With over 500 different products in 200+ countries, the Coca-Cola company has been known to actively invest in researches involving the integration of artificial intelligence algorithms into its processes as well as into its products (AI-powered vending machines).
These researches have propelled the company to come up with unique business strategies for gaining larger market shares and simultaneously retaining old customers. This article talks about one of such strategies used by the company to make its customers loyal to its brand through applied AI.
The Coca-Cola Loyalty campaign
The Coca-Cola company runs a loyalty campaign where consumers can get various rewards upon multiple purchases of Coca-Cola soft drinks. This incentivizes its customers to keep consuming the company’s products and not shift to other competitors’ products.
For this, the company uses a proof of purchase method where they print 14 characters product pin-code on the assorted bottle caps and on fridge-pack cardboards. Consumers can scan these pin-codes through their mobile phones to get various rewards.
Using artificial intelligence to scan pin-codes
In 2018, Patrick Brandt, the IT Director and Solutions Strategist at the Coca-Cola company appeared at the TensorFlow Dev Summit 2018 and presented a talk on how Coca-Cola has been applying AI in their loyalty campaigns.
Since Coca-Cola had been printing the codes in a format that wasn’t easily recognizable for well-known Optical Character Recognition (OCR) methods, the company had to research and develop its own deep learning algorithm for the computer vision task. The high-level overview of their approach looked like the following:
The app’s user would take a picture of the bottle cap and OpenCV is used to detect and locate the region of interest (ROI) in the image, i.e. the bottle cap. The located ROI is then cropped and normalized and character recognition is done through the help of a Convolutional Neural Network (CNN). The CNN outputs a character probability matrix and top 10 predictions are taken out. Then, the pin codes are checked for validity and if they are valid, the user gets the reward.
However, there were two main caveats involved with this approach since the deep learning algorithm had to be light-weight enough to fit on most mobile phones and had to be accurate enough to successfully scan the pin codes. This was certainly a hard task for the R&D team since they had to make sure that the CNN architecture contained a low amount of parameters that could perfectly express the training data.
Three iterations were done to find the best deep learning model
The CNN model development for the Coca-Cola loyalty campaign went through three big iterations.
On the first iteration, a deep learning algorithm was built using Binarization for image normalization but this affected the accuracy of the model as the data was lossy in nature. As a result, Best Channel Conversion was used but it too had its own problem since the model became too large to be stored for a mobile application.
On the second iteration, the previous model was completely discarded and a SqueezeNet model was developed to lower the model size by training it on fewer learnable parameters. This too had its pitfall since the model couldn’t converge due to internal covariate shift.
On the third and final iteration, SqueezeNet with Batch Normalization was used together and the model was observed to converge effectively. This final architecture produced a 5 Mb model which was a 25-fold decrease over the initial model that Coca-Cola had built and the accuracy was now over 95%!
The model worked for all kinds of images!
The model was able to perform well even when the images faced occlusion, translation, rotation and camera focus issues.
The research and development team at Coca Cola had been successful in using applied AI to gain a competitive advantage for the company in the matured market.
In Conclusion
What do you think about the above-discussed method in which Coca Cola has been using Applied AI at scale? Please let us know in the comments.
Do you want to learn Python, Data Science, and Machine Learning while getting certified? Here are some best selling Datacamp courses that we recommend you enroll in:
- Introduction to Python (Free Course) - 1,000,000+ students already enrolled!
- Introduction to Data Science in Python- 400,000+ students already enrolled!
- Introduction to TensorFlow for Deep Learning with Python - 90,000+ students already enrolled!
- Data Science and Machine Learning Bootcamp with R - 70,000+ students already enrolled!