Generative AI at Work
Abstract
We study the staggered introduction of a generative AI-based conversational assistant using data from 5,179 customer support agents. Access to the tool increases productivity by 14 percent on average, with the largest gains for novice and low-skilled workers. AI assistance enables less experienced workers to perform at levels comparable to workers with 6+ months more tenure.
★ Curator Summary
This is the paper that changed my thinking about AI in contact centers. Brynjolfsson's team got access to a real deployment at a Fortune 500 tech company and tracked actual productivity data. The headline finding — 14% average productivity boost — is important, but the real insight is the heterogeneity: novice workers got a 34% boost while experienced workers got essentially zero. AI is a skill equalizer, not a uniform multiplier. That finding has massive implications for workforce strategy, hiring, and training investment.
Why It Matters
For WFM practitioners, this paper fundamentally changes how you model AI-augmented agent productivity. You can't apply a single productivity multiplier across your workforce. Instead, you need to segment by tenure and skill level. The experience curve compression (2 months with AI = 6+ months without) means faster time-to-competency and potentially different hiring profiles. The 40% attrition reduction for new hires is a workforce planning input that belongs in your total cost model.
Caveats
Single firm, single AI tool, specific context (chat-based tech support). The productivity measure is issues resolved per hour, which may not capture quality dimensions. The study period was relatively short. We need replication across industries and AI tools before treating these numbers as universal constants.
Discussion
Loading...