Turning an intranet into an AI assistant

The context

Proximus runs a long-established intranet used by sales and support teams to answer customer questions (pricing, promotion terms, service conditions, etc.). Over the years, the intranet became so large and deep that finding the right information quickly turned into a daily struggle.

The logic

With the arrival of the first GenAI tools, Proximus realized that combining a conversational interface with the intranet knowledge base could dramatically reduce time-to-answer—without forcing employees to browse pages or guess the right keywords.

The solution

We developed MAiA, an AI-powered knowledge assistant trained on the full content of the Proximus intranet. Instead of searching and clicking through multiple pages, employees can ask questions directly in natural language, for example:

• “How much does this pack cost?”
• “What are the conditions to benefit from this promotion?”

The solution combines two complementary components:

• Azure AI Search, which retrieves the most relevant intranet pages for each query.
• A GPT-based model (currently GPT4o), which reads the selected content, extracts the key information, and generates a clear, concise answer employees can use immediately.

To drive continuous improvement, MAiA also includes built-in feedback mechanisms: agents can rate answers as satisfactory/unsatisfactory inside the application. That feedback is then processed through an established improvement loop, including human review where needed, ensuring quality stays high.

Next steps in applied excellence

With MAiA now firmly embedded in daily operations, the next phase focuses on scaling impact through two priorities:

1. Stronger feedback evaluation: improve how feedback is analyzed and translated into model and knowledge-base improvements.
2. Agentic capabilities: enrich MAiA with “agent” features that proactively suggest the right internal tool to use and make it even easier to pull the right information instantly—while keeping the experience frictionless for employees.