Latest / The Productivity Podcast with Fexingo: Output, Efficiency, and Long-Term Economic Growth / How AI Assistants Raised Call Center Productivity 14 Percent
Transcript
- Lucas: There's a number that's been stuck in my head all week: fourteen percent. That's the productivity boost a large call center saw when it gave its customer service agents access to a generative AI assistant. Luna: Fourteen percent is huge in the economics of productivity. Most interventions — new software, training programs — struggle to get half that. Lucas: Exactly. And this wasn't a lab experiment. This was a real company, a Fortune 500 firm, with thousands of agents handling real customer calls. The study was done by researchers from Stanford and MIT, led by Erik Brynjolfsson, who's been studying tech and productivity for decades. Luna: How did they measure it? Call center productivity is notoriously hard to quantify — you have resolution rates, customer satisfaction, handle time… Lucas: They looked at the number of issues resolved per hour — basically, cases closed per agent per shift. And they compared agents who had the AI tool against a control group that didn't. The AI didn't automate the whole job. It gave agents real-time suggestions — responses to customer questions, troubleshooting steps, policy lookups. Luna: So it's more of a copilot than a replacement. That fourteen percent — was it uniform across all agents? Lucas: That's the really interesting part. The productivity gains were heavily skewed. The most experienced, top-performing agents saw almost no improvement. But novice agents — those in their first few months on the job — saw a thirty-five percent boost. And agents with the lowest prior performance caught up to the average much faster. Luna: So the AI essentially compressed the learning curve. That's a different story from 'AI replaces jobs' — it's more about closing skill gaps. Lucas: Right. And that has big implications for how we think about productivity growth in the service sector. For decades, productivity in services has lagged manufacturing. Manufacturing benefits from automation — robots, machines — but services rely on human judgment and conversation. AI seems to be the first technology that genuinely makes service workers more productive without eliminating the human touch. Luna: Quick honest thing — a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many of these. Walking through the economy with us has made something click for some people, and that's why we can keep it ad-free. Lucas: Yeah, it's a small group, but it's exactly what keeps the production engine running. No ads, no sponsors — just listeners who find value in understanding how the actual economy works. Luna: If that's you, we appreciate it more than we can say. And if not, no pressure at all — you're still part of the show. Lucas: Back to the call center study. One thing that stood out to me is how the company rolled it out. They didn't fire anyone. They didn't reduce headcount. They just handled more calls with the same team. Luna: That's a key point. In economics, productivity gains can show up as either more output with same inputs, or same output with fewer inputs. Here it was more output. Lucas: And that meant faster service for customers and lower wait times. The company didn't disclose its name, but it's a large U.S. firm in the business process outsourcing space. Think thousands of agents across multiple sites. Luna: How did they handle the data privacy aspect? Call center conversations are sensitive. Lucas: The AI was trained on the company's own knowledge base — policies, FAQs, product specs — not on live calls. It wasn't recording or analyzing the calls. It just suggested responses based on the customer's typed or spoken question. So it was relatively low-risk. Luna: Makes sense. Now, fourteen percent is impressive, but call centers are a specific use case. How transferable is this to other knowledge jobs? Lucas: That's the billion-dollar question. There's a parallel study on software developers using GitHub Copilot. That found a productivity gain of about fifty-five percent on a specific task — implementing a function. But that's a narrow, well-defined task. Call center work is also relatively structured. For more open-ended knowledge work — strategy, analysis, creative work — the gains are less clear. Luna: So we might see AI boost productivity most in roles that have a high degree of repetition or pattern recognition. That's a lot of jobs — customer support, data entry, basic coding, administrative work. Lucas: Exactly. And if those jobs get a ten to twenty percent productivity boost across the board, that could meaningfully lift overall economic productivity. The service sector is about eighty percent of U.S. employment. Even a small shift there would show up in GDP growth. Luna: But we've been hearing about an AI productivity boom for years, and official productivity statistics haven't shown a big uptick. What gives? Lucas: There's a lag. Brynjolfsson calls it the 'productivity J-curve'. When a new general-purpose technology arrives, we first invest time and money into reworking processes and building complements. Productivity dips before it rises. We saw that with electricity and computers. AI might be in that dip right now. Luna: So the study is a leading indicator. If a single tool in a single company can deliver fourteen percent, imagine what happens when thousands of firms figure it out. Lucas: That's the optimistic view. The pessimistic view is that many of the gains get captured by a few frontier firms, and the average company — the laggard — doesn't adopt at all. We've talked about that gap before, the one between frontier and laggard firms. That gap could widen. Luna: The study did mention that the company had a pretty tech-savvy management and a data-driven culture. Not every call center has that. Lucas: Right. So the glass half empty story is that AI could increase inequality between firms and between workers. The glass half full is that it gives novices a huge leg up, which could be great for entry-level workers and job mobility. Luna: Let's talk about the workers. Did the study look at turnover or satisfaction? Lucas: They did look at agent retention. The AI actually reduced turnover slightly, especially among new hires. The theory is that it reduces stress — agents feel less overwhelmed because they have a reliable assistant. So it might not just raise productivity but also improve job quality. Luna: That's a powerful combination. Higher productivity, lower stress, better customer service. It almost sounds too good to be true. Lucas: Well, it's one study, one company, one AI tool. The replication will be key. But it's one of the first rigorous field experiments on generative AI in the workplace, and the results are genuinely encouraging. Luna: So what's the takeaway for someone who manages a team or runs a business? Should they rush out and buy an AI assistant? Lucas: The study suggests that the biggest gains come when you pair the tool with good training and integration. It's not a plug and play miracle. The company had to customize the suggestions, test them, and train agents on when to trust the AI versus override it. The technology is necessary but not sufficient. Luna: That sounds like a classic productivity lesson: new tools only work if you redesign the process around them. Lucas: Exactly. And that's where the real productivity growth comes from — not the gadget itself, but the rethinking of how work gets done. Luna: Last thing — does this study change how you think about the future of work? I know you're usually skeptical of techno-optimism. Lucas: I remain cautiously optimistic. I think AI will be a big deal, but not in the way the hype suggests. It won't replace whole professions overnight. It will slowly, unevenly boost productivity in specific tasks, and over a decade or two, that adds up. The call center study gives us a glimpse of how that might play out in practice. Luna: And as always, the data keeps coming. We'll be watching for the next wave of field studies. Lucas: Absolutely. That's it for today. Thanks for listening — we'll be back tomorrow with another story about how the world actually works.