thredUp leveraged Harness Feature Management & Experimentation to enhance feature delivery and enable real-time personalization during high-traffic periods like Black Friday. The company used targeting rules and feature flags to dynamically serve unique promotions based on customer attributes and machine learning scores. Split's backend SDKs integrated seamlessly with thredUp's systems, enabling backend A/B testing and collaborative QA where features could be deployed but hidden until fully ready.
The Challenge
Outdated, manual feature rollouts
thredUP initially relied on an in-house feature targeting system to roll out new features gradually. However, the system presented challenges:
- Manual Processes: Rollouts required engineers to whitelist accounts or use user ID-based methods.
- Slow Propagation: Updates took minutes to propagate across their distributed systems, disrupting the user experience.
- Limited Flexibility: Scaling personalization or A/B tests in real-time was nearly impossible.
With hundreds of thousands of daily visitors, thredUP needed a robust, scalable solution to improve rollout speed, enable real-time personalization, and streamline A/B testing.
The Solution
Adopting a unified platform for rollouts and experimentation
By adopting Split, thredUP gained the ability to:
- Execute Phased Rollouts: Engineers could roll out features incrementally with instant changes, reducing risk and improving customer experience.
- Personalize Offers in Real Time: During Black Friday, Split enabled thredUP to dynamically serve unique promotions and experiences based on customer attributes and machine-learning scores.
- Streamline A/B Testing: Split’s backend SDKs (including Ruby) integrated seamlessly with thredUP’s systems, enabling robust tests without manual data tracking.
- Facilitate Collaborative QA: Features could be deployed to production but remain hidden, allowing team-wide QA before customer release.
Results
Real-time promotions on Black Friday
thredUP used Split in innovative ways during Black Friday, dynamically targeting promotions for customers based on attributes like machine-learning scores and date. Promotions and content were updated automatically through Split’s logic, ensuring seamless and personalized experiences for all customers.
“You can pass attributes into Split for targeting... We returned not ‘on’ or ‘off,’ but a promotion code. This allowed dynamic handling of customer-facing content based on real-time conditions.” – Dan DeMeyere, VP Engineering, thredUP
Backend A/B testing at scale
thredUP replaced their existing A/B testing tools with Split for its backend testing capabilities. Split’s decoupling of data gathering and feature serving simplified the process, allowing the app to:
- Focus on customer treatment assignments
- Seamlessly integrate analytics with tools like Redshift and Looker
“Our web apps just care about what customer is in what treatment for what feature. With Split, data integration just happens for free.” – Dan DeMeyere, VP Engineering, thredUP
Collaborative QA and faster deployments
Split enabled thredUP to test features in production environments without exposing them to customers. This allowed team-wide QA collaboration, involving not only engineers but also customer service, product teams, and more.
“Split gives us confidence to ship features to production and make them accessible only to internal employees for testing.” – Dan DeMeyere, VP Engineering, thredUP
The FutureFocus on core competencies with Split
By choosing Split over building an in-house solution, thredUP avoided diverting engineering resources from their core mission.
“If we built this in-house, it would have required a whole team. Split had everything we needed out of the box—developer-friendly, intuitive UI, and robust data capabilities.” – Dan DeMeyere, VP Engineering, thredUP
With Split, thredUP continues to innovate, scale its web app, and enhance the shopping experience for its customers.
