Executive Summary
Coupang wanted to implement an easy-to-use automated solution for time series forecasting. Producing accurate forecasts required the implementation of the latest forecasting methods and a complete data ingestion service. Prophecy was the perfect match for Coupang’s forecasting needs. Coupang needed Prophecy to be easily deployable within their AWS environment. Utilizing AWS services such as EMR and QuickSight, Prophecy delivered fast, accurate results that Coupang’s users could visualize quickly to perform projections and analysis
About the Customer:
Coupang is a large South Korean e-commerce retail company that utilizes modern technology to provide quick delivery and services to its customers. Coupang combines impressive physical infrastructure, such as distribution centers, with state-of-the-art logistics technologies.
Customer Challenge:
Coupang wanted to create a modern forecasting solution for their web traffic orders that improved forecasting time and allowed the company to respond to quick changes in customer trends better. Coupang could potentially lose out on customer demands, over or understock their inventories and see slowdowns in their supply lines if this challenge wasn’t addressed.
About the Partner:
People Tech Group is a full-scale Advanced Tier Consulting Partner with proven expertise in DevOps projects, cloud application development, and infrastructure migrations. With Elliptica, People Tech Group has developed modern Machine Learning and Deep Learning capabilities that give it a unique position. that give it a unique position.
Why AWS:
Coupang was already working within the AWS environment, and as such, AWS was the natural choice for developing a forecasting solution. AWS provided a variety of necessary services which allowed the integration of Prophecy to create an easy to deploy, easy to use forecasting solution. AWS also provided support that helped our team better understand and create a resilient architecture.
Why the Customer Chose the Partner:
People Tech Group has a proven track record of creating and implementing a successful solution with AWS. Coupang chose to work with People Tech because of their longstanding relationship with AWS, their expertise in AWS services, and their knowledge of time-series forecasting.
Engagement Scope
Coupang’s goal was to incorporate a new forecasting solution into their workflow. Part of Prophecy’s engagement scope was to address the following challenges:
Solution Requirements:
- Prophecy must be able to process upwards of 100g of data in one job run
- Training of Machine Learning algorithms can take no longer than 24 hours
- Prophecy job must follow all Coupang practices for security
- Availability and Disaster Recovery must be implemented
- Training jobs must be easily monitored, and support be readily available
- Visualizations of forecasts must be readily available
Proposed Solution
Prophecy’s solution leveraged quick deployment of the following resources using AWS CDK:
- Step Functions for monitoring and job orchestration
- S3 for initial data storage and data staging
- AWS Glue for ETL and data merging, as well as serverless, scalable compute
- EMR provided easy to configure clusters to perform AutoML on a sample of the submitted data, allowing Prophecy to select the best algorithm for the given data quickly
- Provision of an EMR cluster to be optimized for the chosen ML algorithm
- Predictions, once complete were, saved back to S3 for quick visualization in QuickSight
Solution
Coupang considered the implementation of Prophecy a success in creating a forecasting solution that was scalable, resilient, and easy to use for business intelligence users. The teamed learned several valuable lessons, not only in creating modern forecasting algorithms, but also in integrating them within an existing AWS framework. With Prophecy, we were able to achieve a 40% decrease in training time for the machine learning models when compared to earlier implementations. For neural networks, this decrease in time was even more drastic, improving training time but up to 50%. The accuracy of predictions also improved, resulting in up to 30% more accurate results
then competing services.
Key Takeaways
- Prophecy increased accuracy of timeseries forecasting by up to 20% compared to previous solutions
- Allowed for easy connectivity within Coupang’s existing AWS infrastructure
- Jobs can quickly be restarted in case of disaster due to AWS fault tolerance
- Easily compare forecast predictions with incoming results using QuickSight
To know more about People Tech’s Prophecy solution, visit http://peopletech.com/prophecy/
