How Generative AI is Transforming Business Intelligence: Insights from Baidu’s ChatBI

This excerpt is taken from Baidu Technology Salon Issue 115, titled “Data Middle‑Platform in the AI Era – Exploring the Data Middle‑Platform”. It analyzes the business value and technical practice of combining Business Intelligence (BI) with generative models, sharing insights from three perspectives: technology trends and business demand, system design of Baidu’s ChatBI, and practical implementation experiences.

Main contents include:

BI technology development and new opportunities brought by large models

ChatBI design philosophy and platform introduction

Technical secrets behind ChatBI

Landing effects

BI Technology Development and New Opportunities from Large Models

1. Technical Perspective

From a technical viewpoint, a new technology becomes a trend when it achieves widespread accessibility, allowing more people to use it at lower cost and generate greater value. The evolution of BI follows the same principle.

We review the stages BI has undergone over the years:

First Stage – Report‑oriented BI Products

With the emergence of big‑data technologies such as HDFS and MapReduce (MR), early BI products required custom development: analysts or business owners defined data needs, and data engineers wrote extraction code. This resulted in high development cost, long cycles, and limited scalability.

Second Stage – Self‑service BI Products

Recent advances in hardware and query technologies (e.g., MPP architectures, vectorization, in‑memory processing) have improved data extraction speed by more than tenfold compared to the MR era. This performance boost enables dynamic queries on wide datasets, reducing reliance on data engineers. Users can now perform self‑service, visual queries directly on BI platforms, making BI technology more pervasive.

Third Stage – Intelligent BI Products

The rise of large‑model AI introduces a clear trend toward intelligence. By leveraging powerful understanding and reasoning capabilities, modern BI products can hide low‑level details. Users no longer need to choose platforms, locate data sources, or write query dialects; they simply interact via natural‑language dialogue to retrieve data and generate insights, dramatically lowering the barrier to analysis.