01 · Disconnected customer signals
Recommendations were generic or arrived after the shopper had left. Customer behavior, purchase history, and profile data could not be used together quickly enough.
Customer work · Retail and consumer
CaratLane doubled add-to-cart conversion with in-session personalization
BluePi unified CaratLane’s customer signals in an AWS data platform and delivered personalized recommendations while shoppers were still active. The add-to-cart conversion rate on personalized interventions doubled.
Personalization moved into the active shopping session
CaratLane’s customer signals sat across separate systems, and recommendations lacked the context of the current shopping session. BluePi connected those signals to a low-latency delivery path so CaratLane could respond before purchase intent disappeared.
02 · In-session recommendation path
A unified customer profile, AWS data lake, processing layer, and low-latency delivery path turned current behavior into an in-session recommendation.
03 · Add-to-cart conversion
The add-to-cart conversion rate on personalized interventions doubled.
From customer signal to recommendation in the same session
BluePi unified customer signals, generated recommendations from current shopping intent, and delivered them during the active session so the result could be measured before the shopper left.
Build a usable customer profile
Website behavior, purchase history, identity, and preferences contributed to one profile. The recommendation path used current activity alongside previous purchases.
Respond while intent is active
The system delivered the recommendation during the active session, when the shopper could still act on it.
Measure the action that matters
CaratLane measured whether a personalized intervention led to an add-to-cart action. The conversion rate for that action doubled.
Where this pattern fits
This pattern fits retailers whose recommendation systems are limited by fragmented customer identity or slow data movement. A useful first step is to trace one intervention from event capture to an add-to-cart action and measure how response time affects conversion.
Case details
Open a section to review the customer problem, implementation, business change, and architecture.
01The starting point
CaratLane had behavioral and purchase data across separate systems. Existing personalization lacked current session context, and recommendations arrived too late to influence the shopper.
- Incomplete customer context: Website behavior, purchase history, identity, and preferences were stored in separate systems.
- Generic recommendations: Personalization could not use the shopper’s current intent alongside previous behavior.
- Delayed interventions: The existing data path could not process and deliver recommendations quickly enough during the shopping session.
02What BluePi built
BluePi connected customer data, analysis, intervention logic, and storefront delivery in one low-latency personalization path.
- Unified customer profile: Website interactions, purchase history, identity, and preferences contributed to one customer view.
- Real-time event path: An Application Load Balancer, Amazon EC2 Spring Cloud API Gateway, Amazon Kinesis, Amazon EMR, AWS Lambda, and Amazon DynamoDB connected live behavior to an in-session intervention.
- Transactional data integration: AWS Database Migration Service moved MySQL and MongoDB data into Amazon S3 for processing alongside clickstream events.
- Analysis and reporting: Amazon EMR, Amazon Redshift, and Amazon Athena supported processing, customer segmentation, analysis, and stakeholder reporting.
- Outcome feedback: Clicks, add-to-cart actions, orders, and margin measures returned to the customer profile for evaluation.
03Results
The add-to-cart conversion rate on personalized interventions doubled.
- 2× add-to-cart conversion rate: The conversion rate on personalized add-to-cart interventions doubled.
- Higher engagement: Click-through rates improved.
- More cross-sell and upsell orders: Personalized recommendations created additional purchase opportunities.
- Improved margins: More precise targeting made each intervention commercially more effective.