Project Overview
Built a comprehensive data analytics environment to centrally manage and monitor in real-time the vast performance data generated across YouTube PPL, Meta ads, CRM, and other channels. The ultimate goal was to automate the data pipeline, freeing marketers to focus on strategy rather than data processing, by overcoming spreadsheet capacity limits and manual aggregation delays.
The Challenge
- Excessive Analytics Overhead: Manual downloading and merging of media data consumed 6+ hours weekly in repetitive administrative tasks
- Data Fragmentation & Performance Limits: Growing data volumes degraded Google Spreadsheet processing speed and increased I/O costs, making real-time metric monitoring impossible
- Lack of KPIs & Subjective Decisions: Marketing executed without clear data-driven KPIs, lacking objective basis for channel attribution analysis and ad budget allocation
Strategy / Solution
- BigQuery-Centered Data Warehouse: Built ETL pipelines to load fragmented marketing source data into BigQuery, enabling stable large-scale data processing with optimized I/O costs.
- ML-Based Channel Attribution Modeling: Developed machine learning models that calculate each channel's weighted contribution to branding and lead acquisition, establishing objective performance metrics.
- Visualization Dashboard Design & Development: Designed real-time monitoring screens integrated with the frontend, enabling anytime KPI access without weekly reports.
"A data pipeline isn't just storage — it's the fastest and most accurate path to transforming scattered numbers into business growth strategies."
Execution
- Tech Stack: Python (Data Collection), SQL, BigQuery, GA4/GTM
- Key Activities: Dataset configuration and model testing, query cost optimization, dashboard prototyping and KPI calculation guideline creation
Results
| Metric | Before | After | Change |
|---|
| Weekly Report Creation | 6 hours | 10 min | 98.3% Reduction |
| Data Analysis Framework | Manual aggregation | Batch/Data-driven | Max Accuracy |
| Data Processing Performance | Spreadsheet limits hit | BigQuery-stabilized | I/O Efficiency Improved |
| Decision-Making Basis | Subjective judgment | ML model weights | Scientific Budget Allocation |