AI Voice Agents Are Now Taking Broadband Fault Calls: What ISPs Should Learn

On 24 September 2026, Virgin Media O2, one of the UK's largest broadband and mobile operators, started letting an AI voice agent answer some of its broadband fault calls. The agent was built with Sierra, the conversational AI company co-founded by Bret Taylor. It handles routine problems that can be diagnosed without a human adviser and hands anything complicated or sensitive to the support team. For now it covers only a small share of call volumes, and the operator says it is watching performance and customer feedback closely before expanding.

It would be easy to file this under "big European telco does big European telco thing". That would be a mistake. Fault calls are the most expensive, most repetitive and most frustrating part of running any ISP, whether you have three million subscribers or three thousand. When a major operator decides an AI can safely take the first pass at them, smaller operators should pay attention to how it was done, not just that it was done.

What Virgin Media O2 actually rolled out

The details are deliberately modest. The voice agent sits on selected broadband fault journeys, not the whole call centre. Customers can still reach a person, and the operator keeps more than 500 UK-based specialists for complaints, bereavements and vulnerable customers. Alongside the voice agent, it already runs an in-house tool called Lumi, which listens to live conversations and feeds advisers prompts drawn from the customer's earlier interactions.

Chief Operating Officer Alan Stott framed the goal as faster help on routine issues while "freeing up advisors for complex moments." Sierra's Bret Taylor said the operator is starting with one specific broadband support use case to learn from it before widening the rollout. The company also points to Ofcom data showing complaints have fallen over the past 18 months, with a growing share resolved within 24 hours.

That staged, narrow approach is the real lesson. The AI is not trusted with everything. It is trusted with the calls where the answer can be found in the data.

An AI agent is only as good as what it can see

Think about what a fault call actually involves. Before anyone can help, someone has to answer a handful of questions. Is this customer's account active? Is their connection authenticated right now? When did their session last drop? Did their payment go through? Is there a wider outage in their area?

A human agent answers those by clicking through billing screens, router dashboards and payment records. An AI agent has to answer them through data it can reach instantly and reliably. If account status lives in one spreadsheet, session state lives on a router nobody has API access to, and payments sit in an M-Pesa statement someone reconciles on Friday, no voice bot, however clever, can diagnose anything. It can only apologise politely and escalate.

This is why the most important AI-readiness work for an ISP has little to do with AI. It is about getting subscriber, session and payment data into one live system of record that software can query.

Why this matters even more for prepaid ISPs

For PPPoE and hotspot operators in Kenya and across East Africa, a large class of "my internet is down" calls are not network faults at all. The subscription has lapsed. The payment went to the wrong paybill account number. The voucher has expired. The customer paid, but nobody has reconnected them yet.

Every one of those can be answered from billing and RADIUS data in a second, and many of them should never become calls in the first place. If payment and reconnection are automated, the customer who pays is back online before they think about phoning. If session data is visible in real time, whoever picks up the call, human or AI, can see straight away whether the line is authenticated or the account is simply unpaid.

Get that foundation right, and adding an AI layer later, whether a WhatsApp bot, a voice agent or an adviser-assist tool like Lumi, becomes a sensible, low-risk step instead of a gamble.

How XpressRADIUS lays the groundwork

XpressRADIUS keeps billing and network control in one place. PPPoE subscriptions and hotspot vouchers run on real-time RADIUS, so you can see at a glance who is online, who has expired and when each session started. M-Pesa payments through Kopo Kopo, along with other mobile money gateways, are applied to the right account automatically. Paid customers are reconnected without anyone touching a MikroTik, and lapsed accounts are disconnected on schedule. Multi-currency support lets the same setup work across Kenya, the wider region and beyond.

That removes a whole category of support calls today. It also gives you the clean, queryable data any future AI support tool will need to be genuinely useful rather than a polite dead end.

Virgin Media O2's rollout shows where customer support is heading: AI handles the routine first pass and people handle the moments that need judgement. The operators that benefit first won't necessarily be the biggest. They will be the ones whose systems already know the answer before the phone rings.

← Back to all posts