Vendors submitted images instead of structured data, starving the recommender of signal. A metadata mining engine lifted CTR by up to 73%.
Products on the e-commerce platform lacked sufficient text metadata because vendors submitted images rather than structured product data. Product information was embedded within those images as text, invisible to AI recommender systems and search algorithms. This created a cold-start problem: without structured metadata, the platform's discovery and personalization systems had insufficient signal to operate effectively, limiting their value precisely for non-frequent visitors , who represented 97% of the user base.
Profitmind's engagement delivered an AI transformation across four interconnected systems: a personalized product recommender engine, a Knowledge Graph for entity relationships, entity resolution combining deterministic and probabilistic matching, and a metadata mining engine that extracted and structured text data embedded within product images. The solution was integrated via mobile app APIs and validated through rolling four-week CTR measurement across multiple user segments. Deterministic identity unification was achieved at 100% accuracy.
CTR improved between 28% and 73% depending on user segment and time window, simprovements that were sustained across the measurement period rather than representing a one-time spike. The recommender consistently drove performance gains for non-frequent visitors, confirming that the system improves discovery for the long-tail audience most other personalization approaches underserve. Overall click-through rate exceeded 10% across the platform.