For nearly 150 years, KPN, the Netherlands’ largest telecom provider, has responded to every technological shift by leaning in and often leading the way. In recent years, that has included rolling out a future-proof fiber network to two-thirds of the Netherlands and becoming the first European provider to offer real-time text to all its customers during mobile calls.

With new technologies emerging, KPN took a similar approach, acting as an early mover to launch AI-powered chatbot solutions, including an agentic-enabled invoice explainer and a generic Q&A agent available to all customers. As voice models advanced, KPN saw an opportunity to further enhance customer service quality, reduce call center wait times, and position its customer service at the forefront of agentic AI adoption—moving beyond text-based automation to dynamic, voice-to-voice interactions that deliver more connected, high-impact experiences at scale, spanning roughly five million customer calls annually.

“We wanted to make this next step toward the future,” says Jörg Kramer, KPN’s chief of customer and commercial operations. “We see AI as a way to support more natural interaction with customers, while ensuring that human support remains available when it matters most.”

We wanted to make this next step toward the future. We see AI as a way to support more natural interaction with customers, while ensuring that human support remains available when it matters most.

Jörg Kramer

KPN’s Chief of Customer and Commercial Operations

KPN had already applied agentic AI to internal processes, including knowledge-based bots. Building on that experience, the next logical step was to extend AI-enabled interaction to customers—bringing the same capability into the moments that matter most in the customer journey.

To unlock that broader ambition for its customer care function, KPN partnered with McKinsey and its AI arm, QuantumBlack, to move beyond standalone AI features toward a scalable, in-house agentic capability. By making customer care one of the front-runner domains to fundamentally reimagine end to end, the telecom aims to gain insights that it can apply as it deploys agentic AI and evolves its organization more broadly.

“Customer care was the first domain for KPN to start building out agentic capabilities, serving as a proof point for how the company can reimagine other domains to build an AI-native enterprise,” says Maarten Baeten, a McKinsey partner. “KPN made the deliberate choice to build a reusable technology stack and in-house development capabilities to accelerate organization-wide adoption.”

KPN made the deliberate choice to build a reusable technology stack and in-house development capabilities to accelerate organization-wide adoption.

Maarten Baeten

McKinsey Partner

KPN’s transformation centered on four priorities: identifying high-impact opportunities, building the technical foundation, scaling solutions into production, and preparing its workforce for change.

Identifying high-impact opportunities

Using a custom-built large language model framework, the team analyzed large volumes of call transcripts and chat interactions to identify the most common customer needs and how often they occurred, creating a clear, data-driven view of why customers were reaching out. All analyses were conducted on anonymized data in line with strict privacy and European data protection standards.

In collaboration with customer experience and technical groups, they mapped these interactions to identify where processes could be simplified or automated based on impact value, development complexity, and feasibility.

They weighed not only implementation complexity, value at stake, and cost but also the need to keep humans in the loop, such as during sensitive moments like customers moving to a new residence, where churn risk is higher.

This analysis led to a clear set of high-impact use cases, including customer verification, order status inquiries, technician appointment management, and internet connection troubleshooting—all instances in which AI can handle interactions competently while maintaining a strong customer experience.

“KPN saw that agentic AI could fundamentally reshape and reimagine how customer service is delivered,” says Ruben Schaubroeck, a McKinsey senior partner. “It allows them to orchestrate entire journeys around outcomes while keeping human experts focused where they add the most value.”

KPN saw that agentic AI could fundamentally reshape and reimagine how customer service is delivered. It allows them to orchestrate entire journeys around outcomes while keeping human experts focused where they add the most value.

Ruben Schaubroeck

McKinsey Senior Partner

Building the technical foundation

The team began by building the technical foundation needed to support AI-driven voice conversations at scale. This meant setting up a platform that enabled teams to quickly build AI agents; configure customer interactions, including natural-sounding voice responses; and connect to core telco systems.

A key decision was how to structure the conversational layer. From the outset, the team knew that delivering high-quality voice interactions meant minimizing latency and keeping response times below two seconds per turn. Because human conversations typically have response gaps of under a second, even small delays would make interactions feel artificial.

The team also worked extensively to enable “barge-in”—allowing customers to interrupt midsentence—to create a more natural flow, while ensuring the system maintained context and generated accurate responses.

Keeping those requirements in mind, the team evaluated different architectural approaches and ultimately chose an integrated, managed conversational platform—allowing them to deliver natural interactions, centralize speech and AI capabilities, and accelerate development while maintaining consistent quality.

Beyond the conversational layer, the team defined the broader architecture across three areas—agent orchestration, the agent platform, and audio integration—and structured them into connected layers so everything worked together seamlessly. The platform setup was also designed to be reusable, enabling new use cases to be added over time rather than building one-off solutions.

This approach was key for a project of such complexity and ambition. It’s one thing for an organization to deploy individual agents for isolated tasks; what KPN was undertaking at scale was entirely different and higher stakes, requiring new capabilities, such as guardrails and observability.

“We’re building out agents that will be dealing with over 100,000 customers a week,” says Paul Bosch, KPN’s CIO and EVP of B2C Solutions. That meant having to ask: “How do we train and prompt these agents to act within guardrails and in line with our company values in interactions with our customers?”

We’re building out agents that will be dealing with over 100,000 customers a week. That meant having to ask: ‘How do we train and prompt these agents to act within guardrails and in line with our company values in interactions with our customers?’

Paul Bosch

KPN’s CIO and EVP of B2C Solutions

Designing, delivering, and scaling solutions

The team moved quickly to bring these use cases into production, adopting rapid build-test-release cycles and launching minimum viable product versions early. Cross-functional teams tested extensively, including sessions with real customers in a UX lab.

Because AI systems can respond differently to the same input, ensuring reliability required new approaches to quality and control. The team introduced guardrails to ensure agents stayed within defined conversational boundaries and aligned with KPN values, alongside observability tools to monitor behavior and performance.

They also developed an evaluation framework, combining human-led testing with automated quality checks to assess performance, consistency, and safety of interactions.

Once live, the team continued refining the agents. The daily cycle began with reviewing transcripts from up to 100 real customer calls to assess performance, which surfaced unexpected phrasings, edge cases, and improvement opportunities. In the afternoon, they would leverage these insights to change prompts and then release the updates by each evening.

Readying the workforce for AI-native operations

Preparing the organization for this new way of working was just as important as the technology itself. As AI agents began supporting tasks such as customer verification and initial troubleshooting, customer experts increasingly focused on more complex interactions where human judgment and empathy are essential.

KPN created a task force to update operating procedures, develop training programs, and support employees as their roles evolved.

By involving frontline employees in development and testing, KPN created a strong feedback loop that improved both customer and employee experience.

“I am proud that our adoption approach has achieved a success rate of more than 86 percent,” says Kramer. “A clear change narrative, well-defined metrics, and deliberate employee engagement have proven critical to helping teams embrace a new way of working.”

I am proud that our adoption approach has achieved a success rate of more than 86 percent. A clear change narrative, well-defined metrics, and deliberate employee engagement have proven critical to helping teams embrace a new way of working.

Jörg Kramer

KPN’s Chief of Customer and Commercial Operations

KPN’s approach is already delivering tangible results. By reducing handling time, expanding the range of interactions AI can support reliably, and helping resolve issues more efficiently—often without the need for a technician visit—the program is reshaping customer service while maintaining a strong customer experience.

At the same time, it allows human agents to focus on complex issue resolution rather than routine tasks such as customer verification.

KPN has set the ambition for agentic AI to handle 10 to 20 percent of customer service calls by 2027, with early indicators showing rising customer satisfaction.

“Many organizations are still exploring the potential of agentic AI, while KPN is already integrating it into how work gets done every day,” says Dirk Hofland, a McKinsey associate partner. “That kind of alignment between strategy, technology, and execution is what allows these programs to scale and deliver real impact.”

Many organizations are still exploring the potential of agentic AI, while KPN is already integrating it into how work gets done every day. That kind of alignment between strategy, technology, and execution is what allows these programs to scale and deliver real impact.

Dirk Hofland

McKinsey Associate Partner

Just as important as the new platform is the capability KPN has built internally. Through close collaboration with McKinsey and QuantumBlack, teams have developed the skills to design, launch, and scale new agentic AI use cases independently—making this an ongoing capability rather than a one-time initiative.

As Bosch puts it, “It’s not just the technology. It’s about how we scale the mindset of people and the organization to make something like this happen.”

Mate Maczik and Tobias Jongbloets contributed to the development of this article.