What AI Agents Are Actually Doing in Telecom Operations

Will Drewes

Founder, Fern Strategy · 3 min read

The numbers are starting to land

Telecom is a useful industry to watch right now. The workflows are well-defined, the data is dense, and the cost of a bad customer interaction is measurable in churn. That combination tends to produce cleaner outcome data than most sectors. A few recent examples are worth walking through.

China Mobile: first-line deflection and faster repairs

China Mobile's AI customer assistant handled 90% of first-line inquiries in pilot regions and lifted customer satisfaction scores by 10%. Separately, predictive analytics cut network repair times by 30%. Those two numbers belong in the same sentence because they're solving the same underlying problem from different directions: reduce the time between something breaking and a customer noticing it, and reduce the time a customer spends waiting for help when they do. Getting both moving at once is harder than it looks organizationally, which makes this a meaningful data point.

Telstra: internal tooling with measurable retention impact

Telstra built generative AI tools in-house and reported 90% time savings for employees using them, alongside a 20% reduction in follow-up contacts. That second number is the one to focus on. Follow-up contacts are a direct proxy for resolution quality. Fewer callbacks means the first interaction resolved the issue. At Telstra's scale, a 20% reduction in repeat contacts is a retention lever, not just an efficiency metric.

Two unnamed operators: conversion and productivity

One European telco used AI-driven personalized marketing to push conversion rates up 40% while reducing campaign costs. One Latin American operator improved call-center agent productivity by 25% through AI-generated recommendations during live interactions. Neither company is named in the public reporting, but both outcomes are specific enough to be useful benchmarks. A 40% conversion lift in a competitive consumer market is not a rounding error. A 25% productivity gain in a call center changes headcount math.

One North American operator: CapEx reduction through autonomous optimization

A North American operator deployed an autonomous network optimization agent and reduced network capital expenditure by 10%. In telecom, CapEx is a major line item. A 10% reduction through autonomous planning is the kind of outcome that gets a CFO's attention faster than any customer-facing metric. It also points to a category of agent use case that gets less press: not customer-facing automation, but infrastructure decision support running in the background against stable rules and complex inputs.

What this means for telecom operators

The pattern across these cases: the highest-value deployments are not the most visible ones. China Mobile's repair-time reduction happened in the network layer. Telstra's gains came from internal tooling. The North American CapEx story is entirely back-office. Customer-facing chatbots get the demos; the measurable ROI is often upstream.

The governance question that follows is predictable. An autonomous agent recommending network investment decisions or handling 90% of customer inquiries without human review needs an audit trail. Not because regulators are watching today (though in some markets they are), but because when something goes wrong at that deflection rate, you need to know exactly what the agent did and why. A black-box deployment at 90% deflection is a liability. An auditable one is an asset.

If your team is scoping an agent deployment in network ops or retention workflows, start by naming the outcome metric, not the use case. "Reduce follow-up contacts by 15%" is a target you can build governance around. "Improve customer experience" is not.