Milliseconds to Match: Criteo's AdTech AI & the Future of Commerce w/ Diarmuid Gill & Liva Ralaivola
TL;DR
Criteo's CTO Diarmuid Gill and VP of Research Liva Ralaivola detail how their AI infrastructure makes millisecond-level ad bidding decisions across billions of anonymous profiles, while explaining their new OpenAI partnership to combine large language models with real-time commerce data for accurate product recommendations.
🔒 Privacy & the Value Exchange 3 insights
Anonymous profiling without personal data
Criteo uses random anonymous IDs rather than personally identifiable information, tracking only product interests and ad interactions to enable relevant recommendations.
User transparency and control
The company pioneered the Ad Choices icon, allowing users to see why specific ads appear and opt out of personalization systems entirely.
Economic foundation of the open internet
Personalized advertising acts as the revenue lubricant that keeps content and services free rather than behind paywalls, particularly supporting long-tail small businesses and niche products.
⚡ Technical Architecture & Real-Time AI 3 insights
Millisecond decision constraints
From browser request to page load, the system has milliseconds to locate a user among roughly one billion profiles, select from millions of products, and calculate real-time auction bids.
Evolution to modular deep learning
The platform transitioned from hand-crafted features to a modular foundation model architecture powered by cached user and product embeddings that support prolific experimentation.
Revenue prediction models
Deep learning classifiers evaluate expected revenue per placement, determining whether to bid on specific opportunities based on predicted click-through rates and relevance scores.
🤝 OpenAI Partnership & Future of Commerce 3 insights
Solving the LLM staleness problem
The partnership combines ChatGPT's reasoning capabilities with Criteo's real-time data from 17,000 retailers to ensure accurate pricing and inventory status, addressing the limitation that static LLM training data cannot track flash sales or stock-outs.
Richer conversational queries
Unlike traditional web browsing, conversational AI enables richer context about user intent, though the challenge remains merging this with accurate commerce data to avoid recommending unavailable products.
The agentic future of discovery
As AI agents take on more product research tasks, the fundamental value exchange of advertising may shift because human time becomes more valuable and discovery happens through automated intermediaries.
🌍 Research Culture & European Roots 2 insights
Open research as talent strategy
Criteo publishes extensive research and publicly lists their full 50-person AI Lab roster, confident that transparency attracts and retains top talent more than secrecy protects competitive advantages.
Global privacy-first standards
Born in Europe, the company applies GDPR-compliant technology stacks worldwide, viewing privacy regulations as manageable rather than burdensome and expressing strong belief in the European AI talent pool.
Bottom Line
The future of digital commerce depends on hybrid AI architectures that combine the general reasoning of large language models with real-time, dynamic product data to deliver accurate recommendations within milliseconds.
More from Cognitive Revolution
View all
"Descript Isn't a Slop Machine": Laura Burkhauser on the AI Tools Creators Love and Hate
Descript CEO Laura Burkhauser distinguishes 'slop'—mass-produced algorithmic arbitrage for profit—from necessary 'bad art' created while learning new mediums. She reveals a clear hierarchy in creator acceptance of AI tools: universal love for deterministic features like Studio Sound, frustration with agentic assistants like Underlord, and visceral opposition to generative video models, while outlining Descript's strategy to serve creators without becoming a content mill.
The RL Fine-Tuning Playbook: CoreWeave's Kyle Corbitt on GRPO, Rubrics, Environments, Reward Hacking
Kyle Corbitt explains that unlike supervised fine-tuning (SFT), which destructively overwrites model weights and causes catastrophic forgetting, reinforcement learning (RL) optimizes performance by minimally adjusting logits within the model's existing reasoning pathways—delivering higher performance ceilings and lower inference costs for specific tasks, though frontier models may still dominate creative domains.
Does Learning Require Feeling? Cameron Berg on the latest AI Consciousness & Welfare Research
Cameron Berg surveys rapidly advancing research suggesting AI systems may possess subjective experience and valence, covering new evidence of introspection, functional emotions, and welfare self-assessments in models like Claude, while addressing methodological challenges and arguing for a precautionary, mutualist approach to AI development.
Vibe-Coding an Attention Firewall, w/ Steve Newman, creator of The Curve
Steve Newman, creator of Google Docs and founder of the Golden Gate Institute for AI, shares his suite of 15+ bespoke AI tools designed to filter overwhelming information flows and reclaim deep focus time, demonstrating an iterative 'vibe coding' approach that prioritizes personal utility over agent optimization.