Latest / The Diversity Career Podcast with Fexingo: Inclusion, Equity, and Underrepresented Professionals / How One Company Uses AI to Reduce Bias in Hiring
Transcript
- Lucas: So there's this company called PivotPath — about two thousand employees, based in Austin — and they rolled out an ai driven hiring platform about two years ago. The idea was to strip out any demographic identifiers from resumes: names, schools, even zip codes. Then the algorithm scores candidates purely on skills and experience. Luna: I've heard of this approach — it's basically blind hiring on steroids. But did it actually move the needle? Lucas: According to their internal data, yes. Within eighteen months, the share of candidates from underrepresented groups who made it to final-round interviews went up by 40 percent. And offers extended to those groups increased by about 30 percent. Luna: That's a pretty dramatic shift. But I'm guessing it wasn't all smooth sailing? Lucas: Not at all. The biggest pushback came from hiring managers who felt the algorithm was rejecting candidates they considered a strong 'cultural fit.' One engineering manager told HR that the AI passed over someone who had worked at a competitor and seemed like a perfect match — but the algorithm scored them lower because their resume didn't have specific keywords. Luna: Right — so the algorithm might miss things that a human recruiter would pick up on. But isn't that kind of the point? To remove human bias that often favors people who look and act like the hiring manager? Lucas: Exactly. And PivotPath's head of diversity told me that the phrase 'cultural fit' has historically been used to justify hiring people who fit an existing mold — which often means less diversity. So they pushed back and said, 'Let's define cultural fit more concretely, as values and work style, not shared hobbies or background.' Luna: So they retrained the algorithm to incorporate those concrete definitions? Lucas: They did. They worked with the vendor to adjust the model so that certain soft skills — like collaboration or adaptability — were weighted in, but only if they were backed up by specific examples in the resume. They didn't just let managers override the algorithm willy-nilly. Luna: I wonder about the training data though. If the AI was trained on historical hiring data from PivotPath, it might have learned the same biases that existed before. Did they address that? Lucas: Great question. They actually used a synthetic data set created by the vendor — designed to be bias-free — plus they did regular audits to check for disparate impact. For instance, they looked at whether the algorithm was more likely to flag certain names or phrases associated with particular groups. They found one early version was penalizing candidates who listed 'mentoring' as an activity because it was more common on resumes from women and people of color. They fixed that. Luna: Wow, so even subtle patterns like that get baked in. That's a good reminder that AI isn't a magic bullet — it requires constant oversight. Lucas: Absolutely. The whole process is iterative. They have a weekly meeting where the HR analytics team reviews any flags from the algorithm and discusses adjustments. It's not set-and-forget. Luna: So what's the bottom line for listeners who might be considering a similar tool? Is it worth it? Lucas: I think the key takeaway is that AI can be a powerful tool for reducing bias, but only if you're transparent about how it works and you're willing to invest in ongoing monitoring. PivotPath saw real gains, but it took a lot of internal education — helping managers understand that a lower score from the algorithm doesn't mean a bad candidate, it just means the resume didn't match the objective criteria. Luna: And that takes trust. Which is hard to build when people feel like a machine is making decisions about their team. Lucas: Exactly. That's why they didn't just hand the final decision to the AI. The algorithm shortlisted candidates, but human recruiters still did the phone screens and interviews. The difference was that the pool they started from was more diverse. Luna: That seems like a smart balance. Use the tech to widen the funnel, but keep humans in the loop for the nuanced evaluation. Lucas: Right. And I think that's a model a lot of companies could emulate. If you're thinking about implementing something similar, start with a pilot in one department, measure the outcomes, and be prepared to iterate based on feedback. Luna: Speaking of feedback, we always love hearing from our listeners. If you've tried an AI hiring tool or have thoughts on this, let us know. And if you find value in these conversations, consider supporting the show at buy me a coffee dot com slash fexingo. It helps us keep the podcast ad-free and focused on real stories. Lucas: Yeah, that's a big help. So to wrap up — PivotPath's experience shows that AI can be part of the solution to hiring bias, but it's not a silver bullet. It takes commitment, transparency, and a willingness to challenge your own assumptions. Luna: And maybe most importantly, it requires defining what you really mean by 'the best candidate' — because if you don't, the algorithm will make that decision for you, with all the hidden biases that entails. Lucas: Well said. Thanks for listening, everyone. We'll catch you next time.