Business Impacts
95%
String accuracy
99.6%
Character accuracy
5MB
Model file size
Customer Key Facts
- Rank : Fortune 100
- Size : 80,000+ employees
- Location : Atlanta, Georgia
- Industry : Food & Beverages
Problem Context
Coca-Cola, an American multinational beverage corporation, had launched a loyalty reward redemption program, called MyCokeRewards.com, in which customers had to manually enter 14-character pin codes printed inside their purchased beverage’s bottle caps into a mobile device.
Coca-Cola wanted to eradicate this manual data entry process and aspired to enhance the end consumer experience by building a swift image scanning solution.
Challenges
- Non-standard fonts on the bottle caps
- Synthetic data generation for varying lighting, scales, rotations, tilts, modes
- 14^20 possible character combinations
- Small model size needed to help fit on edge devices
Technologies Used
Amazon EC2
AWS Lambda
Amazon S3
TensorFlow
Python
Delivering 'Magic Wand' Proof of Purchase Experience on Phone-Based App
Solution
Quantiphi developed a custom computer vision solution which enabled customers to easily scan their bottle cap on the mobile app, accurately processing the pin code with sub-second response time for both Android and iOS devices. This saved the company millions of dollars by avoiding the requirement to update printers in production lines to support higher-fidelity fonts that would work with existing off-the-shelf OCR software
Result
- Effortless proof of purchase submission
- Seamless customer experience
- Greater cost savings
