Transforming data analysis with natural language processing for $30M savings at a global bank

A large Global Banking Institution approached EXL to converse with data in natural language and generate powerful insights.

Challenge

This financial institution’s primary challenge lay in multiple central reporting dashboards with inefficient data exploration & deep dive capabilities. Extracting efficacious insights necessitated delving into complex data structures, making this task time-consuming and often reliant on many analysts. There was a need for a user-friendly data analysis and dashboarding tool that democratized data, and enabled users without coding backgrounds to query data using plain English.

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

Solution

EXL Conversational BI is a powerful solution enabling financial institutions to conduct in-depth data analysis, all through a very simple and intuitive interface that allows users to use English as a coding language. For this global banking client, the solution empowered them with faster insight generation, ad-hoc querying capability, and self-service BI. By leveraging solution’s accuracy in translating natural language queries into SQL statements, the client achieved significant breakthroughs:

Conversational BI

Knowledge Graph Creation: Developed a knowledge graph integrating banking and client domain expertise, encompassing 600+ variables across multiple tables. This enabled relational inference spanning customer acquisition, portfolio management, performance, spend, customer complaints, and revenue.

Custom RAG Framework

Custom RAG Framework: EXL created a custom, robust RAG Framework utilizing the knowledge graph to precisely identify tables and variables based on user context and intent. This addressed ambiguity caused by variables with similar definitions but different interpretations, and distinct definitions but similar interpretations.

Flexibility

Flexibility: Implemented multi-step methodology to convert English inputs into SQL queries. The first step involves accurately interpreting user context to augment the query. Subsequent steps focuses on translating this refined user query into SQL leveraging the RAG framework to identify the tables and variables. This approach empowers users to pose same question in different ways.

Hybrid Deployment

Hybrid Deployment: The solution was deployed in a hybrid way. Backend Gen AI modules were hosted on EXL cloud while execution happened in client’s environment.

Outcomes

By implementing EXL’s Conversational BI solution, the client achieved significant productivity and cost benefits. The user-friendly interface allowed users to ask questions in natural language, triggering the automatic generation of SQL queries. Upon execution, these queries yielded visual and tabular outputs, accompanied by insightful textual analyses. The client achieved the following:

  • Estimated $ 30 M+ saving in expenses over a period of 3 years
  • 75%+ reduction in the time required for ad-hoc analysis
  • 2X acceleration in generating insights for strategic decision-making
  • Facilitating the democratization of insights by enabling self-service capabilities for generating valuable insights
  • Fostering improved collaboration between teams, enhancing overall operational efficiency

EXL Conversational BI: Democratize Data Analysis and Insights generation

  • EXL Conversation BI solution democratizes insight generation by arming decision makers with a powerful tool at their fingertips. As demonstrated in the above case study, the users are no longer constrained by technical limitations or dependencies on specialized teams. Instead, they can directly engage with and query the data, achieving quicker insight generation. This not only streamlines the decision-making process, but also elevates the quality of decisions by ensuring it is based on robust data analyses.