Zalora, a fashion e-tailer operating in Southeast Asia, transformed its analytics stack to streamline reporting and make it easier to follow users “as they moved through various journeys” on its e-commerce platforms.
Principal product manager of growth and analytics May Chin told Twilio’s Signal 2022 conference that the changes had enabled an “analytics-driven culture” to develop at Zalora over the past year.
“We have seen 150 percent growth in the month-on-month internal users of our analytics platform, which goes to show you how top-of-mind analytics and metrics have become for our Zalora team members,” Chin said.
The main change is the establishment of Twilio’s Segment customer data platform (CDP) as the “single source of truth” for events data. Events in Segment’s language, are “any actions users perform”.
A year ago, events and transaction data were passed between multiple systems in the analytics stack.
This structure created “duplicate sources of truth that [could] only be fully reconciled in our data warehouse [Google BigQuery] and fully visualised on Tableau, which can be a little bit technically difficult to use because not all team members have the expertise to do.”
It also made reporting and “insights generation” challenging.
“Just to answer a simple analytics question would take a typical [staff member] hours of manual data set reconciliation, sanity checking across multiple data sources, memorisation of various database table names, and even sometimes having to scan our code base to make sure we were using the correct event names and labels,” Chin said.
There were also differences in how each system handled user identification.
“We weren’t able to truly identify and follow a user as they moved through the various journeys on our e-commerce platforms,” Chin said.
The identification issue also “led to inconsistent experiment experiences”.
“Because of the user identification mismatch, we were inadvertently re-bucketing and reallocating users to experiment variations, pre- and post-login,” she said.
“For anyone familiar with A-B testing, you would know this is a huge no-no, and inconsistent exposure to experiment variance will be very harmful to the robustness of your experimentation.”
The solution to this was to establish Segment “right in the middle of our analytics setup”, Chin said.
All events and transaction data now pass through Segment, which re-routes it appropriately to other systems.
Chin said that meant “objectively identifying what were the highest priority events that we needed to have to answer the typical analytics questions that we have in our day-to-day”, ranking them, and then having “every single funnel owner” across Zalora implement them.
“Throughout this entire process, our team was also there in the background to constantly govern and provide continuous education on the proper way to implement these events to encourage that consistency across implementation, which was important because we had multiple different teams and engineers implementing each of these events,” Chin said.
As each event became trackable through Segment, reports were iteratively built to generate insights for the business.
She said the result is a cultural change within Zalora, with a broad embrace of data analytics.