What Should Telecom Companies Automate First with AI? 

Not every support request requires the same level of AI. 

“What roaming plans do you offer?”, “Is roaming enabled for my number?”, and “Please enable roaming for me”—these requests may seem similar, yet from a technical standpoint, they are three completely different tasks. The answer to the first question is the same for every customer. The second requires retrieving current data from a specific account. The third, on the other hand, involves actively making changes to that account.

Every request in a telecommunications company’s support queue falls into one of these three categories, which is the reason why automation projects often fail to meet expectations when these categories are conflated. For now, AI in customer support primarily works by taking on part of the workload rather than replacing staff: in a survey from Gartner conducted in October 2025 among 321 customer service executives, only 20% reported staff reductions due to the implementation of AI. But 55% noted that staffing levels remained the same, despite an increase in workload.

This kind of “absorption” occurs only when the appropriate level is automated, and the telecom industry is a good test case, since in a single support queue, all three levels are regularly combined within a single call.

Level 1: Knowledge base 

To answer the question “What roaming plans do you offer?”, there’s no need to log into the customer’s account. The answer is the same for everyone. This is the level of automation with the lowest risk and, as a rule, the best place to start, since it reduces the load on the queue without affecting the billing or service delivery systems.

But if this level doesn’t work properly, everything that follows becomes more complicated. A good knowledge-based chatbot resolves such issues in a single message exchange—without redirects and without phrases like “Let me check that for you.”

Level 2: Real-time lookup 

“Is roaming enabled on my number? ” This question looks almost the same as the previous one, but now the AI must call a function, retrieve data for a specific account, and return an answer that wasn’t pre-programmed. The same logic applies to questions like “How much data do I have left?” or “Why isn’t my internet working right now?”

Some of these inquiries require more than one call. The question “Why isn’t my internet working right now?” may require a series of steps: checking the account status, then checking for network outages, followed by checking the status of the device or service, and then running a series of diagnostics, before enough information is gathered to provide the customer with a specific answer. Properly organizing this sequence in the correct order constitutes the bulk of the engineering work at this level. Usage data, billing discrepancies, network status: all of this must be returned as concrete answers, not as a link to check the app.

Level 3: Action and escalation 

“Enable roaming for me” is the level that companies approach with the greatest caution. At this level, AI can trigger predefined actions using the company's existing APIs and workflows: activate a SIM card, change a rate plan, and suspend or block the SIM. This is also the level with the greatest return on investment, since it goes beyond simply describing the task and actually carries it out.

Safety measures are especially important here. Changing a rate plan involves billing and user rights. Reporting a lost phone affects account security. Identity verification and confirmation steps must be part of the workflow from the very beginning.

How this plays out in practice 

“Why is my internet not working?” First, the support agent checks the network status for the specific account or address. Known outage → this is reported immediately, along with the current status, without the need to follow a standard troubleshooting procedure. No outage → a diagnostic algorithm is launched. Technical assistance is required → in the next step, a support ticket is opened or a service visit is scheduled.

“Why is my bill 20 euros higher this month?” The agent retrieves the current and previous bills, the rate plan, additional services, and roaming charges, and then explains the specific difference. If something looks suspicious, the assistant escalates the issue to a higher level rather than trying to guess.

“My eSIM isn’t working.” The agent checks the activation status and settings, then either provides specific instructions, initiates the reactivation process, or escalates the issue to a higher level.

“I lost my phone.” This scenario requires the utmost caution. First and foremost, identity verification must be conducted through the carrier’s existing process, and only then can the agent proceed to block the SIM card or initiate the replacement process. Skipping this step poses a risk of fraud, not a convenient way to streamline the process.

Where the AI sits 

AI does not require unrestricted access to the BSS or OSS. Tools such as Echomind operate through predefined functions and APIs provided by the telecommunications company. The model determines when a particular function is needed, but the function itself defines what data can be retrieved or what actions can be performed.

For example, AI can call the check roaming status or enable roaming functions, but it does not have unrestricted access to the customer’s account or the billing database.

Where to Start

Start with Level 1: low risk, immediate reduction in the volume of inquiries. 

Level 2: as soon as the knowledge base (KB) begins to operate consistently, since it is precisely when searching for information in real time that customers notice the difference. 

Level 3: last, with built-in verification and confirmation mechanisms from the very beginning, since it saves the most time but causes the greatest damage if an error occurs.