Pre-Code Planning Stopped Me From Getting Stuck on a 3-Hour ETL Pipeline

I recently finished building a multi-channel data pipeline to solve a specific problem: consolidating fragmented monthly sales data from an in-store till, Uber Eats, and Deliveroo into a single master source for Looker Studio.

My Dashboard

The technical work took me slightly under 3 hours to build. But the biggest takeaway for me wasn't the speed, it was the shift in how I approached the build.

The Problem with "Figuring it out as you go"

On previous projects, my default habit was to open a blank IDE and just start typing. I’d try to figure out the data schemas and transformations while mid script.

When I did that, I ended up hitting a wall. I’d confuse myself, lose track of the data flow, and get stuck in a loop of debugging things I hadn't fully defined yet.

What I Did Differently This Time

Before touching a single line of Python for this dashboard, I forced myself to properly define the parameters of the project in a document:

• The Exact Inputs: Mapping out the column mismatches between the three different CSV exports (Till vs. Uber vs. Deliveroo).

• The Exact Transformation Steps: Deciding how to handle commission deductions and currency formatting before writing the cleaning functions.

• The Destination: Defining how the final master source needed to be structured for Looker Studio to read it cleanly.

The Result

Because the blueprint was already written, the actual coding process was just pure execution. I stayed hyper-focused for the full 3 hours because there was zero guesswork involved.

It showed me that my biggest bottleneck usually isn't my technical ability or my understanding of code; it's the framework I use to approach a problem. Spending time upfront to structure the data pipeline logically turned the coding phase into a straightforward task rather than a guessing game.

For anyone else working through data projects or personal builds: if you find yourself constantly hitting mental blocks mid code, try closing the IDE and explicitly defining the data flow first. It saves a massive amount of cognitive energy.