
The Gumbel‑Max Trick Made Intuitive
When we work with machine learning models, we constantly turn probabilities into discrete choices. Which ad do we show? Which action do we sample in a reinforcement learning policy? Which word comes next in a language model? All these problems share the same core operation: We have a list of options, each with an associated…

Readable Unit Tests: make tests you actually want to write (and read)
You’ll read this test five times more than you’ll write it. Future-you will thank present-you for mocking it readable. This article distills and expands the talk I gave internally at Criteo: Readable Unit Tests. It’s aimed at developers who know unit testing but avoid it because tests feel hard to write, hard to maintain, and — let’s…

Agentic AI Symposium 2026
A science‑fueled, multidisciplinary approach to Agentic AI Authors: Jean-Yves Franceschi & Mariia Vladimirova At the Criteo AI Lab, we believe that making AI both trustworthy and useful requires a continuous and open dialogue between scientists, engineers, policymakers, regulators, and industry leaders. As AI systems evolve from passive predictors to agentic systems capable of planning, using tools,…

Our Experience at Snowcamp 2026
Ten years of the coolest dev conference in Grenoble 🏂 Authors: Nicolas Terreyre, Sylvain Dedieu, Arnaud Becquet, Nicolas Drufin, Nicolas Beaudouin, and Léo Gatellier For the fourth consecutive year, Criteo was proud to partner with the Snowcamp conference in Grenoble. Furthermore, we had the privilege of hosting the “Snowcamp Tremplin” at our Grenoble office last November,…
Nicolas Terreyre & 5 others
Leveraging Commerce Data for Outcome-Based Relevancy in Agentic Recommendation Systems
Learn how commerce data improves outcome-based relevance in agentic recommendation systems, delivering 37% better retrieval and 60% better SKU re-ranking.

🏆R&D Community Awards 2025
Our annual ceremony, where we celebrate achievements together. For the third consecutive year, we gathered to celebrate our achievements within the R&D community and recognize the contributions of all our R&D Ambassadors. At Criteo, expressing gratitude to our teammates is a consistent practice throughout the entire company. In our R&D department, our expert community embraces…

Video Ads: Introducing video interactivity in production
At the end of 2024, Criteo’s Video R&D team started working on a new goal: adding recommended products on top of videos, so that each display delivered to someone on the Open Web would become the most relevant possible. Over the last two decades, Criteo has continuously advanced its recommendation engine. Good news: From now…

Off-Policy Learning in Large Action Spaces
Optimization Matters More Than Estimation This Research Card shows that in large-scale recommendation systems, like those used at Criteo, focusing on how we optimize algorithms is more impactful than trying to perfectly estimate outcomes. By using simpler, more stable training objectives, we can achieve better click-through rates, faster experimentation, and more scalable performance across millions…

Criteo Boosts Ad Performance with Latest AI Advancements
Abstract. Everyone talks about AI today, and it seems AI has become an off-the-shelf tool providing plug-and-play capabilities. However, when addressing the hardest problems, such as those arising in the Commerce Media and AdTech industry, things are far more complex and require advanced AI (more specifically, machine learning) expertise. In this post, we share the…

AuthZilla, Rethinking Authorization For The Modern Era
Authorization is a cornerstone of the Identity and Access Management (IAM) stack. Yet, despite its critical role in securing digital systems, it suffers from a lack of standardized mechanisms, protocols, and formats to communicate authorization-related information across platforms. This gap has real consequences. According to the OWASP Top 10, Broken Access Control is the #1…










