Customer care has long played second fiddle to network build-out and coverage in mobile operator strategy. That is changing fast as 5G networks converge on similar speeds and coverage, forcing operators to compete on the experience around the network - and AI is now the technology reshaping that fight, according to a recent Mobile World Live panel featuring executives from CSG, Red Hat, and GSMA Intelligence.
The cost reduction opportunity hiding in plain sight
Customer care represents one of the biggest, most improvable cost lines on an operator's books. GSMA Intelligence maintains a network economics model built on operators' real costs and revenues, and the range it uncovers is striking. "Some operators spend a relatively small amount, while others devote between 3% and 5% of operating expenditure to customer care, call centres and customer interactions," said Emanuel Kolta, lead analyst at GSMA Intelligence. "That creates a substantial opportunity for AI to reduce costs and improve outcomes."
GSMA Intelligence's own tracking of live telco AI deployments finds customer care is the single largest use case category, accounting for roughly half of all projects operators have underway. About three quarters are already running in production rather than stuck in trial.
From canned responses to orchestrated action
What is changing is not just where AI sits, but what it can do. "Traditional chatbots are often frustrating because they provide canned responses that do not fully address the customer's need," said Chad Dunavant, chief strategy and product officer at CSG. The alternative now emerging is agentic and orchestrated AI. A customer asking why their bill changed can trigger a billing agent, then a catalogue or rate-change agent, chained together into something that "begins to resemble an interaction with an experienced call centre representative who can take several actions on the customer's behalf."
Dunavant added: "The important difference is that the customer gets something done instead of receiving a generic answer and calling the contact centre anyway." Making orchestration work depends on feeding these agents the right information, not simply more of it. "The best context is not the largest context," said Fatih Nah, distinguished chief architect in Red Hat's CTO office. "It is the most relevant and precise context." Dumping every available data point into a model dilutes what matters and wastes computing resources. "Operators do not need a gigantic hammer for a small nail," Nah said.
That orchestration increasingly spans more than text. Customers can choose whichever mode fits the problem, from a short instructional video to an audio walkthrough that adapts as it goes - "a two-way, interaction-based model rather than a one-way question-and-answer exchange," Nah said.
Distributed intelligence and the commoditisation of connectivity
Because some interactions are latency-critical, intelligence needs to be distributed across the network rather than concentrated in one distant cloud. This ranges from the radio edge for urgent cases through to regional and core data centres for more complex tasks. The shift matters more than ever because connectivity itself is becoming commoditised. "The industry is entering a different phase of 5G," said Kolta. "Competition is moving beyond speed and traditional service quality towards reliability, responsiveness and the quality of the relationship with the operator."
Where infrastructure sharing narrows the gap between networks, poor service becomes a powerful push factor for churn. AI-driven customer care may be one of the missing links in a complete digital transformation at operators whose contact centres still look much as they did 20 years ago. Telcos are not alone in this: banks faced a similar problem of customers scattered across mortgages, loans and current accounts, which is partly why they moved first, needing, as Dunavant said, "to recognise one customer across many silos and manage the interaction consistently at the front door." Telcos, he added, are "not far behind."
Guardrails, trust, and the agent-to-agent future
None of this works without trust, and generative AI is not without its trust issues. "A hallucination could cause financial loss, delete data, disrupt a production environment or reduce service availability," Nah said. Guardrails are a vital part of the picture. Dunavant described a workflow in which an offer agent recommending a 20% discount, above an authorised 10% limit, is automatically routed to a person. He noted the control was "no different in principle from a call centre, where agents can approve discounts only up to a defined threshold and must refer larger concessions to a supervisor."
Looking ahead, the panel turned to what happens when customers start sending their own AI agents to deal with operators. Dunavant thinks pricing and plans will need to become legible to machines as well as people, warning that operators invisible to shopping agents risk being bypassed altogether. "Search engine optimisation could be followed by a new discipline of agent optimisation," he said. Kolta suggested today's crowded app ecosystem is ripe for transformation, while Nah described a future built around AI-native 6G networks in which intelligence is designed in from the start rather than patched on afterwards. "In many respects, the AI agents will become part of the network itself," he said.
The panel also covered a concrete result from a live deployment, where placing an intelligent agent at the front of the interactive voice-response process cut calls reaching human agents by 71%. They discussed the early rules now being drafted for agent-to-agent communication between companies, and how operators are combining generative and deterministic AI to keep decisions auditable in a regulated industry.
Why this matters for customer support teams
The 71% reduction in calls reaching human agents is not a theoretical projection - it is a measured outcome from a live deployment. For support leaders, the implication is clear: orchestrated AI that chains multiple agents together to complete tasks, rather than serving static answers, can shift contact volumes at scale. The guardrail patterns described - such as automatically escalating decisions that exceed authorised thresholds - offer a practical model for teams building AI workflows that remain auditable and safe, which is especially relevant for those exploring AI for Customer Support. For supervisors managing these transitions, the panel's emphasis on agent orchestration and distributed intelligence aligns directly with the skills covered in the AI Learning Path for Call Center Supervisors.
Your membership also unlocks:
