up to 29% of the candidate’s score. Content, language and vocal features (like tone and pace) make up the rest.
In expression and language analysis, the system has been trained to look for specific parlances. These are factors like active vs. passive wording, technical language, long sentences, vocabulary, and several other parameters. Depending on the job that's being hired for, different aspects are benchmarked, and new candidates are compared to existing data which has been compiled from existing employees.
A report is then automatically generated, showcasing the candidate’s traits and competencies such as “willingness to learn,” “conscientiousness & responsibility” and “personal stability” from the candidate. Later, candidates are sorted into categories of high, medium or low probability of being a high achiever. Finally, it's up to the hiring company to use the AI-generated results for their continued hiring process. There's nothing stopping them from pursuing a candidate in the lower tier, but of course, people in the high group are most often favored.
It's difficult to pinpoint exactly what makes a perfect candidate for a job - sometimes even for AI system designers themselves. As with all artificial intelligence systems using neural networks, there’s no actual algorithm deciding what's good and what's not. Neural networks simply try to reproduce the same results as the defined goal on a specific set of training data that it has been trained on. This also means that if none of your employees are high achievers, the chance of this system finding one for you is very close to zero.
In other words: neural network systems never become better than the training data you put into it.
When a company agrees to use AI assessments, the system is trained on the existing employees in your team. That includes everyone on the performance spectrum, from low to high achievers. All employees are benchmarked on a scale and asked to complete the same AI assessment.
A model is then constructed from the profile of the best employees in the firm and the AI looks for the same types of behavior, competencies, and attributes in the interviews. The more employees in the organization the better; it means more data to train the AI model on.
But as we’ve all been taught, correlation does not always imply causation. There is a chance that AI’s will find patterns that make no sense in the real world, but have a strong correlation in a data set.
There is no shortage of examples of what the outcome can be from failed attempts of using AI for recruiting.
If you're thinking 'Hey, should an AI decide who gets a job or not? Isn’t this pretty serious stuff?' well, you're not alone. In fact, policymakers in both the EU and the US have started constructing regulations on how AI should be used used in the recruitment sector. And it’s not easy being an AI assessment vendor these days.
Recently in Brussels, big-shot tech leaders from San Fransisco queued around the block in an attempt to lobby against the newly presented regulations. But according to Thierry Breton, European Commissioner for Internal Market and Services, the attempts have been more or less futile. The existing platforms will have to adapt to the policies in Europe - not the other way around.
We still don't know the details of how these regulations will impact existing AI recruiting solutions. In spite of this, the consensus surrounding HR and artificial intelligence is clear: AI tech will, to a greater and greater extent, determine who gets a job and who doesn’t. Colleges around the US are beginning to adapt to this reality, and are now preparing their students for AI interviews. The only problem? Very little is known about how the systems actually work.
Some counselors suggest using industry-specific keywords - while others make recommendations regarding AI-approved clothing. A new tool called Symplicity was developed to help prepare students on how to behave in tech-supported interviews. The software allows for candidates to record their answers, review the footage, and improve.
AI assessment vendor HireVue gives registered users access to resources meant to prepare them for coming interviews. Try recording yourself when responding to a couple of different questions to see what kind of a first impression you make. Once you know how you do on the first try, you can work on polishing your responses.
The best-known service provider in AI assessments today is HireVue, which has revolutionized the way employers conduct and prepare for interviews. HireVue doesn't disclose a full list of their clients, but they do showcase a few of them.
Hilton Hotels has been a frequent user for a long time, and say they have automated interviews to thank for decreasing their hiring process for reservation-booking, revenue management and call center positions from 6 weeks to just 5 days.
In a recent interview with The Washington Post, the global head of recruiting at Hilton Hotels, Sarah Smart, says that the system has allowed them to evaluate candidates at a far higher pace than before. She also mentions that it has become rare for recruiters to go beyond the recommendations of the system, thereby drastically reducing the workload when screening.
Another HireVue client, Unilever, also praises the system and claims that they've saved around 1 million dollars in recruiting costs as well as 100.000 hours in interviewing time by using the platform. Leena Nair, CHRO at Unilever, compliments AI assessment on helping them succeed with workforce diversification. She estimates that they've increased diversification throughout the company by approximately 16 % with HireVue.
Another smaller, yet very loyal, follower of HireVue’s practices, Re:work, a non-profit organization helping people get into the tech sector, have found relief through the app. At Re:work many course participants could not handle the intense 8-week program leading to burnout in several cases.
Re:work reports that with HireVue it became possible to forecast which candidates were more likely to succeed after the program, based on their problem-solving skills and ability to negotiate. As such, Re:work was able to focus all their attention on recruiting candidates with the right skillset and avoid leading people into exhaustion.
In the case of HireVue, there is some confusion surrounding which of their customers actually use the software's AI component. This is based on the fact that the AI assessments are an add-on service to the video interview that sits at the core of HireVue's business.
Goldman Sachs have previously been criticized for being associated with HireVue's AI interviews. Recently, however, The Washington Post announced that "due to incorrect information from Goldman Sachs", the accusations are inaccurate; the banking group only uses the video interview platform.
If and when you are faced with attending a HireVue interview, be sure to investigate if the AI service is active or not, as this will have an impact on your results.
A common recommendation, according to Michael Kalish who is the associate director at Baruch’s on-campus recruiting department, is to dress in a full suit.
Pants are optional as long as you aim the camera above the belt.
Firstly, it's important to be aware of the ongoing war between proponents and opponents of AI recruitment systems on the topic of bias.
On the one hand, proponents and vendors argue for how their systems actually reduce bias and increase diversity in recruiting. Opponents, on the other hand, are outraged by the fact that there's no traceability or transparency surrounding the functionality of many of these systems.
As more and more AI-driven tech makes its way into our daily lives, the attitude towards system accuracy needs to change. Up until now, most traditional systems have had a very predictable accuracy level that won't change with every new training data input.
AI systems don’t function like that.
A model based on neural networks is comparable to a black box, where the computer is provided with a set of training data and a desired outcome. The patterns the system decides is most significant to achieving this outcome are not given, and may be totally insignificant when applied to another data set.
Unless you have an infinite amount of training data (which no one does) occasional errors may happen. In other words: The computer might find a pattern that's reliable in one case, but which may still be flawed in a different context.
When AI is applied to make an actual decision that has the potential to impact people's lives, it's always met with strong skepticism and rigorous evaluations. It's an interesting paradox. Human recruiters are human, and so inevitably, they will make mistakes. Those same faulty decisions from an AI system, though, are still very controversial.
Recently, Tesla’s autopilot system faced this exact problem, when a pedestrian was hit and killed while the system was engaged. Although human drivers hit and kill thousands of people each year, the news is received in an entirely different light when a computer causes the fatality. No matter how you look at it, it's always considered worse when a program, rather than a human, makes a mistake.
In the context of human resources, candidates are rejected daily on the basis of ridiculous reasons, outspoken or subconscious. It could be an ill-fitting shirt, a weird haircut, a Texas accent or just looks. According to service providers, their AI model sees past these factors, and solely focuses on the traits that are proven to have an effect on future work performance. As such, it effectively reduces bias.
Still, these systems need to be built by a human. And if you're not careful, the inherent bias of the designer may be reflected or even amplified in the platform. In addition to this, as we mentioned previously, the system is trained on existing employees', and so it's likely that an AI system will find candidates that are closely matched with them.
According to the vendors: No. But as we've already established, an AI recruitment system is not like a regular computer program. It's designed to remove bias, yes. But then again, it's also based on neural networks, and so has very sophisticated in-built analysis functionality.
Service providers say you can't fake your way through an AI-powered interview, because it can pick up on things like too much eye contact, being overly friendly or smiling too much. Going in, then, it seems that the best way to 'hack' the call is by approaching it the same way you would in a human-to-human interview: by giving your best answers. The perk is that you can also feel confident that your interviewer won't judge you on the basis of looks, clothing, hair style or race - and instead will focus solely on the content of the answers you provide.
In many cases - yes it is. Imagine how often candidates are dismissed based on bad chemistry with the recruiter and his or her personal references. An untucked shirt, chewing gum, typing errors in the resume, ugly shoes.
If the AI that is scoring candidates can be described as a black box, then the recruiter's thinking is often the ultimate black box. For the candidates, it is, however, hard to improve your performance until the next interview if you didn’t get the job. Hirevue, for instance, does not explain their decisions and gives no feedback to the candidate. Not even their final score or ranking.
Vendors would probably do well to focus on the candidates, too, and make sure an AI assessment benefits both the company hiring and the people applying.
AI and machine learning are only as good as the training data. In other words: interviews conducted by an AI are only as good, unbiased, effective and meaningful as the training data the system is based on. Even carefully trained AI systems can potentially lead to unwanted outcomes.
Having all this said, there's no doubt we're headed in a direction where AI assessments will become a standard procedure. Who knows, in a not so distant future, maybe you will face an AI recruiter.