Fueling Efficiency From Pump to Profit

Overview

Pilot Company built a GenAI-powered document processing solution on the Databricks Data + AI Platform (using Agent Bricks) to automate handling of handwritten fuel bills of lading across over 300 document formats. Using a multi-modal few-shot approach with fuzzy matching, Pilot achieved 98.6% accuracy, a 99% reduction in manual processing, and a 90% decrease in overall processing costs.

The challenge

Pilot Company faced major inefficiencies in its fuel billing process, with truck drivers uploading handwritten delivery forms that required time-consuming, error-prone manual data entry. Early attempts to automate the process with OCR and RPA failed to handle the complexity and variability of over 300 document formats.

The solution

With the Databricks Data + AI Platform, Pilot built a GenAI-powered document processing pipeline using Agent Bricks, progressing from classic OCR (76.2% accuracy) to few-shot prompting (84.7%) to a multi-modal few-shot approach feeding the model actual photos (94.3%), then implementing fuzzy matching against Delta Lake reference tables to boost accuracy to 98.6%. The team used the AI Gateway to connect to foundation models hosted in Amazon Bedrock, MLflow and MLflow Tracing to manage prompt versioning and model evaluation, Unity Catalog for governance and access control, and Monte Carlo to monitor the quality of vectorized document data and detect schema drift.

Reported business value

The system delivered a 99% reduction in manual document processing and a 90% decrease in overall processing costs, with 98.6% accuracy accelerating revenue recognition. The GenAI app is delivering value in the low seven figures through a combination of cost savings and increased revenue.