Unlocking Data Accessibility: How Accenture Leverages Generative AI RAG for Finance Teams

[akshay_bs]

The Challenge

Finance teams need quick access to crucial information on best practices, accounting guidelines, compliance procedures, and policy requirements. However, this information is often scattered across various platforms and formats, making retrieval difficult, especially under time constraints.

The challenge of navigating this complex landscape can lead to slower decision-making and inconsistencies in global policy application. As teams grow and technologies evolve, ensuring that everyone has the latest material can become a burden on experienced team members.

A streamlined, accessible, and consistently updated knowledge base is needed to empower finance team members. A solution that consolidates information and enhances accessibility through intuitive, everyday language searches would alleviate information overload and enable efficient, data-driven decision-making across Accenture's global finance operations.

The Solution

Accenture has implemented an innovative pilot solution leveraging Generative AI, specifically Retrieval-Augmented Generation (RAG) technology. This advanced approach integrates Large Language Models (LLMs) with information retrieval systems, enabling finance teams to access and interact with vast amounts of data using natural, everyday business language.

At the core of this solution is the utilization of vector embeddings and similarity search within a vector database. By employing chunking strategies, large documents and datasets are broken down into manageable pieces, or "chunks." Each chunk is then converted into a high-dimensional vector using embedding functions, which preserve the semantic similarity of the content. These vectors are stored and managed within the SAP HANA Cloud vector engine, which supports create, read, update, and delete (CRUD) operations using SQL.

When a user inputs a query, the system converts this query into a vector representation. Through similarity search, the system compares this query vector against the stored vectors to find the most relevant chunks of information. This process enables the retrieval of highly pertinent data from across the knowledge base, which the LLM then incorporates to generate accurate and contextually enriched responses.

Key Features of the Solution:

• Semantic Search and Similarity Matching: By understanding the semantic meaning behind queries and documents, the system can retrieve information that is contextually relevant, even if the exact keywords do not match.
• Natural Language Processing (NLP): Users can interact with the system using natural language, making the search process intuitive and user-friendly.
• Consistency and Up-to-Date Guidance: The solution taps into an agreed-upon "best practice knowledge and rules database," ensuring that the guidance provided is consistent and reflects the most current policies and procedures.
• Empowering New-Joiners & Experts Alike: By providing quick and accurate answers, the tool empowers users to be more self-sufficient, reducing the need for constant support from seasoned Finance professionals.

•  Enhanced Onboarding Experience: New team members access information seamlessly, facilitating quicker integration and contribution to finance operations.
  
  

To consolidate disparate information sources and enhance accessibility through intuitive search capabilities using everyday business language, Accenture explored and piloted an innovative solution utilizing SAP HANA Cloud's Vector Database (Vector DB) offering.

Exploring the SAP HANA Vector DB Offering

At the heart of Accenture's solution lies the integration of Generative AI with advanced data retrieval systems. Specifically, the team leveraged Retrieval-Augmented Generation (RAG) technology, which combines Large Language Models (LLMs) with information retrieval mechanisms. This integration enables finance teams to interact with vast amounts of data using natural language, making the search process more intuitive and efficient.

The SAP HANA Cloud Vector DB plays a crucial role in this setup. By utilizing vector embeddings and similarity search capabilities, the system transforms large documents and datasets into high-dimensional vectors that capture the semantic essence of the content. This process involves:

1. Chunking: Breaking down large documents into manageable pieces or "chunks" to facilitate efficient processing. In our case the chunk size of 1000 words was found to be effective. You may need to experiment a bit to find the size that works for you.
2. Embedding: Useinf the ADA models from openai we converted each chunk into a numerical vector representation using embedding models that understand semantic relationships.
3. Vector Storage: SAP HANA Cloud Vector DB was leveraged, which supports comprehensive create, read, update, and delete (CRUD) operations through SQL
4. Retrieval: This is where similarity search plays an important role, we needed to use a reranking step before the final results were generated for the user.

Key Technical Considerations

When designing this solution, these were a few critical factors to work through:

• Security: It was an important requirement to maintain security between the documents of different teams which is established by ensuring different vector tables for different teams.
• Vector DB – Auto Updates: Automated chunking process and update of Vector DB tables based on user uploaded files on SharePoint. No IT administration is needed, and the business team is directly responsible for maintaining the document store through this process.
• Authorization: The system validates if a user belongs to a specific team and ensures only the vector tables which contains documents for that specific team is used for retrieval
  
  

Responsible AI, Data Privacy and Security are at the core of how Accenture designs Generative AI solutions. Ensuring that sensitive financial data remains secure within the vector embeddings that are required robust access control mechanisms.

Conclusion

Accenture's innovative Generative AI solution with Retrieval-Augmented Generation revolutionizes financial data access by enabling intuitive and efficient natural language searches. By integrating SAP HANA Cloud's Vector Database, which uses vector embeddings and similarity search, the system ensures secure, team-specific document retrieval.

This approach not only enhances internal processes but also reduces reliance on specialized support, promoting efficiency and consistency across global finance operations. With a focus on responsible AI, Accenture ensures robust data privacy and security, making their solution both powerful and trustworthy.