A new study from HiringBranch challenges the conventional approach of scoring soft skills in isolation during pre-hire assessments, suggesting that this practice undermines the ability to predict on-the-job performance for frontline roles. The research, discussed in a recent episode of the podcast 'You Should Know,' indicates that evaluating empathy, acknowledgment, active listening, and reassurance as separate metrics yields only moderate correlation with human judgment, whereas a combined proprietary model shows much stronger predictive power.
Assaf Bar-Moshe, Chief of Research and Development Officer at HiringBranch, explained the implications during the podcast. 'If a candidate can express empathy, but is unable to solve the issue correctly or to comprehend the issue correctly or to reassure the customer, then this empathy is nice, but it's actually useless,' he said. The company's approach diverges from personality-based assessments, instead using a linguistic analysis of candidates' actual words in open-ended voice and writing scenarios. This method, described as a 'sociopragmatic analysis,' allows the assessment to mirror the integrated nature of skills required in customer-facing interactions.
The study, which will be published under the AI research tab on the HiringBranch website, is grounded in real-world customer service scenarios. For example, a retail confrontation over a mispriced item may require diplomacy more than strict adherence to policy. Bar-Moshe emphasized that the assessment translates job descriptions into conversation flows and scenario-based tasks calibrated per client, region, and role. This calibration accounts for regional variations, such as differences in scoring weights for the same role in Vancouver, Toronto, and Montreal.
The research team, comprising IO psychologists and linguists, uses years of textual data to build machine learning models that predict skills like empathy and acknowledgment. These predictions are then validated against actual on-the-job performance months after hire, ensuring the model's relevance. The findings suggest that when skills are assessed in combination, the predictive validity strengthens significantly, offering a more accurate picture of a candidate's potential.
Bar-Moshe also hinted at a forthcoming self-serve capability that would allow hiring managers to build assessments from a library of conversation flows and skills, reducing reliance on weak or generic job descriptions. This development could democratize access to sophisticated assessment tools, enabling smaller companies to benefit from the same predictive power.
The episode, hosted by William Tincup and co-hosted by Ryan Leary, is part of the WRKdefined Podcast Network based in Arlington, Texas, and reaches over 3.9 million verified listeners monthly. The full episode is available on the You Should Know Podcast page and wherever podcasts are heard.
