Enhancing Financial Accuracy with AI: LLM-Powered GL Review System

About Client

A Texas-based private firm investing in and managing properties across the U.S., Canada, and Europe to deliver long-term value for investors.

Background

  • Client’s finance team handles vast volumes of financial data every month, particularly through General Ledger (GL) entries recorded across multiple properties. These records are critical for accurate reporting and compliance, but the manual review process is time-intensive and susceptible to human error.

Challenge

Every month, thousands of GL entries flow in from systems like Yardi. Some entries inevitably end up in the wrong account or category due to:

  • Ambiguous descriptions
  • Human oversight
  • Limited contextual understanding of prior entries

Correcting these issues manually consumes valuable time and energy. Worse, manual reviews themselves can introduce new mistakes. Client wanted a smarter way to accelerate reviews, ensure accuracy, and free up their finance team to focus on higher-value work, without losing control of the final decision-making.

Our Solution

Client partnered with BizAcuity to implement an LLM-driven GL Review System that blends automation with human oversight.

Secure File Upload

  • Finance teams upload monthly GL files via a Streamlit web app integrated with Azure EntraID for secure login.
  • The system validates the file structure and ingests data into Postgres with vector embeddings, while safely backing it up in AWS S3.

LLM-Powered Analysis

  • Entries are normalized, deduplicated, and embedded for context.
  • Using AWS Bedrock, the system runs GL descriptions against historical data with models such as Cohere, Claude, and LLaMA 2.
  • The LLM flags potentially miscoded entries, suggests corrections, provides a confidence score, and explains its reasoning.

Review & Feedback Loop

  • Finance teams download a flagged-entry report (Excel/CSV) with clear recommendations.
  • Users can mark entries as “correct” or “incorrect” and re-upload annotated files.
  • This feedback is fed back into the pipeline, allowing the model to learn and continuously improve over time.

Decision Authority Stays with Finance

  • The system provides insights but never makes changes directly in Yardi.

The client’s finance team retains final control, ensuring confidence and compliance.

Outcome

The new system delivers:

  • Faster monthly financial reviews with significantly reduced manual effort
  • Early detection of misclassified entries, minimizing downstream errors
  • Cloud-native design that scales seamlessly across multiple properties
  • Continuous improvement through feedback loops that make the model smarter over time
  • Finance team retains full control, with AI acting as a supportive assistant rather than a replacement

Leave a Reply

Your email address will not be published. Required fields are marked *

BizAcuity
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.