Dutch Telecom Builds an AI System for Millions of Customer Calls

KPN, the largest telecommunications firm in the Netherlands, is putting agentic artificial intelligence to work where customer service meets one of the industry’s biggest operational challenges: millions of routine conversations that still require time, systems access and human attention.

KPN’s agentic AI case study, done in partnership with McKinsey, shows how an AI system can move beyond answering customer questions to completing tasks, connecting with back-end systems and handing more complicated interactions to employees. Building that capability involved redesigning workflows, establishing safeguards and changing how customer service teams work alongside AI.

KPN’s customer service operation handles roughly 5 million calls a year, making even routine interactions a significant operational undertaking. The firm had already introduced AI-powered chatbots, including an invoice explainer and a general question-and-answer agent. As voice technology improved, KPN began looking for a way to extend AI into the phone conversations that still formed a major part of its customer service operation.

The challenge was to make those conversations useful without turning them into another layer of automation for customers to navigate. KPN partnered with McKinsey and its AI arm, QuantumBlack, to build an agentic AI capability that could handle customer interactions directly while keeping human employees involved in more sensitive or complicated situations.

Deploying Agentic AI

The teams started with KPN’s existing customer interactions. They analyzed large volumes of anonymized call transcripts and chat records to identify recurring customer needs and determine which processes could be simplified. The resulting use cases included customer verification, order-status inquiries, technician appointment management and internet troubleshooting.

That analysis shaped the technology that followed. KPN built a platform connecting AI agents with the company’s core telecommunications systems, allowing the agents to do more than provide information. The system was designed for voice conversations, with response times targeted below two seconds per turn and the ability for customers to interrupt an agent mid-sentence while the system retained the conversation’s context. The architecture was also designed to support additional use cases rather than requiring KPN to build each application separately.

Operating an AI system at that scale required controls alongside automation. KPN introduced guardrails to keep agents within defined boundaries and developed tools to monitor their performance. Human testers and automated checks evaluated the interactions for quality and safety. After deployment, teams reviewed transcripts from up to 100 customer calls each day, used those findings to refine prompts and released updates by the end of the day.

The workforce was another part of the implementation. KPN created a task force to update procedures and training as AI began taking on routine activities. Frontline employees participated in development and testing, while customer-service specialists increasingly focused on more complicated interactions requiring human judgment and empathy. KPN says its employee adoption approach achieved a success rate of more than 86%.

Cost Reductions

The early results have encouraged KPN to expand the program. The company says artificial intelligence has reduced handling times, helped resolve some issues without technician visits and allowed human agents to spend more time on complex problems. KPN has set a goal for agentic AI to handle 10% to 20% of customer-service calls by 2027, while early indicators show customer satisfaction comparable with human-handled calls.

KPN’s experience shows that putting agentic AI into customer service involves as much operational redesign as technology deployment. The company is expanding the system while maintaining human involvement for more complex interactions and building processes to monitor and improve its performance. As KPN continues to expand its agentic AI capacity, its approach provides a view of how telecom operators can introduce AI into high-volume service operations without removing people from the parts of the job that require human judgment.