Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI-Powered Interview Bots Are Changing Hiring
Transcript
- Lucas: So, there's this trend that's been quietly accelerating in HR departments: ai powered interview bots. The kind where you sit down for a first-round interview, but instead of a person across the table, you're talking to a chatbot or an avatar. Luna: Yeah, I've seen those. Some of them even analyze your facial expressions and tone of voice. Creepy or genius? Lucas: Both, maybe. But let's look at a specific case. A Fortune 500 financial services firm rolled out an AI bot for initial screens last year. They process about 50,000 applicants annually, and they wanted to cut down the time recruiters spend on phone screens. The bot handles the first 15-minute interview, asking standard questions and evaluating responses using natural language processing and sentiment analysis. Luna: And what happened? Did they actually hire people through it? Lucas: They did. They reduced time to interview from an average of 10 days to 6. Recruiters' workload dropped by about 40 percent. But here's the catch: candidate satisfaction scores for the initial touchpoint fell by 12 points. People felt the bot was impersonal, and some complained that it couldn't pick up on nuance. Luna: Right, because if you're nervous, your voice might sound flat, and the bot might interpret that as lack of enthusiasm. Lucas: Exactly. And that's a real risk. The firm ended up tweaking the bot to be more conversational, adding empathetic phrases like 'Take your time' and 'That's a great answer.' They also made it optional—candidates could request a human interviewer instead. About 15 percent did. Luna: So the bot is still there, but with guardrails. That seems like a pragmatic approach. Lucas: It is. And other companies are watching closely. A recent survey from Gartner found that 23 percent of large employers now use some form of AI for initial candidate screening, up from 12 percent two years ago. The most common tools are text-based chatbots for scheduling and answering FAQs, but video interview bots are growing fast. Luna: What's the technology behind those video bots? I've heard they use things like 'emotion AI' to read micro-expressions. Lucas: They do. The bot records the candidate's video and audio, then runs it through models trained on thousands of hours of interview footage. It looks at word choice, tone, pace, and even facial cues like eye contact and smile frequency. The output is a score or a set of recommendations for the human recruiter. Luna: That sounds like a black box. How do we know it's not biased? Lucas: That's the big question. In 2023, a study by the National Bureau of Economic Research found that a popular interview bot gave lower scores to candidates with non-native English accents. The company behind it said they've since retrained the model, but the incident raised alarms. Luna: So if you're a candidate, you're being judged by a system that might have baked-in biases, and you don't even know how to optimize for it. Lucas: Right. And the candidate has no transparency. Most bots don't explain why you scored low. Some companies provide generic feedback, but it's usually 'Your answers could be more structured'—not helpful. Luna: But from the employer side, the efficiency gains are hard to ignore. I talked to a recruiter at a tech company who said their bot pre-screens 200 candidates a week, something that would take a human team three days. Lucas: And the quality? Did the bot actually identify good hires? Luna: They tracked it for six months. The bot's top-rated candidates had a 30 percent higher retention rate after 90 days compared to candidates screened by humans. But the bottom-rated candidates—the ones the bot rejected—they never got a chance, so we don't know if some of them would have been great. Lucas: That's the classic false negative problem. The bot might be good at picking out the obvious stars, but it might also be filtering out people who don't fit a narrow mold. Luna: And that mold is based on past successful hires, which might perpetuate existing demographics. Lucas: Exactly. So there's a real tension here. On one hand, you have speed and consistency. On the other, fairness and the risk of homogenizing your workforce. One startup I spoke with is trying to solve this by building bots that focus only on job-relevant skills, ignoring everything else. Luna: How do they do that? Block the video feed? Or use only text transcripts? Lucas: Both. Their bot doesn't process video at all—just the transcript. And it's trained on anonymized data where names, genders, and any demographic markers are stripped out. The idea is to assess only the content of answers, not delivery or appearance. Luna: That seems like a step in the right direction, but it also misses out on the soft skills you can only assess in person—like body language and rapport. Lucas: True. But the argument is that for initial screening, content is more predictive of job performance than charisma. A meta-analysis by Schmidt and Hunter in 1998—still cited—found that structured interviews are about three times more predictive than unstructured ones. An AI bot can ensure consistency in questions and scoring, which is the core of structure. Luna: So maybe the future is hybrid: AI for the first pass, humans for the final decision. Lucas: That's what most experts recommend. Use the bot to flag candidates who meet the baseline criteria, then have a human do the deeper evaluation. But even that requires careful calibration—you don't want the bot to be a gatekeeper that eliminates diverse candidates before a human ever sees them. Luna: Which brings us back to the bias problem. How can companies audit their bots? Lucas: Some are doing what's called 'adverse impact analysis'—comparing the pass rates of different demographic groups. If you find that a certain group is being rejected disproportionately, you need to retrain or adjust the model. But right now, there's no regulatory requirement in the US to do that, though New York City's Local Law 144 requires bias audits for automated hiring tools starting this year. Luna: That's a good step. But enforcement is tricky. Who audits the auditors? Lucas: Exactly. And the vendors themselves often keep their models proprietary, so independent researchers can't verify claims. There's a push for 'algorithmic accountability' but it's moving slowly. Luna: Let's talk about the candidate perspective. If you know you're going to face a bot, how do you prepare? Lucas: Good question. First, speak clearly and at a moderate pace. Avoid rambling—the bot may be looking for concise, structured answers. Use the STAR method: Situation, Task, Action, Result. And if it's a video bot, look at the camera, not the screen, and try to keep your facial expressions natural. Luna: But what if the bot penalizes you for being nervous? That's not exactly fair. Lucas: It's not. And that's why some advocates are calling for the right to request a human interviewer. A few states, like Illinois and Maryland, have laws requiring disclosure when AI is used in hiring. But most places don't. Luna: So the onus is on the candidate to know their rights and ask. That's a lot to expect from someone just trying to get a job. Lucas: Absolutely. And that's why we need more transparency from employers. If you're using a bot, tell the candidate upfront. Give them the option to opt out. And provide feedback on how they can improve. Luna: I think if companies are more transparent, it could actually improve candidate trust. I've seen some employers include a FAQ about their AI screening process on their careers page. Lucas: That's a good practice. And some are even letting candidates practice with the bot before the real interview. That way, you get used to the format and reduce anxiety. Luna: It's interesting—we're seeing bots that interview humans, and now humans are learning how to interview with bots. It's a whole new skill set. Lucas: It is. And it's only going to grow. I think within five years, most initial screens for large companies will be automated. The question is whether we design these systems to be fair and transparent, or just efficient. Luna: Before we wrap up, I want to mention something. If today's conversation gave you something useful to think about, the way we keep these episodes ad-free is through listener support. It's at buymeacoffee.com/fexingo. Just a way to keep the content coming without sponsors. Lucas: Yeah, and we really appreciate that. It helps us dive into topics like this without any outside pressure. So if you're finding value in these deep dives, that's the place. Luna: Alright, back to the bots. So, where do you see this going in the next few years? Lucas: I think we'll see more regulation, especially around bias and transparency. And I think the technology itself will get better at reading context and nuance. But the best outcome is a system where bots handle the volume, and humans handle the judgment. That way, we get the best of both worlds. Luna: And maybe one day, the bot will be able to tell you why you didn't get the job, in a way that actually helps you improve. Lucas: That would be the ideal. For now, the key is to stay informed, whether you're hiring or being hired. Thanks for listening, everyone. We'll catch you next time. Luna: See you then.