Uploaded July 2026 | Updated September 2026, 2 weeks ago
97% Data Accuracy, 3x Cheaper, and 10x More Scale: Why This Engineering Team Dumped Google Analytics.
When Delivery Hero reached the architectural limits of Google Analytics (GA) - facing data latency, strict event caps, and GDPR compliance bottlenecks - they didn’t just migrate to GA4. They built their own real-time tracking engine from scratch.
In this InfoQ talk, Alina Krasavina (Engineering Manager at Delivery Hero) pulls back the curtain on how her team deprecated GA to build an internal user-tracking infrastructure using GCP, BigQuery, and Pub/Sub. She breaks down the exact MVP strategy, architectural optimizations, and data governance frameworks that allowed a custom, in-house solution to out-perform Google's own platform.
⏱️ Video Timestamps (For Navigation)
00:00 - The Problem: Why We Deprecated Google Analytics
01:14 - GA Limits: Universal Analytics Migration, GDPR, & Real-Time Data Gaps
02:08 - The Architecture: Building a High-Scale Tracking MVP with BigQuery & Pub/Sub
03:22 - Order Match Rates: Defining the True Metric of Data Quality
04:14 - The High Cost of Parallel Testing (How to Remove Google Tag Manager Safely)
05:01 - Why We Chose a Custom SDK Over Open-Source (The Snowplow Dilemma)
05:48 - Key Takeaways: Surviving Peak Loads & Progressive Rollouts
06:45 - The Post-MVP Evolution: Overcoming Out-of-Memory (OOM) Pod Restarts
07:34 - Automating Data Governance: Solving the "String Null" Problem with Schema Code-Gen
08:45 - Cost Optimization Hacks: JSON Storage and On-Demand GCP Nodes
09:50 - Building Non-Blocking Mobile SDKs & Event Prioritization
11:02 - The Results: 97% Accuracy, 10x Load Capacity, and 3x Cost Reductions
12:20 - Q&A: Looker Studio Dashboards & Synchronous Fallback Solutions
🔗 Transcript & slides available on InfoQ: bit.ly/3SHkZf2
#SoftwareArchitecture #DataEngineering #BigData #GoogleAnalytics #SystemDesign
97% Data Accuracy, 3x Cheaper, and 10x More Scale: Why This Engineering Team Dumped Google Analytics.
When Delivery Hero reached the architectural limits of Google Analytics (GA) - facing data latency, strict event caps, and GDPR compliance bottlenecks - they didn’t just migrate to GA4. They built their own real-time tracking engine from scratch.
In this InfoQ talk, Alina Krasavina (Engineering Manager at Delivery Hero) pulls back the curtain on how her team deprecated GA to build an internal user-tracking infrastructure using GCP, BigQuery, and Pub/Sub. She breaks down the exact MVP strategy, architectural optimizations, and data governance frameworks that allowed a custom, in-house solution to out-perform Google's own platform.
⏱️ Video Timestamps (For Navigation)
00:00 - The Problem: Why We Deprecated Google Analytics
01:14 - GA Limits: Universal Analytics Migration, GDPR, & Real-Time Data Gaps
02:08 - The Architecture: Building a High-Scale Tracking MVP with BigQuery & Pub/Sub
03:22 - Order Match Rates: Defining the True Metric of Data Quality
04:14 - The High Cost of Parallel Testing (How to Remove Google Tag Manager Safely)
05:01 - Why We Chose a Custom SDK Over Open-Source (The Snowplow Dilemma)
05:48 - Key Takeaways: Surviving Peak Loads & Progressive Rollouts
06:45 - The Post-MVP Evolution: Overcoming Out-of-Memory (OOM) Pod Restarts
07:34 - Automating Data Governance: Solving the "String Null" Problem with Schema Code-Gen
08:45 - Cost Optimization Hacks: JSON Storage and On-Demand GCP Nodes
09:50 - Building Non-Blocking Mobile SDKs & Event Prioritization
11:02 - The Results: 97% Accuracy, 10x Load Capacity, and 3x Cost Reductions
12:20 - Q&A: Looker Studio Dashboards & Synchronous Fallback Solutions
🔗 Transcript & slides available on InfoQ: bit.ly/3SHkZf2
#SoftwareArchitecture #DataEngineering #BigData #GoogleAnalytics #SystemDesign
