An A/B test of AI-personalized banners on the o2 online store achieved a 17% click-through rate, a 275% conversion-rate uplift in the hero use case and a return on investment of 939%. The earlier MVP phase had already raised online-store conversions by 66% with an ROI above 550%. The page also reports 9 million calculated recommendations in 30 days.
"With Vertex AI Recommendations, we deliver personalized product recommendations in real time that instantly adapt to user behavior—and reliably drive conversions," said Steven Burkhardt, Head of Digital Analytics & Performance Management at Telefónica Deutschland.
The main challenge was personalizing the experience for visitors who are not logged in, who account for the majority of commercially valuable traffic. Recommendations rested on standard quotes and manually maintained, rule-based logic, which could not react to intent shifts within a session, such as a user moving from Apple to Samsung devices. The solution had to work without login, interpret browsing behavior with zero latency, automatically identify the highest purchase probability and raise revenue without manual effort.
e-dialog and Telefónica Deutschland built an event-driven, automated architecture on the Google Cloud Platform, called OREO (On-site Recommendation Experience Optimization), using Vertex AI Search for Commerce. In Phase 1, a lean MVP served "Frequently bought together" accessory recommendations on product detail pages. Specialists had previously entered rankings in Excel, converted them to JSON and uploaded them by hand. e-dialog combined Google Analytics behavioral data with product-feed data into continuous streams; the Vertex AI Retail API ranked products automatically and returned dynamic JSON to the product pages. The model draws on the product feed and real-time behavior delivered through Google Tag Manager.
In Phase 2, the hero use case personalized the homepage banner spaces with the highest reach using real-time session data. The infrastructure is defined as Infrastructure as Code and integrated into an automated CI/CD pipeline, making it auditable, reproducible and transferable to other brands. e-dialog also paired the onsite AI with Google Ads campaigns such as Performance Max and Demand Gen, so that campaign signals feed directly into onsite recommendations.
The next step is to make real-time browsing preferences available in the Mein O2 app, in-store retail and the service hotline, with the long-term goal of page-level optimization in which AI models arrange the entire page layout.