Document to Transaction Automation with Vision AI: A Case Study

Organization

A large multinational pharmaceutical company with over 200 SKUs and more than 20 locations, processing orders from over 3000 customers and stockists.

Objective

To explore a solution that automates the order entry process with minimal human intervention.

Challenge

Manual order entry for over 200 products (SKUs) across 20 locations.
Orders received in various formats (PDF, Excel, images) with inconsistent layouts (500+ formats).
Orders lacked standardized product names and SKUs, making processing difficult.
Existing OCR solutions failed to achieve high accuracy (below 50%).
Needed a solution with 99%+ accuracy to automate order entry.

Solution

Proteus implemented a custom-built Vision AI solution:

Automated learning: The system learned to recognize order formats and extract relevant data.
High accuracy: Achieved 99%+ accuracy in capturing order details.
Reduced manual work: Automated order entry process, saving significant time and resources.

How Vision AI Made It Happen

1. Legacy Data Learning

The company shared a set of Mumbai sales data with the Vision Team for AI learning. Over 300 orders were analyzed, and an AI model was formulated. The AI algorithm was designed to determine customer codes based on layout. Different format handling was implemented for PDF, text, Word, Excel, image, and PDF with image.

2. Learning Process

The Vision Team and the customer’s team uploaded the legacy data and completed the learning process in the Vision AI module. Orders from two months were analyzed to ensure that the Vision AI engine could interpret data irrespective of the format. The AI algorithm was designed to translate customer descriptions (such as case/strip) into the customer’s unit. On-site verification and enhancement of the AI model were performed to achieve 99% accuracy.

3. Successful Data Interpretation

The Vision AI Data Interpretation Module successfully achieved a 99% accuracy. Pre-trained customer orders were processed seamlessly without any intervention, and data was exported to Excel for order processing in SAP.