Agentic AI: driving a deep transformation in customer experience

Year after year, all barometers agree: when it comes to mass-market customer experience, the telecom industry as a whole is lagging behind. Among the culprits: long processing times, opaque bills, and journeys riddled with friction. However, where operators do stand out is in their ability to quickly harness new technologies. Agentic AI appears to be no exception. With its promise to deliver automated, personalized experiences, could it be the long-awaited key for operators to finally catch up with the best-in-class in customer experience? Better yet, will it enable them to differentiate themselves in a context where the increasingly shared infrastructures and networks reinforce the commoditization of services?

From words to actions

Since the advent of LLMs, we already knew that AIs are particularly “smart” and comfortable with naturallanguage interactions. Now, with agentic AI, we are discovering their ability to move from words to action, by leveraging internal and external functions and tools, with minimal or even no human supervision. This tool-augmented intelligence has two fundamental characteristics that strongly distinguish it from rule-based automation technologies:

• Advanced mimicry of human intelligence and action, including the ability to plan and adapt. Faced with a goal to achieve, an AI agent will iterate until it finds the best sequence of actions to perform and create a workflow on the fly. If necessary, it will collaborate in a team with other AI agents, in a division of labor and hierarchy that mirrors human collaboration.
• A high level of standardization of mainly open source interfaces (APIs) and communication protocols (MCP and A2A), which ensures interoperability within agentic systems, facilitates their evolution, and encourages the reuse of their components – as seen in the emergence of AIagent marketplaces.

Acceleration of customer relationship automation

For operators, these characteristics open up the prospect of further automating end-to-end customer processes and interactions. In this context, conversational AIs (chatbots, voicebots and callbots) are gradually becoming the preferred interfaces in this new human-AI agent relationship. They also have the advantage of integrating into numerous touchpoints (IVR, mobile applications, WhatsApp, etc.) as well as in the back office, in “agent assist” and “copilot” type solutions. But the impact of AI agents on customer relationships goes beyond efficiency gains.

Smoother and personalized customer journeys

From customers’ perspective, journeys will be dynamically and automatically crafted on a case-by-case basis, depending on the context. The benefits for them are numerous:

• Permanent and immediate availability (selfservice),
• The possibility of expressing a request in a simple way (in natural language),
• More relevant responses and proposals, because of personalization,
• Requests handled more quickly, endtoend and often on first contact.

Ultimately, agentic AI[1] helps establish or strengthen every one of the six pillars of customer experience highlighted by KPMG, compared to "classic" AI:

Operators’ hybrid approaches

However, operators’ interest in automation and conversational AI for customer experience is not new. Many have long been offering virtual assistants based on business rules to automatically handle simple requests. Rather than throwing away these existing assets, most are now opting to first embed generative AI into them, then agentic capabilities, thereby automating more support tasks without deviating (too much) from the established rules and journeys. Among other benefits, this hybridization strategy helps keep economic and environmental costs under control, and it is beginning to bear fruit. Vodafone, for instance, claims a 70% resolution rate and an 8 point NPS improvement with SuperTOBi, an AI agent deployed across all its European subsidiaries and built as an evolution of its historical virtual assistant TOBi. Deutsche Telekom is taking a similar path with Frag Magenta, which has evolved from a simple virtual assistant into an AI agent capable of handling customer requests for which there is no predefined script. Frag Magenta has reportedly saved 133,000 contact center agent hours in the first half of 2025. Meanwhile, a few other operators are making a bolder bet.

A personal AI agent always at hand

Imagine having a personal AI agent in your smartphone that would perform a whole range of time-consuming daily tasks for you, in every area of life, following a brief voice instruction from you: choosing and booking a restaurant, or making a doctor’s appointment for your youngest child. Nothing revolutionary there, you might say. And yet, this personal AI agent would not just be a simple executor, skillfully navigating your applications or the Web, with access to your credit card details and your calendar. Its main promise would be to free you from the mental load associated with organizing daily life, day after day. Moreover, it would know you so well that it would anticipate your needs even before you articulate them. This is the vision driving South Korean operator SK Telecom’s “Aster” experiment launched in March 2025 in North America. To ensure the success of its smartphone-based personal AI agent, SK Telecom has developed a whole ecosystem of partnerships and integrations, including Perplexity, Google Calendar, Yelp, Uber and Lyft.

Competition from the Big Tech companies

In the market for personal AI agents, however, operators face stiff competition – who wouldn’t want to position themselves as the orchestrator of their customers’ lives? For more than a decade, tech giants have been pursuing this holy grail with embedded AI assistants. To equip the latter with agentic capabilities and encourage adoption, they are now banking on a wider variety of connected wearable devices, such as glasses – whose prices have dropped fivefold in 10 years – or pocket devices, and even brain computer interfaces. However, Big Tech companies’ strategy revolves less around agentic AI and more around AI models designed to understand and interact with the physical world. This is one of the main goals of Google DeepMind’s Astra research project, in particular.

New revenue streams for operators

Faced with Big Tech companies, operators that already have Super Apps may be in the best position to assert a personal and universal AI agent. These operators can rely on an active user base, already operational marketplace vendors, and possibly mobile money solutions. An integrated personal AI agent will simplify user journeys, especially for purchases: it will help customers more easily find the product or service they are looking for within the Super App and may even complete the purchase on their behalf. In the form of a multilingual voicebot, especially when speaking languages still poorly mastered by LLMs/SLMs, this agent will also drive inclusion and accessibility, thereby expanding the addressable customer base. For operators, this will translate into additional direct revenue, mainly within marketplaces.

Amplification of existing risks and new challenges

Operators must nonetheless ensure user trust and employee engagement by adopting an appropriate governance model for agentic AI, in which humans retain the final say and can pull the plug on AI agents that have become rogue. Indeed, connecting tools and functions amplifies the potential impact of known generative AI issues, chief among them being the risk of hallucination. The likelihood of these issues also increases, due to the complexity of such systems and notably their expanded attack surface, creating new vulnerabilities to cyber threats. But the greatest challenge for operators will probably be transforming their organizations and supporting the evolution of roles and skills for their employees, especially frontline staff, so that they find their place in a customer experience reshaped by agentic AI: experts in complex cases, VIP advisors, designers of agentic workflows or AI agent managers – the range of possibilities remains to be explored together with the very people who are most affected.

Will we bond with AI agents in the future?

In short, agentic AI is poised to fundamentally reshape the mass-market customer experience in telecoms, delivering automated personalization, seamless journeys, and measurable operational gains. This promise, however, comes with amplified risks and growing competitive pressure from Big Tech companies. For agentic AI to become a true differentiator, speed of execution will be key, as always. Additionally, this speed must go hand-in-hand with a parallel transformation of governance, organization, and skills – and crucially, with support for teams navigating the change. Yet to make this advantage truly last, isn’t building an emotional bond with customers – the very key to engagement and loyalty – also essential? And can AI agents be the catalysts of this bond? With the rise of parasocial relationships – the one-sided, imagined bonds people form with virtual beings or out-of-reach celebrities – the concept no longer seems like pure science fiction.