Title

AWS re:Invent 2022 - Poshmark accelerates growth via real-time analytics & personalization (ANT342)

Summary

Poshmark, a leading social marketplace for fashion, has over 80 million users and uses data-driven decision-making.
Initially, Poshmark faced challenges with growth, compliance, fraud, spam, and user experience, which they addressed through batch computing and analytics.
They transitioned to real-time analytics to improve latency and reusability, enabling faster customer response and fraud detection.
AWS architects collaborated with Poshmark through the Data Lab program to design a real-time analytics solution.
The solution involved Kafka for sub-second latency, Flink for event-based processing, and microservices for enrichment.
Poshmark implemented a feature store engine using Flink and ElastiCache to store attributes for entities like users and listings.
They developed use cases for compliance (CCPA, de-identification), fraud mitigation (ATO attacks), and personalization (LTR).
The real-time analytics system improved fraud detection from 45% to 80% and increased top-of-funnel click-through by 8%.
Amazon Managed Streaming for Kafka (MSK) was crucial for setting up, maintaining, and scaling Kafka clusters.

Real-time analytics and personalization are critical for improving customer experience and accelerating business growth.
The transition from batch processing to real-time event-based processing can significantly reduce latency and improve the ability to respond to customer actions and detect fraud.
The use of microservices and a smart client approach for load balancing can enhance the performance of real-time systems.
A feature store engine is a powerful tool for aggregating and storing entity attributes, which can be leveraged for various applications, including machine learning models for fraud detection and personalization.
The collaboration between AWS architects and Poshmark through the Data Lab program exemplifies the benefits of close partnership in designing and implementing complex technical solutions.
The use of managed services like Amazon MSK can simplify the management of complex streaming data infrastructure, allowing companies to focus on business logic and customer experience rather than infrastructure maintenance.