Criteo views the "human in the loop" not as a fallback but as a fundamental design requirement for all AI systems. Their development process explicitly focuses on identifying the correct place for human intervention and decision-making, believing that full automation is both risky and less effective.
Criteo builds multiple, specialized foundation models (for products, user timelines, etc.) rather than a single monolithic one. The embeddings from these models are made available across the company, serving as a "warm start" to accelerate the development and improve the performance of new AI products.
Criteo successfully retains its 50-person AI lab team by fostering a culture similar to academia. Researchers are encouraged to publish their work, make it reproducible, and maintain a public presence. This commitment to open science and challenging problems is a key differentiator in attracting and keeping top talent.
The future of personalization may involve a two-step process. A centralized AI (like Criteo's) will provide strong recommendations. Then, a smaller, privacy-centric model running locally on the user's device (e.g., in their glasses) will perform the final, hyper-personalized adjustments, keeping the most sensitive data private.
Criteo has just milliseconds to respond to an ad request. This extreme speed requirement dictates their AI architecture, forcing them to pre-compute and cache user and product embeddings. Real-time inference is limited to fast operations with only marginal updates for the user's latest action.
Contrary to the view that European regulations stifle innovation, Criteo leverages its European roots. They built a single, global tech stack compliant with the highest privacy standards (like GDPR) from the start. This privacy-first approach is applied worldwide, simplifying operations and building user trust.
Criteo’s strategy with OpenAI is to create a hybrid system. LLMs provide general reasoning and conversational ability, but their knowledge quickly becomes stale for dynamic commerce data like pricing and stock. Criteo provides the real-time data layer to ensure accuracy and avoid bad user experiences.
Criteo's models moved from using manually crafted, extremely high-dimensional sparse vectors (e.g., 2^12 features) with linear models to dense vectors (a few hundred features) automatically computed by deep learning algorithms. This shift eliminated manual feature engineering and improved model adaptability.
As AI assistants become more capable, the fundamental advertising dynamic may invert. Instead of being passively shown ads, users might actively instruct their agents to "go find me five options for shoes," effectively requesting advertising. The value exchange changes to one where users want curated commercial options.