Hiring managers: We have seen the future and it is hybrid.
A study by Ishita Chakraborty of the Wisconsin School of Business suggests that AI-human collaboration during the recruitment process may save managerial time—and still be cost-effective—by allowing AI to weed out the bottom candidates from the pool, leaving managers to focus on the top tier requiring human judgment.
Set in the context of salesforce hiring using interview videos, the study is the first known work of its kind to use video data to better understand sales recruitment as well as the audio-visual cues and linguistic features that can make a sales candidate successful.
“This space is very exciting overall, because I feel that the future of hiring, or AI-based hiring, is actually hybrid,” says Chakraborty, the Thomas and Charlene Landsberg Smith Faculty Fellow and an assistant professor of marketing. As the technological landscape changes over the next five to 10 years toward being fully AI, trust must evolve along with it, she believes.
“People have to be able to trust these models enough to say, ‘Hey, I don’t want to be part of this hiring process at all. Let the machine take care of it.’ Until we reach that point, the role of combining human and AI judgment is critical.”
Examining interactive videos
The study used 195 videos of interactive sales-pitch roleplays between a student “seller” and a “corporate buyer.” The buyers were played by actual company representatives (such as managers, VPs, CEOs) and the filmed interaction was ranked by real-world industry judges as part of an internship drive for students, resulting in more than 1,752 evaluations. From the resulting interview dataset, which included video, transcript text, and judges’ scoring matrices, two models were created—a fully AI model and a hybrid AI-human model.
The AI model was designed to predict candidate interview performance: What aspects of a candidate’s interview, including voice data, text, and body language, are the makings of a successful salesperson? The data suggested that early screening decisions were influenced by factors such as “energetic voice” and “open posture,” while the final round of selection decisions tended to hinge upon conversational interactions, hand gestures, and qualitative linguistic features.
“I’m personally very interested in sales negotiation, and I’ve often wondered what really works in sales and what does not. In my own experience, I saw that there are different types of salespeople, and they succeed in different ways.”
—Ishita Chakraborty
The hybrid AI-human model used the AI predictions while adding in human contributions to the process. The findings suggested that this model increased accuracy: having humans in the loop made for more precise judgements and mitigated against the false positives of the less-qualified candidates advancing to the next round while the better candidates slipped through the cracks. However, including human labor made it more expensive than just AI alone.
Based on the results of these two models, the study findings pointed to utilizing AI in the initial stages of salesforce interviews to remove the least-qualified candidates while reserving the final selection round of candidates for sales managers to determine, since it requires subjective and careful human judgment.
It’s an efficient solution that lowers stress for managers “who don’t want to spend more time with the bad candidates,” says Chakraborty, while also being more cost-effective than solely human labor and more accurate than AI alone would be.
“The gap we found in the literature is that we have not really experimented and validated the extent of human intervention that is needed, and at what stage it’s needed. That’s how we started thinking of a hybrid model. Given that you have these AI tools, do you still benefit from, for example, having a human judge?”
Chakraborty says the study aligns with her research around unstructured data (such as texts, audio, and social media) and also intrigued her because her first job was as a sales management trainee.
“I’m personally very interested in sales negotiation, and I’ve often wondered what really works in sales and what does not,” she says. “In my own experience, I saw that there are different types of salespeople, and they succeed in different ways. Not every extrovert is a great salesperson; it could be someone who is really calm, a good listener versus someone who throws their weight around.”
The timestamped video data helps show what styles might work and in what stages of the interview process, she says. “Our goal was to understand the sales pitching process in a more scientific way, a more quantitative way. We wanted to really understand what micro success factors matter.”
Planning for the future
While Chakraborty is hesitant to extrapolate the study’s results to other fields, she notes that the hybrid approach might work particularly well in specialized occupations where the qualified candidate pool tends to be small to begin with and can’t afford any false negatives, such as academic faculty hires. She would also be eager to see how hybrid models might be applied to online teaching, a “personal favorite,” with considerable potential impact.
Data privacy is another key consideration. The hiring process ranks right up there with healthcare in terms of handling sensitive information, Chakraborty says, and states are moving to regulate data privacy laws, with Europe being even more stringent. Companies should minimize the volume of personal information they keep for liability reasons, and with video capturing the largest amount of sensitive data, it’s a medium to use sparingly. More work needs to be done, Chakraborty says, but her study suggests that interview text alone may work well in lieu of video and carries less of a privacy risk.
“That’s why I said the future is going to be hybrid; it’s really in our best interest to understand that. How much human intervention do we need and how can we best combine the two, AI and human contributions? Because for right now, it’s unlikely that this will become totally automated technology.”
Read the new working paper: Chakraborty, I. & Chiong, K. & Dover, H. & Sudhir, K. “AI and AI-Human Based Salesforce Hiring Using Interview Videos.”
Ishita Chakraborty is the Thomas and Charlene Landsberg Smith Faculty Fellow and an assistant professor in the Department of Marketing at the Wisconsin School of Business.