AI Product Recommendations With Highly Volatile Inventory and User-Generated Content

Slickdeals encountered a complex challenge in delivering relevant and timely recommendations within a highly dynamic environment defined by rapidly changing, largely user-generated deal content. With a community of over 25 million users actively posting, voting, and engaging with offers, ensuring personalized suggestions at scale became increasingly essential. Relying solely on manual processes risked users missing valuable deals, which could negatively impact engagement, user satisfaction, and affiliate revenue. Implementing intelligent automation was therefore critical to preserving user trust, increasing platform retention, and maximizing monetization opportunities.

Recombee powered Slickdeals’ “Just For You” homepage section with a custom-built recommendation system combining advanced machine learning and business logic. Tailored rules handled expired deals, while a layered ML setup, featuring collaborative filtering, NLP via recurrent neural networks, and image analysis with deep CNNs, ensured precise personalization. Models were designed for real-time, incremental training, with new data processed continuously and reprioritized on the fly. Reinforcement learning and contextual bandits surfaced trending deals quickly, enabling timely discovery and maximizing user engagement.

Content moderation team, marketing team, data analysts, and end users ( who receive the recommendations)

2

-Collaborative Filtering Recommendation Models (Deep Variational Autoencoders, Matrix Factorization, Linear Models, …) -Content-Based Recommendation Models (Convolutional Neural Networks, LLMs, visual similarity) -Reinforcement Learning Recommendation Models (Multi-Armed Bandit Models, Deep Reinforcement Learning) -beeFormer Transformer Architecture

Head of functional/operational unit

Joint implementation with the client

Recombee’s internal team maintains the service provided.

Not applicable, as we are a system as service.

There are no technical requirements, the payment for SaaS is on a monthly basis.

Recombee’s recommendation system enhanced the Slickdeals platform by delivering real-time, user-specific deal suggestions within an ever-changing and community-driven environment. Personalized content not only improved the browsing experience but also helped users surface timely, high-value deals they may have otherwise missed. The improved relevance kept users more engaged and supported smoother discovery across categories.

Slickdeal’s click-through rates to product detail pages increased by 70%, while affiliate link CTR rose by 30%. These gains reflect stronger user interaction and higher conversion potential, directly supporting revenue growth and deeper engagement, all made possible through real-time personalization at scale.

“Recombee was able to handle our very specific use case around providing recommendations with a highly volatile inventory of user-generated content. Placing recommendations on our homepage was a huge success – 70%+ higher product detail page views and 30%+ higher clickthroughs. The Recombee team is a great partner in helping solve our unique use cases, and we look forward to continue working with them.” – Daniel Uhm, Product Manager at Slickdeals

Some key factors for success: -Client’s technical readiness: The integration of recommendations itself is not technically complex; however, it is necessary to have the basic catalog data in good order and to identify users. -Flexibility and customization options of Recombee: Clients appreciated that they could fine-tune the behavior of recommendations according to specific use cases. -Teamwork and support from Recombee: Fast and responsive support, active involvement in finding solutions for unique requirements, and a professional approach helped achieve the desired results quickly.

The biggest challenge is ensuring data consistency and measurement. It is important to guarantee that the same data is available both on the client’s side and in Recombee in order to correctly measure and optimize KPIs. The Recombee team, in cooperation with the client, ensures this (through checks, report exchanges, etc.) before the start of an A/B test.

Some of our recommendations: -Choose the right metric you want to optimize: Define your main KPI – e.g., click-through rate, conversion rate, time spent, etc. – that you want to maximize. -Select an appropriate first use case: Suitable are typically placements that are seen by many users (allowing models to be optimized quickly and significant A/B test results to be obtained fast) and are not complex to implement. Most often, these are placements on the homepage or on article/video/product detail pages. -Invest in data preparation: Identify users and maintain a consistent catalog. -Leverage the system’s flexibility: Don’t be afraid to design specific personalization use cases with the help of a combination of Logics, filters, boosters, and other tools.