Increasingly, customers turn to AI assistants for help before opening an app or calling their operator. They want to compare offers, change a tariff or understand a bill. Gartner expects 70% of customer service journeys to begin and be resolved through AI assistants built into mobile devices by 2028.

Operators’ own self-service channels are not keeping up. Only 14% of journeys are fully resolved through self-service, even though three in four customers try these channels first. In just one year, the use of third-party AI assistants for customer service has roughly doubled, while usage of operators’ own chatbots has remained flat since 2022.

Today, most AI assistants cannot connect directly to an operator’s systems. They can only provide information available on the operator’s public website, which may be outdated, and they cannot complete the request. Customers are then sent back to the app, a shop or the call centre.

ACX, Agentic Customer eXperience, is Celfocus answer to this gap. It enables AI assistants, both the operator’s own, such as Vodafone TOBi, and those customers already use, such as Gemini or Claude, to carry out real requests within the operator’s systems: check a bill, change a tariff, replace a SIM or buy a new line. The assistant carries out the request, while the operator’s rules determine what it is allowed to do.

ACX does not add a new technology stack or duplicate existing business rules. It builds on the operator’s existing digital stack, reusing the services that already support channels such as the app, web, shops and call centre, and extending them to a new channel: AI assistants.

Over the past few months, the team has built a working prototype that can be demonstrated and tested with customers and internal teams. Between the AI assistant and the operator’s systems sit three layers:

1. Agentic Integration Platform: the single entry point. Every assistant connects through this layer using MCP (Model Context Protocol), the open standard that allows AI assistants to connect to external systems. The platform identifies the customer, requests an OTP code when needed, records consent and maintains an audit trail of every action. When a request requires human support, it hands the conversation over to an agent, together with the full history.
2. Agentic Journey Workflow Engine: keeping the assistant on track. Each customer journey, such as changing a tariff, is designed as a sequence of steps: identify the customer, check eligibility, show the available options, get the customer’s confirmation and submit the order. Each step is a tool the assistant can call. The engine, rather than the AI model, determines which step can come next and what the customer needs to confirm. The assistant can converse freely, but it can only take actions within the journey defined by the operator. This prevents it from improvising when acting on a customer’s account.
3. Existing digital services: already in place. These are the services that already power the operator’s app, web, shops and call centre. They connect to the operator’s BSS, including catalogue, CRM, billing and ordering systems, through TM Forum Open APIs. The AI assistant never accesses the BSS directly.

‍

‍

The prototype runs on Celfocus premises over a simulated BSS, with no client code or data. Today it covers journeys such as looking up a bill, comparing and recommending offers, changing a tariff, buying a new line with delivery, replacing a SIM and opening a support ticket.

• A channel that works better than apps and web. Customers can say what they want in their own words, and the assistant takes care of the steps. Requests that currently require a shop visit or a phone call, such as bringing a friend’s number over from another operator, can be completed online in a single conversation. Fewer assisted interactions can also mean a lower cost to serve: McKinsey reports a 35% reduction in cost to serve for a targeted telco care group when support journeys are automated end to end.

• More sales from campaigns and offers. The assistant can present an offer that matches the customer’s current plan and complete the purchase in the same conversation. McKinsey reports a 40% increase in campaign conversion when recommendations are relevant to the customer.

• Present where customers ask. When customers ask their AI assistant for a better deal, an operator connected through ACX can provide a price tailored to that customer and complete the switch within the same conversation. An operator that is not connected can only be represented by the information available on its public website, if it is surfaced at all.

• No new stack. The same services and business rules that support the app, web, shops and call centre can now support AI assistants too.

• Safe by design. Customer identification, OTP verification, consent, audit trails and handover to a human agent are built into every journey from the start.

• Working at the forefront of technology. The team works hands-on with AI agents and MCP, testing how agents behave in real scenarios. This creates opportunities for our people to work with emerging technologies and helps attract new talent.

• More sales opportunities. ACX is a new offer we can take to existing and new clients, while also creating opportunities for conversations around other Celfocus offers.

• Extending our existing projects. ACX reuses the digital services we already deliver and run for our clients, extending them into a new channel rather than competing with them.

• Reusable assets. The workflow engine, integration platform and journey designs can be built once and adapted to each operator’s rules and catalogue.

ACX was built by a small team combining architecture and engineering:

• João Colaço, Architect

• Roberto Silva, Architect

• Sérgio Sá, Architect

• Diogo Sousa, Software Engineer

• Bruna Leitão, Software Engineer

Next, we will take the prototype to customers. We are planning customer demos to validate the solution against real operator needs, find out which journeys matter most to them, and learn what they need to see before they let AI assistants serve their own customers. Their feedback will decide which journeys we build next and how we prepare security and governance for production.

If you would like to see ACX in action, or you have a client conversation where it could fit, talk to the team.