Client background

The client is an innovative, global experience and sponsorship agency that creates, consults and activates advertising worldwide, covering every continent with 23 offices in 12 countries. They are headquartered in the United States and are a part of Omnicom Group, Inc.

The business challenge

 The client had a highly effective proof-of-concept machine learning model which provided key insights. However, because it was only a proof-of-concept model, they were unable to meet full production standards and obtain ongoing insights from new data. 

Strategy and solution

Baker Tilly implemented a class system to organize the proof-of-concept code base and increase readability. This class structure is framed within a fail gracefully system which outputs smart logging for high level tracking and simplified troubleshooting. Additionally, we restructured naming conventions throughout the code to match data team norms to streamline use. The proof-of-concept model was updated to meet full production standards with the following improvements: 

• Dynamic code structure for flexibility with inputs, providing the client with the ability to naturally scale the growth alongside their data 
• Elastic topic identification methodology, allowing the client the power to scale with the natural growth of their program 
• MLflow, an open-source machine learning operations tool for model tracking and deployment, giving the client a clean history of implementation and performance, while leveraging simplified deployment tools 

The client was able to meet full production standards and is saving time and money through: 

• Daily insights and analysis through a machine learning model fully in production 
• User friendly and minimal effort updates via scalable code structure 
• Robust model tracking and deployment utilizing MLflow