Background: This project was my first large-scope and leading project at Garmin. The work allow me to hone my analytics skills by applying different Machine Learning methods to generate business insights. Garmin Japan has reached out to me for a solution to analyze the newly developed web chat system to improve and enhance the existing functions that the web chat provides. The aim was to replace a 3rd party outsource platform to cost down and maximize efficacy.

The project was separated into two phases from Developing to Production:
- Cooperated with Garmin Japan’s customer support team, sales team, IT team to discuss the analysis metrics and algorithms on Natural Language Processing (NLP). The dashboard scope is across log data, text data, network graph, and tree structure data. Conducted weekly meeting with stakeholders to validate weekly progress.
- Automate and designed the Extract, Transform, Load (ETL) process to update the dashboard twice per day. Transformed raw data from MySQL to MariaDB for log data manipulation, processed python scripts for text data mining, and updated them through Airflow. Embedded Tableau dashboard into Microsoft SharePoint application for internal access.

Motivation: The long lasting issue on manpower shortage within customer support function is especially severe in developed country such as Japan. It is simple to find a modern solution such as chatbot, yet how to analyze and how to interpret the data is something that worth diving in.

Project Timeline

Check out my Github

The top 5 most desirable features within an individual health monitor dashboard is:

• · Overall usage calculated by day/ month/ year
• · Overall chat amount/ unique user/ email user
• · Overall chat response rate/ problem solved rate
• · Problem solved rate by individual chat
• · Drop-out ratio by individual chat
• · Tree structure analysis on question layer
• · Keyword analysis using Part of Speech Tagging and wordcloud
• · Network Graph analysis based on dynamic TF-IDF value

Outcome