Data Engineering for Growth Analytics — Part 2: Attribution Modeling with GA4 Data

​In Part 1, we threw out the idea of rebuilding the GA4 interface in SQL. Instead, we designed what actually matters: clean anonymous entities and structured touchpoints from raw event data. All on the whiteboard, no slides.

​

​Now here's where it gets interesting. In Part 2, we take those building blocks and answer the question everyone actually cares about: where are my conversions coming from?

​

​Here's the thing — most attribution setups I see in the wild are either blindly trusting Google's default channel grouping or building overly complex multi-touch models that nobody on the marketing team understands. Both miss the point.

​

​What we'll design together:

​

• ​A channel classification model that you actually control — no black-box defaults

• ​First-touch and last-touch attribution from the touchpoint model we built in Part 1

• ​Why the attribution waterfall matters: click IDs > UTMs > referrer > direct — and how to think about that priority

• ​Where GA4's built-in session scoping breaks down — and when you need to roll your own

• ​When multi-touch makes sense and when it's just noise

​Same format as Part 1: mostly whiteboard design with some SQL to make it concrete. The goal is that you walk away understanding the architecture — generating the queries from there is the easy part.

​

​You don't need to have attended Part 1 to follow along, but it helps. The entity and touchpoint models from that session are the foundation we'll build on here.