This top Indonesian e-commerce platform served over 35 million monthly active users and listed more than 120 million products across categories ranging from electronics to fashion to groceries. Despite their massive user base, the platform's product recommendation engine was a basic collaborative filtering model that had not been significantly updated since 2020. It treated all users within broad demographic segments similarly and could not adapt to individual browsing behavior in real time.
Conversion rates on recommended products were declining — from 4.2% to 2.8% over 18 months — as competitors with more sophisticated personalization captured market share. The average order value had plateaued at IDR 185,000, and the platform's data science team estimated that poor recommendations were leaving approximately USD 12 million in annual revenue unrealized. Internal A/B tests showed that users who received irrelevant recommendations were 2.3 times more likely to abandon their session.
The platform's data science team of 12 engineers had attempted to build an improved recommendation system internally but struggled with the sheer scale of the product catalog, the diversity of user behavior patterns across Indonesia's varied regional markets, and the computational infrastructure required to serve personalized recommendations to millions of concurrent users with sub-200-millisecond latency.
Pertama Partners conducted an AI Readiness Audit that analyzed the platform's existing recommendation infrastructure, data pipeline architecture, and user interaction data. We discovered that the platform was sitting on extraordinarily rich behavioral data — clickstream data, purchase history, search queries, wishlist additions, cart abandonments, and review patterns — but only a fraction was being utilized by the existing model.
Our AI Pilot Program developed a multi-stage recommendation engine using a hybrid approach combining deep learning collaborative filtering, content-based similarity, and real-time contextual signals. The system incorporated user session behavior (what someone browsed in the last 10 minutes), temporal patterns (shopping behavior varies dramatically between Ramadan, payday periods, and regular days in Indonesia), geographic preferences (product popularity varies significantly across Java, Sumatra, and Kalimantan), and price sensitivity signals.
The AI Transformation Program scaled the recommendation engine across all product categories and touchpoints — homepage, product detail pages, cart page, and post-purchase emails. We worked with the platform's infrastructure team to deploy the model on an optimized serving architecture that maintained sub-150-millisecond response times at peak traffic. Executive Training sessions helped the leadership team understand how to measure recommendation quality beyond click-through rates, including downstream metrics like customer lifetime value impact.
"Our previous recommendation engine treated Indonesia as one market. Pertama Partners helped us understand that a mother in Surabaya and a student in Jakarta want entirely different things — and now our platform knows that too."— Budi Santoso, Chief Product Officer
Every transformation starts with a conversation. Let's discuss your challenges and opportunities.
Discuss Your Challenge