With subscribers, an AI agent answers information requests on its own, at any hour and in their language. Connected to the operator’s systems, it can also look up a bill or trigger an action, within a scope set in advance. Whatever falls outside that scope, it hands to an adviser, together with what it has already checked.
At an operator, some messages call for no decision at all: where an offer stands, how a procedure works, what to do after an outage. Others touch the subscriber’s account, their money or a commercial gesture. An AI agent does not treat these two parts the same way, and that split decides what a deployment actually brings.
On information requests, the agent answers alone, in the subscriber’s language, on the web and on WhatsApp, at night or on a Sunday. Nobody is pulled in, and no ticket is opened.
This is the starting point Monaco Telecom chose. Its virtual assistant, built in Monaco with Altores and launched in June 2025, answers around the clock on the operator’s website and on WhatsApp, in several languages. At launch, it relied on the content published on Monaco Telecom’s website and stuck to information requests: no personal requests, no commercial advice, no prices. The operator also warned its customers that the assistant can make mistakes.
An agent that only reads public pages answers general questions. Connected to the operator’s systems, it answers the subscriber’s own question: it looks up the bill, the usage or the network status before answering, and it can write to the customer record or trigger an action when the operator allows it.
What makes this acceptable is the scope. It is declared in advance: what the agent reads, what it can change, what it never touches. Outside that scope, the agent does not guess.
Personal, commercial or sensitive requests go back to a person. What changes is how they arrive. The agent passes on a summary of the exchange, the subscriber’s state of mind and what it has already checked. The adviser picks up a conversation, not an empty ticket, and the subscriber does not have to repeat everything.
At Monaco Telecom, at launch, a request that went beyond information was pointed to the chat with an adviser, open from 9am to 6pm, or to a phone callback.
The same foundation can run another agent, on the team’s side. That one answers the adviser during the exchange, from the offers, procedures and internal documents, and names the document each answer comes from.
After the exchange, the follow-up work gets done too: the customer record is updated, the history is written into the operator’s systems, and recurring reasons are spotted along the way.
The lowest-risk order is the one Monaco Telecom followed: information already published first, on the channels where subscribers write, with a clear route to the advisers. Once that base is in service, each new type of request is added with its access to the systems and its scope, starting with those that only require reading.
It gives no volume or gain figures: a figure that was not measured has no place here. It covers neither voice nor the cost of a deployment, which depends on the channels opened, the systems to connect and the hosting location chosen.
It takes over the repetitive part: information requests, at any hour, and the work that follows each exchange. Personal, commercial or sensitive requests stay with the advisers, who receive them already prepared. The split is set request by request, and the operator sets it.
Yes, if it is connected to the operator’s systems and the operator allows it: it can read usage or a bill line, and write to the customer record. Outside the declared scope, it hands over rather than improvising. At launch, Monaco Telecom’s assistant stuck to the information published on the operator’s website.
No. The agent connects to existing systems and reads what it needs before answering. At Monaco Telecom, it fitted into the existing support processes without replacing any of them.
Wherever the operator decides. The Altores AI foundation can run in Monaco, in the operator’s own environment, or in the closed environment its regulator requires.
Tell us how your work actually runs, and we will tell you what an agent can take off your hands.
Talk to usCookies
We use them to improve our services.Learn more