Building Search and Personalization at Nordstrom Rack | Hautelook
Download as PPTX, PDF1 like293 views
The document outlines the journey of a senior data scientist at Nordstrom Rack, focusing on improving search relevance and personalization through machine learning and customer behavior analysis. It discusses challenges such as poor customer experience, lack of a data-driven culture, and management issues, as well as the strategies implemented to overcome these hurdles. Noteworthy results include a significant increase in search conversion rates and improved business agility through collaborative efforts and enhanced data management.
Building Search and Personalization at Nordstrom Rack | Hautelook
- 2. My Journey PA N K A J A N D H A L E Senior Data Scientist Nordstrom Rack | Hautelook B U I L D I N G S E A R C H & P E R S O N A L I Z AT I O N J O U R N E Y • Background : Computer Science Engineering + Machine Learning • Experienced working on textual data enrichment leveraging Crowdsourcing approach and external data sources to create quality datasets to enable Machine learning @Intuit • Using scientific techniques to measure and tune the relevance of search algorithm @ Intuit • Currently working on improving relevance by leveraging customer behavior & building recommender systems
- 3. Outline • Challenges • Road to Relevance • Fusion • Results • Q & A
- 6. Challenge #1 Bad Customer Experience • Untuned Solr 6 • Customer complaints • No dedicated search team
- 7. Challenge #2 Data Driven Culture • KPIs not defined • No way to AB test performance of various algorithms • No dashboards to track quality, performance, or surface opportunities • How do we measure success? Search Visitors Non Search Visitors 1x 2x to 3x Conversion Rates (Industry) 1.67x
- 8. Challenge #3 Speed & Flexibility • No synergy between Search and Merchandising • Simple tweaks required code changes and deployments • Slow to fix simple issues
- 9. Challenge #4 Data Management & Processing at Scale • Processing customer behavior & leveraging in near real time • Infrastructure did not support
- 11. • 1 Product Manager • 1 Engineering Manager • 1 Architect / Principle Engineer • 1 Solr Engineer • 1 QA • 1 Data Scientist Forming a Team Road to Relevance
- 12. Fusion Platform Road to Relevance • Rapidprototyping • Testing • Deployment • EasyAB testing + • Signalaggregation • Versioning • Managedrules • Machine learning
- 13. Measuring Quality Road to Relevance • 42 KPIs and supporting metrics defined, audited, and tracked • Unlocked power of Google Analytics • NewSearchDashboardautomatedtime-consuminganalysis consuming analysis • Democratized search domain analysis • AB testing on iOS
- 14. Cross-Functional Feedback • 30+ Merch Team issues fixed in two weeks • Reciprocal sharing of knowledge led to quick understanding • Most new issues can now be resolved in the same day • Giant leap in business agility MerchandisersRoad to Relevance
- 15. Signal Aggregation Road to Relevance • Query:red dress
- 16. Noteworthy Example of Relevance Tuning Road To Relevance • Deep dive into the query logs • Understanding the customer intent • Leveraging the query pipeline stage • Significant increase in the search count
- 20. THANK YOU
- 21. Questions