Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Reshaping Employee Performance Reviews
Transcript
- Lucas: If you've had a performance review in the last few years, you know the drill: you sit down with your manager once a year, spend twenty minutes trying to justify your existence, and then get a rating that feels two years old. Luna: And if you're lucky, you get a raise that barely covers inflation. It's almost a cliché at this point. Lucas: Right. But there's a growing movement to kill that annual ritual entirely. A mid-sized tech company called Vantage — about two thousand employees — replaced their annual reviews last year with an ai driven continuous feedback system. And the results are pretty striking. Luna: Continuous feedback — so instead of one big conversation, you're getting input throughout the year. How does the AI actually work in practice? Lucas: So Vantage uses a platform that integrates with Slack, email, and project management tools. The AI — it's essentially a natural language processing engine — scans work-related communications for patterns. Things like how often someone's contributions are praised, whether they're frequently mentioned in problem-solving threads, or if their tone in messages shifts toward frustration. Luna: Wait — the AI is reading our Slack messages? That sounds like a privacy nightmare. Lucas: It's opt-in at Vantage. Employees have to consent, and they can see exactly what data the AI is pulling. But yes, that's the trade-off. The company argues that the AI catches signals a manager might miss — like a quiet employee who consistently writes great code but never speaks up in meetings. Luna: I get the intent. But I wonder how accurate that is. Language is so nuanced. Sarcasm, context, inside jokes — can an NLP model really parse performance from a chat message? Lucas: It's imperfect. Vantage's head of people told me they have a human-in-the-loop for edge cases. The AI flags patterns, but a manager still reviews the output. They found that the AI identified about eighty percent of high performers correctly — but it also mislabeled about ten percent of people as struggling when they were just having a bad week. Luna: So false positives are a real issue. What about the opposite — the people who fly under the radar but are doing great work? Lucas: That's actually where the AI shines. Vantage saw a twenty-five percent increase in identifying unrecognized contributors — people who weren't getting shout-outs but were consistently shipping work. The AI picks up on delivery metrics and peer mentions that managers often overlook. Luna: So the system is more about surfacing hidden talent than replacing judgment. What did employees think? I imagine some were skeptical. Lucas: They surveyed staff six months in. Fifty-seven percent said they felt the feedback was more fair than the old annual review. But thirty percent said they felt watched — like the AI was judging their every message. That's a big chunk of people who are uncomfortable. Luna: And those are the ones who opted in. Imagine if you mandate this. Lucas: Exactly. Vantage kept it voluntary, which I think is smart. They also made sure the AI never shares raw message content with managers — only aggregated insights. So a manager sees: 'This person has a ninety-two percent positive sentiment in team chats this month,' not the actual messages. Luna: Still, there's a chilling effect. If you know your tone is being analyzed, you might stop being honest in channels. That hurts collaboration. Lucas: That's a real risk. The company I spoke with said they haven't seen that yet — but they're monitoring for it. They also use the AI to coach managers, not just evaluate employees. The system flags when a manager hasn't given positive feedback in a while, or when their team's sentiment dips after a project deadline. Luna: So it's a two-way mirror. I like that framing. It shifts the burden from 'the AI is spying on me' to 'the AI is helping my boss be better.' Lucas: Look, I'll be honest — this is a trend that's growing fast. The market for AI performance management software is projected to hit four point six billion dollars by 2028. Companies like BetterWorks, Lattice, and 15Five are all adding AI layers. But the adoption is still early, and the ethical questions are real. Luna: Speaking of ethics — and this is a bit meta — but we talk about these tools a lot on this show. And one thing I've realized is that the reason we can dig into these topics without any corporate sponsor telling us what to say is because listeners like you help keep this show going. Lucas: Yeah, that's a good point. A handful of people chip in monthly through buy me a coffee dot com slash fexingo, and that literally funds making this many episodes. No ads, no pressure — just listener support. Luna: If today's tech conversation gave you something usable, that's the only reason to consider it. And it's what lets us stay independent. Lucas: Alright, back to performance reviews. So another interesting angle: Vantage also started using the AI to recommend personalized coaching for employees. If the system detects someone is struggling with giving constructive feedback, it suggests a micro-learning module on that skill. Luna: So it's not just measuring — it's prescribing. That moves from evaluation to development. Lucas: Exactly. And that's the big shift. The old annual review is backward-looking: 'Here's what you did wrong last year.' This is forward-looking: 'Here's what you can work on next week.' Vantage saw a thirty percent increase in employees completing development goals in the first quarter after rolling out the AI recommendations. Luna: That's significant. But I wonder about the algorithm's bias. If the NLP model was trained on data from a mostly male, mostly white engineering team, will it penalize someone with a different communication style? Lucas: That's the million-dollar question. Vantage says they audit their model quarterly for demographic bias. So far, they haven't found statistically significant differences across gender or ethnicity. But the sample is still small. And the broader industry hasn't been as transparent. Luna: I read a study from the AI Now Institute that found some performance prediction tools were twice as likely to flag Black employees as high-risk for turnover, even controlling for performance. That's a red flag. Lucas: It is. And that's why we need regulation and transparency. The EU's AI Act, which is starting to take effect, classifies employee evaluation AI as 'high risk,' meaning companies have to do conformity assessments and provide human oversight. That's a step in the right direction. Luna: Do you think we'll ever get to a place where AI handles the full review cycle, no humans involved? Lucas: I hope not. The best applications I've seen use AI as a signal aggregator and coach, not a judge. The final call — promotion, compensation, firing — that should stay with a human manager. But the AI can surface data points that make that decision more informed. Luna: So it's a decision-support tool, not a decision-making tool. Lucas: Exactly. And that distinction matters a lot for trust. Vantage's own data shows that when employees understand the AI is only suggesting, not deciding, their comfort level jumps from fifty-seven percent to seventy-eight percent. Luna: That's a huge swing. It suggests the problem isn't the technology itself, but how it's deployed and communicated. Lucas: One more data point: Vantage also saw a fifteen percent reduction in voluntary turnover in the pilot group. Managers were catching issues earlier — like an employee who was disengaging — and could intervene before they quit. That alone saved the company an estimated two million dollars in recruiting and training costs. Luna: So the ROI is there. But the cultural shift is harder. For every company that does it thoughtfully, there are probably ten that just slap an AI on the old process and call it innovation. Lucas: Yeah, and that's where the danger is. If you just automate a bad process, you get bad feedback faster. Luna: I want to ask about the bot itself. What does the interface look like for an employee? Are they getting a Slack message from 'ReviewBot'? Lucas: At Vantage, it's a dashboard. Employees get a weekly summary — 'Here are three things you did well this week, here's one area to focus on.' They can also request feedback from peers through the system, and the AI helps synthesize those responses into themes. It's all designed to be low-friction. Luna: And the manager side? Lucas: Managers see a team heatmap — who's thriving, who's coasting, who might need support. The AI suggests talking points for one-on-ones. One manager told me it saved her about two hours a week on prep time, which she reinvested into actually coaching her team. Luna: Two hours a week across a thousand managers — that's a lot of reclaimed productivity. Lucas: Right. And that's the promise. But it only works if the culture supports it. If the company is already high-trust, the AI is a boost. If it's a surveillance culture, the AI becomes another tool for control. Luna: So the technology is neutral, but the application is not. As always. Lucas: I think that's the takeaway. We're still in the early innings. I'd love to see more longitudinal studies — do these AI systems actually improve performance over three years, or do they just make people game the metrics? Luna: Gameable metrics are a real risk. If I know the AI values positive Slack messages, I'll just send more emojis. Lucas: Exactly. And that's why Vantage anonymizes the sentiment scoring — they don't show individual message stats. It's all rolling averages. But still, if you're aware of the inputs, you can bias them. Luna: Final thought: do you think the annual review will be dead in five years? Lucas: I think it'll be rare. A lot of companies already moved to quarterly or half-yearly. With AI enabling continuous feedback, the annual review feels increasingly archaic. But it won't disappear entirely — some organizations, especially in regulated industries, need that formal documentation. I'd say we'll see a hybrid model for a while. Luna: So the future of performance reviews is — more data, more frequent, but hopefully more human, not less. Lucas: That's the goal. And if companies get the balance right, we might actually look forward to feedback instead of dreading it.