Results
The complexity of analyses was reduced by a factor of ten. Analyses are now faster, more consistent, and the external tool is no longer needed (saving costs). Business users can explore data and generate insights themselves. But Bol.com is always moving forward. Engineers quickly replaced the in-house Kafka solution with RudderStack, proving that data work is never really finished.
The Process
Analyze
The internal tool had much more information than the external one, but more is not always better. Too much detail made it hard to see the big picture. The goal was to simplify the tool so that analyses could be faster and easier.
Strategy
We focused on two main areas: user behavior and financial results. Together with internal stakeholders, we designed a modular structure and built a proof of concept to analyze A/B tests thoroughly.
Execute
We broke the huge amount of data into different modules using dbt and BigQuery. We created a clear data flow, making A/B tests faster, more consistent, and more accurate. We introduced data tests and streamlined the entire process. Dashboards were built to work on this clean, organized data. The biggest challenge was the sheer volume: 100TB per analysis was not unusual.
Maintain & Scale
The system was built for a production environment to run permanently and automatically, no manual data input (very handy with 100TB!). We organized internal knowledge-sharing sessions and discussed successes and improvements with business owners, data owners, analysts, and developers.
The Technology
What the client says
"What really stood out was Floris' technical knowledge, his ability to easily connect with the relevant people in many places within our organization, and the enthusiasm he brings on everything data-related."
"Floris really makes it easy to understand complex data topics and brings a critical mind to the job."