Skip to main content

Meet Raiff #7: How to talk to Tabular Data – a chatbot case study from Raiffeisen Tech and the RBI Group

Large Language Models (LLMs) are playing an increasingly important role in business, yet their practical application within corporate systems remains limited. The most common approach — Retrieval-Augmented Generation (RAG) — works perfectly for processing documents, but in many organizations, key data exists in tabular form. So, how can we enable natural language queries for such data? Here are the key takeaways from the latest Meet Raiff #7 session.

  • By Wojciech Nowak
  • Meet Raiff

Artificial Intelligence as a key driver of transformation

During the seventh Meet Raiff session, organized in collaboration with Sages, Wojciech Nowak – GenAI Tech Lead at Raiffeisen Tech – presented a practical approach to using Large Language Models (LLMs) for working with tabular data. His talk was framed within the broader context of Raiffeisen Bank International Group’s strategy, which views AI as a crucial catalyst for organizational transformation.

Wojciech Nowak emphasized that RBI has implemented an AI development program based on two pillars:

  • the AI Center of Excellence, and
  • pioneering projects that allow the organization to test new solutions and observe their real impact on daily operations.

Sales assistant supporting CRM and productivity

The project presented during the session featured a chatbot acting as a sales assistant. Its role is to deliver insights and analyses based on business data — without requiring users to know SQL or database technologies.

The solution operates on CRM data and sales productivity metrics, helping employees in their everyday decision-making processes. Data access is enabled through queries automatically generated from natural language questions. The LLM analyzes the user’s intent, interprets the context, performs semantic translation, and ultimately generates SQL queries, which are then executed on real data sets.

Agent-based architecture

At the heart of the solution lies an agent-based architecture. The agent not only understands the user’s question but also plans the path to the answer. As part of this process, the agent:

  • identifies the data needed to provide the answer
  • breaks down complex questions into smaller subproblems
  • selects the right processing tools
  • generates and executes SQL code
  • combines and interprets query results
  • uses conversational context memory

The project was designed to be scalable. The architecture allows new databases and functionalities to be added without rebuilding the entire system. The agent acts as an orchestrator, autonomously deciding which tools to use and in what order.

MVP as a testing ground

The presented solution was developed as an MVP with a clear goal — to test whether LLMs can safely and efficiently handle tabular data used in sales operations. The experiment aimed to validate both the technical feasibility and business value of the tool.

The results demonstrated that the solution significantly lowers the entry barrier to data analytics and allows users to obtain business-critical insights faster, particularly in the context of customer relationship management.

Missed the event? Watch the recording!

Watch the recording and find out how to make LLMs “talk” to tabular data.

More about Meet Raiff

Meet Raiff is a series of meetups where Raiffeisen Tech experts share their knowledge in IT and banking. The events are organized in collaboration with Sages. If you missed any of the previous meetups, you can catch up by watching the recordings on our YouTube channel.