Vestiaire Collective leads retail innovation with A/B testing and experimentation

Vestiaire Collective is the leading global platform for pre-loved luxury fashion. The company’s mission is to transform the fashion industry for a more sustainable future, through empowering its community to shop more consciously. Co-founded in 2009 by two female entrepreneurs in Paris, Vestiaire Collective is a Certified B Corporation® and is active in 70 countries worldwide. 

Driven by the philosophy ‘Think First, Buy Second’, Vestiaire Collective offers a trusted space for its community to prolong the life of its most-loved fashion pieces. The platform's innovative features simplify the selling and buying process, as well as giving its members access to one-of-a-kind wardrobes from around the world. The company boasts a curation of 5 million desirable items.

As Vestiaire Collective grew, the company recognized the need for a more unified and efficient approach to feature management and experimentation. The existing setup was fragmented, with different solutions used across various parts of the platform.

"With all the friction that we had, to [be able to] do some custom implementation or analysis, or debug potential issues with the data — that’s also time lost toward understanding if a feature is working and what we should do next," explained David Tieba, Head of Product Analytics at Vestiaire Collective.

The company faced several challenges:

• Fragmented approach to feature flagging and A/B testing
• Limited control over feature rollouts
• Manual and time-consuming experimentation process
• Lack of autonomy for experimentation teams
• Difficulty correlating feature changes with platform performance

A/B testing was primarily limited to mobile apps using Firebase, with no robust solution for web or backend testing. Analyzing experiment results involved manual data exports and analysis in Tableau, making it difficult to iterate quickly. David described the process: "We were already exporting Firebase data with a daily batch, having some tables in Snowflake and trying to connect this to front-end tracking data and analyze the result in Tableau. But it was still a bit manual." Feature flagging was fragmented, with gaps and inconsistencies across systems.

Reliance on third-party tools like Algolia for search-related A/B testing limited the team’s flexibility. David says, "We didn't have anything for web. We didn't have anything for backend. Back in time, we also had a third-party search engine, which was Algolia, and which had its own A/B testing capabilities built in. But we migrated to the in-house search system, and we also needed to have a replacement solution."

Vestiaire Collective implemented LaunchDarkly to address these challenges and create a more centralized, scalable system for feature management and experimentation. LaunchDarkly provided a single, consistent platform for feature flagging and A/B testing across all Vestiaire Collective client apps (iOS, Android, and web) and select backend services.

David emphasized the importance of this unified approach:

Vestiaire Collective integrated LaunchDarkly with several tools in its tech stack, including Datadog for monitoring and Slack for notifications. The team leveraged the LaunchDarkly data export feature to send experiment data to their data warehouse (Snowflake) for in-depth analysis.

"Data export capability was one key differentiator," David explained. "It enabled us to measure the result of the A/B test, but also to go much deeper into why we think it's working or not working, by doing a lot of breakdown and additional analysis."

LaunchDarkly enabled the team to implement phased rollouts, starting with internal testing and gradually increasing exposure to users. The platform allowed for targeting based on various factors such as geography, platform, and user attributes. David elaborated on their targeting approach, explaining that his team often targets by region and platform, and occasionally differentiates between logged-in and non-logged-in users, as well as between different app versions.

LaunchDarkly transformed the Vestiaire Collective approach to feature management and experimentation, delivering significant improvements in several areas:

• Unified framework for feature management and experimentation
• Granular control over feature rollouts
• Streamlined, autonomous experimentation process
• Increased collaboration between product, engineering, and data teams
• Real-time visibility into feature impact on platform performance

LaunchDarkly streamlined the development process, allowing developers to focus on writing code without worrying about the complexities of feature flag management. Hugo Almeida, Vestiaire Collective’s VP of Engineering, noted: “It’s easier to set up and experiment with multiple, different ways of targeting, which was not possible before. Right now, you have the context in LaunchDarkly, and then you can play with the variables in the context and targets, target specific users… you don’t need to deploy anything. It’s way more convenient than it was before.”

The experimentation team now has greater autonomy; they can make changes and run tests without relying heavily on engineering resources. The ability to gradually roll out features (and quickly roll them back if needed) has accelerated the Vestiaire Collective time-to-market for new features.

Integration with monitoring tools has improved the team's ability to correlate feature changes with platform performance. As Hugo explained, "Integrating [Datadog] with LaunchDarkly was super useful, because nowadays we can overlay LaunchDarkly activations on top of health charts and metrics that we have on a timeline. And we can easily correlate any degradation in performance with the possibility of that having been triggered by an activation or deactivation of a given flag."

LaunchDarkly also supports Vestiaire Collective AI and machine learning efforts. David shared,

The platform has been particularly useful for testing and optimizing pricing recommendations. David explained, "We are guiding the sellers to recommend what the right price is that they should set for an item to maximize the chance to sell [it] fast at the best price for them. So it's also powered by some machine learning, and we also use LaunchDarkly to do some A/B tests at the item level."

The ability to access raw data has been essential for building trust and understanding experiment results. David emphasized, "Having access to the whole level data is helping [us] to identify potential issues in terms of setup, and overall, over time, improve the setup and improve the trust."

LaunchDarkly helped Vestiaire Collective transform its strategy for feature management and experimentation, delivering significant improvements in developer productivity, release speed, and cross-functional collaboration. Using a unified platform for feature flagging and A/B testing, Vestiaire Collective is positioned to innovate faster and deliver better experiences to their customers.

Hugo summarized the impact: