EXL Conversational BI in Action: A Global Bank’s Journey to Data Excellence

A leading global bank partnered with EXL to enable natural language data interactions, democratize access to insights, and eliminate the need for traditional SQL queries.

The Challenge

The primary challenge for this financial institution was the presence of multiple central reporting dashboards that offered limited data exploration and deep-dive capabilities. Extracting meaningful insights required navigating complex data structures, making the process time-consuming and heavily reliant on a large team of analysts. There was a need for a user-friendly data analysis and dashboarding tool that could democratize data access, allowing users without coding expertise to query data using plain English.

In addition, the client lacked a GPU-based infrastructure to host the backend, which was developed using open-source LLMs.

Solution

The client implemented EXL Conversational BI, a pre-trained solution that simplifies data analysis by allowing users to query data in plain English. Tailored to the client-specific use cases, the solution enabled faster insight generation, ad-hoc querying, and self-service BI, empowering users to address their queries instantly. By accurately translating natural language into SQL statements, the client achieved significant breakthroughs in efficiency and decision-making:

• Knowledge Graph Creation: EXL developed a knowledge graph that integrated client data spread across different tables, enabling the understanding of its context and relationships to other data by linking over 600 variables. This advanced relational inference across customer acquisition, portfolio management, performance, spending, complaints, and revenue. The knowledge graph provided a semantic, interconnected map of data, enabling deeper insights from complex datasets
• Custom RAG Framework: EXL developed a custom RAG (Retrieve, Augment, Generate) framework using a knowledge graph to accurately identify tables and variables based on user context and intent. This system resolved ambiguities from variables with similar definitions but different interpretations, enabling efficient data retrieval, contextual augmentation, and the generation of insightful, conversational responses.
• Transforming English Queries into SQL: Implemented a multi-step methodology to convert English inputs into SQL queries. The process begins by accurately interpreting user context to refine the query. The refined query is then translated into SQL using the RAG framework to identify relevant tables and variables, allowing users to ask the same question in different ways.
• Hybrid Deployment: The solution was deployed in a hybrid manner: backend Gen AI modules were hosted on EXL’s cloud, while execution occurred in the client’s environment

Outcomes

By implementing EXL’s Conversational BI solution, the client realized substantial productivity and cost benefits. The intuitive interface enabled users to pose questions in natural language, automatically generating SQL queries. These queries produced visual and tabular outputs, supplemented by insightful textual analyses. The results were significant:

$30M+
Estimated savings in expenses over three years

75%+
Reduction in time required for ad-hoc analysis

2X
Acceleration in generating insights for strategic decision-making

Enhanced democratization of insights through self-service capabilities

Improved team collaboration, leading to greater operational efficiency

EXL Conversational BI: Empowering Data Analysis and Insight Generation

EXL Conversational BI transforms the way organizations approach data analysis by putting powerful tools directly in the hands of decision- makers. As illustrated in the case study, this solution removes the barriers of technical constraints and reliance on specialized teams. Users are empowered to interact with and query data directly, leading to faster and more informed insights. This approach not only accelerates decision-making but also enhances decision quality by leveraging thorough and accessible data analysis.