Behavioral Signals of Ecommerce Conversion
Conversion is not one magic funnel metric; it is a pattern of intent, comparison, friction, and trust.
Stack
Description
A COGS 108 project analyzing large ecommerce clickstream datasets to understand which pre-purchase behaviors best predict online purchase conversion.
Context
The research question focused on behavioral signals: views, cart actions, removals, session duration, product/category/brand comparison, returning-user signals where available, and viewed-product price context.
System
The analysis aggregated raw event logs into session-level datasets with leakage-aware pre-purchase features. The primary multi-category dataset covered 9,244,421 sessions, with additional cosmetics and jewelry datasets used for cross-category comparison and high-value order context.
Intelligence
Interpretable modeling and exploratory analysis pointed to cart activity, session duration, engagement intensity, product/category comparison, and price context as the clearest signals. Cart activity was especially strong, but cart drop-off also showed how much high-intent friction remains before checkout.
Iteration
The project kept the claims observational rather than causal, using the model as an interpretive check and framing the findings around better ecommerce design: reduce uncertainty, build trust, and make the next step easier for high-intent users.