We helped a leading B2C retailer modernize their e-commerce personalization. By replacing manual curation and black box solutions with an LLM-infused engine, the customer can now generate high-quality, context-aware recommendations for thousands of products daily — improving User Experience (UX) at a fraction of the cost of human labor.

Balancing quality with massive scale

As a leading Finnish B2C retailer, our customer faced a difficult trade-off between scale and quality. With a massive and constantly changing product catalog, maintaining relevant recommendations was a significant hurdle. Manual management was impossible due to the sheer volume of products, and existing third-party tools were “black boxes” that offered no visibility into why specific products were recommended.

The goal was not just to increase the Average Order Value (AOV) by building larger baskets, but also to improve the overall User Experience (UX). The customer wanted to ensure that every visitor received logical, helpful suggestions that would build trust and drive long-term retention. They needed a partner who could deliver a transparent, high-volume solution that understood product context without requiring a massive manual workforce.

A semantic engine built for retail complexity

We designed a tailored, LLM-infused architecture to tackle the volume challenge. Taking a phased approach—starting with a Proof of Concept and moving to a managed rollout—we deployed a solution capable of deep semantic analysis, instantly connecting seasonal context with product intent. Unlike standard keyword matching, the engine “understands” that winter boots relate to wool socks, not beach towels, even if they share overlapping tags.

The technical solution prioritized precision and cost-efficiency. By using a multi-stage AI workflow, the system anchors recommendations in the retailer’s specific catalog structure while applying advanced logic to filter for business priorities. This ensures the creativity of an LLM is balanced with strict inventory controls. The output integrates seamlessly into the customer’s existing personalization engine, minimizing overhead while processing tens of thousands of products at a fraction of typical cloud costs.

Unmatched efficiency and immediate expansion

The project has delivered distinct operational efficiency and validated quality. The system currently generates recommendations for over 5,000 products per 24 hours—a volume unreachable for human teams. The cost savings are substantial: the cloud cost to process this volume is equivalent to just a few hours of a product marketing specialist’s salary.

The impact on the customer’s business was immediate. Within days of the latest recommendations going live, the quality was deemed high enough to fast-track the expansion of the solution. The retailer has now expanded to generating recommendations for additional product categories, confident in the system’s ability to improve customer experience and increase AOV — both of which contribute to higher revenue. Additionally, the “reasoning data” generated by the AI (explaining why items fit together) continues to provide strategic value, enabling automated buying guides and smarter customer service tools.