17.13 Customer service and contact centres
Contact-centre operations were one of the first places generative AI showed large measurable productivity effects in production. Brynjolfsson, Li and Raymond's 2023 NBER paper studied a Fortune 500 company's deployment of a generative-AI agent assistant for 5,179 customer-support agents, measuring productivity (issues resolved per hour) over 11 months. The AI assistant, which suggested responses in real time and surfaced relevant knowledge-base articles, increased productivity by 14% on average, with the largest effects on novice agents (34% improvement) and minimal effects on the most experienced agents.
By 2026 most major contact-centre platforms, Genesys, NICE CXone, Five9, Salesforce Service Cloud, Zendesk, have integrated LLM-based agent assistance, conversation summarisation, post-call wrap-up automation and self-service chatbot capabilities. The end-to-end automation push, in which the AI handles the customer interaction without a human agent, has been more cautious; tier-one support automation rates of 30–50% are now common for well-defined product domains, but the long tail of unusual cases continues to require human handling.
The economic effects on the contact-centre workforce are visible. Klarna's February 2024 announcement that its OpenAI-powered chatbot was handling two-thirds of customer service chats and doing the work of 700 full-time agents was widely reported. India's BPO sector, employing several million people, has seen significant pressure from offshore-onshore agentic-AI substitution, though the net employment effect over the full transition remains uncertain. Klarna's subsequent acknowledgement in 2025 that it had over-automated and was rehiring human agents to handle quality concerns is a useful corrective: customer service involves emotional regulation, escalation handling and complex problem-solving that current AI handles unevenly, and platforms that remove humans from the loop entirely tend to discover the long tail at customer cost.