Project: Stitch Fix
Client Engagement Acceleration System
Stitch Fix
A client engagement acceleration system that monitors customer engagement, identifies at-risk clients, and automatically generates personalized re-engagement emails using AI.
Project Demo
Project Metrics
Project Details
The Client Engagement Acceleration System reduces client churn and increases retention by proactively identifying disengagement risk and automating personalized re-engagement, directly addressing a key business risk identified in Stitch Fix's SEC report.
Business Value
- Customer Retention
- Engagement Analytics
Key Features
Engagement Monitoring System
Sophisticated scoring algorithm based on order recency, frequency, and value with real-time tracking of client engagement metrics.
Automated Re-engagement System
AI-powered generation of highly personalized emails with structured output in consistent JSON format and HTML-formatted email content with personalized recommendations.
Management Dashboard
Real-time visibility into client engagement metrics with user management interface for creating and updating clients.
Event-Driven Architecture
DynamoDB Streams for capturing data changes, SNS/SQS for reliable message delivery, and Lambda functions for event processing.
Technologies Used
Frontend
Backend
DevOps
Other
Challenges & Solutions
Email generation not working for users with low engagement scores
Modified the email processor Lambda to use the existing engagement score from the database instead of recalculating it, ensuring emails are generated for users with low engagement scores.
Email sending failing due to SES verification issues
Removed SES dependency and modified the Lambda to skip actual email sending and just mark emails as "SENT" in DynamoDB, eliminating the need for email verification in AWS SES.
Email content not properly formatted as HTML
Implemented structured output with JSON Schema validation, switched to openai/gpt-4o model for better structured output support, and enhanced the prompt to explicitly require JSON output.
Key Learnings
- Event-Driven Architecture: The event-driven architecture provided a clean separation of concerns, enabling loose coupling, scalability, and resilience.
- AI-Powered Content Generation: Structured output with JSON Schema validation proved essential for reliable AI-generated content.
- Multi-Language Implementation: Using Go for the email processor alongside TypeScript for other components demonstrated the value of selecting the right language for specific workloads.
- Executive Communication Style: The Stitch Fix project featured the strongest email communication so far, adopting a style inspired by Mark Zuckerberg.