Why Telcos Need Guardrails For Agentic AI Customer Service

Agentic AI is rapidly becoming central to telco customer service, managing voice and digital interactions with growing autonomy. But as deployments scale, many telcos still lack clear operational oversight.

Sebastian Glock, Director of Product Marketing at NiCE Cognigy, warns that without proper guardrails, small failures in speech services or external APIs can disrupt entire customer journeys.

Responsibility is often split between IT and experience teams, slowing response when issues arise. As agentic AI evolves from support tool to service backbone, telcos must treat it as core infrastructure – governed, monitored and resilient. But, are operations ready to keep up?

Many telcos are deploying agentic AI, yet a significant proportion still lack operational oversight. Why has this gap emerged?

A key reason is timing. Agentic AI has only really been deployed at scale in the past 12 to 18 months, particularly in voice based customer service. In that initial phase, organisations were understandably focused on proving value and deploying quickly.

In many cases, the operational oversight of all the dependencies involved came later. There is also a tendency to assume that APIs and third party services will always be available, stable and performant. In practice, that is not always the case, which is why visibility across both internal and external systems becomes important.

What kinds of failures are you seeing that most directly affect customer service performance?

In many situations, the AI agent itself is not the root cause. Instead, issues arise in the surrounding ecosystem. For example, a speech to text service might hit a quota limit, causing a voice agent to stop responding mid call.

An external API may become overloaded and prevent a transaction from completing. There are also well known IT issues such as routing or DNS problems. From the customer’s perspective, the interaction either fails or slows down.

Without clear insight, it can take time for customer experience or IT teams to identify where the issue sits.

Why do these problems often escalate so quickly inside organisations?

Responsibility for AI driven customer service is often split. Customer experience teams focus on tone, quality and metrics like satisfaction and resolution. IT teams are responsible for the underlying infrastructure and integrations. When something goes wrong, it is not always immediately clear which team should act first or which dependency is responsible. That initial uncertainty can delay recovery and increase the impact on customers.

How does the risk profile change as AI takes on a larger role in service delivery?

The impact of failure increases as AI becomes more embedded. Agentic is expected to handle up to 80% of customer enquiries within telcos by 2029.

Earlier generations of chatbots were relatively isolated and limited in scope. If they were unavailable, the effect on overall service levels was minimal. Today, many organisations are designing their customer service operations around AI.

They use it to manage demand, handle peak volumes and extend service availability. In that context, an AI outage can have a much wider effect, because alternative capacity may no longer be available at short notice.

How should telcos approach monitoring and oversight without creating additional complexity?

The goal should be clarity rather than volume. Teams need insight that helps them understand what is happening across the system and who needs to act.

That information should be accessible to people with different levels of technical expertise and should fit into existing operational processes.

Oversight works best when it supports established incident management and escalation models, rather than requiring entirely new workflows.

Looking ahead, how might increasing sophistication in agentic systems affect operational requirements?

Most organisations are still focused on running a single AI agent reliably. More complex models, such as multiple agents collaborating across tasks or handing work between AI and humans, are still emerging.

As those patterns mature, operational visibility will become more important. You may have synchronous and asynchronous interactions happening at the same time, with context passed between systems and people. Being able to trace those journeys and understand where delays or failures occur will matter more as complexity increases.

At a strategic level, what should telco leaders be thinking about now as they scale agentic AI?

AI should be considered part of the operational fabric, not just a technology experiment. As its role in customer service grows, questions around ownership, accountability and resilience become more important.

Leaders need to think about how AI systems are governed, how failures are handled and how transparency is maintained across teams. In highly regulated and customer sensitive environments like telecoms, taking a structured approach early can help organisations scale AI in a way that supports reliability and trust over time.