Mehmet Furkan
Akpınar

An e-commerce retailer had 4M+ behavioural events but no structured way to differentiate customers. Marketing spend was applied uniformly regardless of customer value or intent.

Built a hybrid segmentation pipeline combining K-Means clustering (silhouette score: 0.57) with rule-based segmentation. Engineered features including session duration, cart count, conversion rate, and purchase hour distribution per user.

41.2% cart abandonment rate — the largest re-engagement opportunity. Only 24.9% of users made more than one purchase, but repeat buyers spent 5× more on average. Purchase activity peaked 8–11 AM, directly actionable for campaign timing. Repeat buyers averaged a repurchase cycle of 98 hours.

What are customers actually complaining about vs. praising — and which themes drive low ratings? Aggregate star ratings hide the signal; the text is where the insight lives.

Sentiment analysis and topic modelling of Google Reviews using BERTopic to surface recurring pain points and positive drivers invisible in aggregate star ratings. Results visualised interactively with Plotly.

In progress — scraping and preprocessing complete, topic modelling underway.