Thursday 13 March 2025
LinkedIn’s experiment platform has transformed how the company evaluates new features and products, allowing it to make data-driven decisions more quickly and efficiently.
For years, LinkedIn had been struggling to conduct reliable online experiments, known as A/B tests, for its enterprise-facing features. These tests involve randomly assigning users to different groups, with one group receiving a new feature or design change and the other group serving as a control. By comparing the results, companies can determine whether the changes have a positive impact.
However, LinkedIn’s hierarchical entity relationships – think of them like layers of complexity within its products – made it difficult to run these experiments effectively. The company has many enterprise customers who purchase its products under contracts or accounts, and each contract or account contains multiple seats or profiles. This meant that traditional A/B testing methods couldn’t be applied directly.
To overcome this challenge, LinkedIn developed a new experiment platform, known as EEP, which takes into account the hierarchical entity relationships within its products. The platform uses a taxonomy-based design setup, refining analysis methodologies and implementing advanced variance reduction techniques to ensure reliable results.
A key feature of EEP is its ability to detect Sample Size Ratio Mismatch (SSRM), a phenomenon where the number of units in one group differs significantly from the other, affecting the accuracy of the results. The platform uses a two-level SSRM detection methodology, operating at both randomization unit and analysis unit levels, to identify and mitigate this issue.
EEP has had a significant impact on LinkedIn’s experimentation process, allowing it to make data-driven decisions more quickly and efficiently. The company is now able to run thousands of experiments per quarter, with each experiment lasting only two to four weeks. This has streamlined the post-review and quality check process, enabling product managers to self-serve and expedite the business decision-making process.
LinkedIn’s EEP platform is an example of how a well-designed experiment platform can transform a company’s approach to experimentation. By acknowledging and addressing the complexities within its products, LinkedIn has been able to make more informed decisions, drive innovation, and ultimately improve customer experiences.
Cite this article: “LinkedIns Experiment Platform Revolutionizes Data-Driven Decision Making”, The Science Archive, 2025.
Experimentation, Linkedin, A/B Testing, Eep, Experiment Platform, Product Development, Data-Driven Decisions, Hierarchical Entity Relationships, Sample Size Ratio Mismatch, Taxonomy-Based Design
