Telecom operators are among the industries using artificial intelligence at several levels at once. On one side, AI and machine learning support network planning, optimisation and monitoring, activities operators began long before the current wave of generative AI. On the other, the latest models accelerate product development, testing of new services and customer support. Matija Ražem, Chief Telecom Commercial Officer at Infobip, sees this combination as one of the most important changes for the sector.
“AI allows you to build a prototype of an idea literally in a few hours,” Ražem tells ICTbusiness Media – ICTbusiness.info & ICTbusiness TV. That is not unique to telecoms, but it matters greatly in this industry because operators are large organisations with complex infrastructure and slower development cycles. AI does not mean telecoms will suddenly behave like startups, but it gives them more room to validate ideas before entering traditional development.
In practice, an operator can design a product, create a prototype, gather feedback and refine it before committing significant resources to full development. Ražem sees this as particularly relevant in business development, when telecoms approach the market with new offers. AI shortens the path from idea to testing and reduces the risk of complex projects being built for months without clear evidence of customer demand.
Another major area is network operations. Telecoms have long used machine learning for routine network tasks: traffic optimisation, selecting better routes, detecting failures, and switching to healthy links. The most advanced generative models are not always required for such use cases, Ražem says, but AI and machine learning clearly help. Regulation also remains crucial in telecoms because every solution has to operate within a strict framework.
The most visible change for end users is in communication. Conversational AI and agentic systems allow a brand, such as a telecom operator, to build an agent that understands common user problems and can provide a concrete answer. If a customer’s line is not working, the customer can contact the operator through a digital channel, send a photo of a router or its indicator lights, and receive instructions. The solution may be to reset the device, check a connection, or perform another standard step.
Ražem says Infobip has tested such scenarios with some telecom operators outside Europe, where more than 90 percent of questions were resolved by AI and a smaller share was routed to a live agent. “I think it was around 92 percent of resolved inquiries, while around eight percent went to a live agent,” he says. The figure should be understood in the context of standard telecom questions: why a phone has been disconnected, why a service is not working, how to access invoices from the past six months, or how to change a package.
AI is not used only to answer customers automatically. It can also support live agents by preparing several possible answers or solutions. The agent still communicates with the customer, but reaches relevant information faster and can focus on the actual problem. This is gradually changing contact centres, with routine questions automated and people taking on more complex or sensitive cases.
Ražem does not argue that agents have already taken over most of the work. He expects that shift to happen gradually, first through further technological development and then through lower costs. At present, the cost of using large language models remains high and, in his view, is still partly subsidised by major model providers. The market therefore needs to reach a more stable level before broader use can expand without uncertainty over the economics.
The next step may be even more profound. Ražem expects end users to have their own digital agents, a kind of digital twin on their mobile phones, performing tasks on their behalf. These agents could communicate with agents from telecom operators, banks, airlines and other large organisations. In such a world, users would not always have to move through apps, forms and call centres themselves. Their agent could talk to the provider’s agent, gather information and complete tasks that today still require multiple steps and human time.
Telecom operators are no longer using AI only for network optimisation and routine operations. They are applying it to faster product development and customer conversations. Matija Ražem says AI enables prototypes in hours, quicker market testing, and automation of many standard support questions. The next step may be a world where personal digital agents interact with agents from telecoms, banks and other large organisations.
ReversingLabs is developing agent-based approaches to security operations, while warning that the expertise needed to manage such systems will become even more critical.
Infobip has officially introduced RCS BusinessOS, an end-to-end operating layer designed to assist mobile network operators in launching, managing, and monetizing RCS for Business.
