DataTwin Enabled Informed Decisions through GenAI-based Business Transactions Analysis

About Customer

DataTwin's is an intelligent modern finance platform equipped with strong data management capabilities to tackle the challenges of today. It helps finance teams manage the business transaction data, and automate processes including Accounts Payable, Accounts Receivable, Fixed Assets, Reconciliations across Revenue, Collections, Taxation, Fixed Assets, Leases, etc. It also provides the Finance Leaders with the insights for them to accelerate their success and growth.

Business Problem

DataTwin finance platform used an AI-based invoice processing system to automate data extraction from supplier invoices. However, the challenge arose as the AI required manual retraining to adapt to evolving invoice formats, consuming 5 man-months for about 50,000 invoices. The rigidity of their AI model pushed DataTwin to seek ShellKode's help in building a Generative AI solution that could minimize the training times while being able to extract the data with high accuracy.

Solution

To address the challenges above, ShellKode built an integrated Amazon Bedrock driven workflow that automatically extracts relevant information from invoices, such as vendor details, invoice numbers, dates, line items, and amounts and processes it based on predefined checks & balances. The solution also helped to categorize documents based on content, type, or purpose using prompt tuning which streamlined the sorting process and improved efficiency.

  • Efficient Data Extraction with Langchain: Leveraging the Langchain orchestration tool, we achieved efficient extraction of vital details (invoice numbers, due dates, amounts, vendor info) from supplier emails and attachments.

  • Structured Data with Amazon Bedrock: Amazon Bedrock was employed to structure the extracted data into a standardized JSON format, ensuring organization and readability.

  • Robust Data Storage and Visualization: The JSON-structured data found a home in a robust database, enabling streamlined visualization, real-time tracking of invoice status, and proactive monitoring of payment due dates.

    • Intelligent Email Responses with Language Model (LLM): Harnessing the data, a Language Model (LLM) was employed to generate intelligent email responses tailored to internal agents, based on customer queries.

      • Seamless Workflow Integration: Operational efficiency was elevated through a seamless workflow integrating Langchain, Amazon Bedrock, and the LLM model, ensuring a comprehensive solution to the invoice management challenge.

      Business Outcome

      • The solution helped to achieve an 80% reduction in invoice processing time, with the ability to adapt to the changing invoice formats.
      • Automation helped their users close financial books 50% faster and also minimized errors.
      • The solution provided real-time visibility into the status of invoices, enabling better tracking, monitoring, and decision-making related to financial transactions.
      • Efficient procure-to-pay cycle: Optimize supplier payments and minimize financial risk.
      • The solution contributes to improved compliance by maintaining a comprehensive audit trail of invoice processing activities and facilitating regulatory requirements and audits.