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.

The primary obstacle for OpenAI's shopping features isn't the transaction layer, but the complex task of standardizing inconsistent product data (sizing, pricing, inventory) across millions of merchants. This foundational data problem requires deep collaboration with partners and explains the slow, deliberate rollout.

Generative AI models like ChatGPT predict the next logical word based on vast, generic datasets. A more advanced approach uses predictive models trained on a brand's specific performance data—opens, clicks, conversions—to forecast which content variants will actually drive business outcomes, not just sound plausible.

Participating in AI commerce isn't just about capturing inbound data. Brands must structure and provide outbound data feeds of their inventory and product details in a format that LLMs can readily access for recommendations and transactions. This represents a significant new technical requirement for marketing teams.

OpenAI plans to personalize ads not just on immediate queries but by analyzing a user's entire chat history. This creates a powerful hybrid of Google's intent-based advertising and Meta's interest-based profiling, going beyond simple sponsored links to offer deeply contextual promotions.

Building reliable AI agents for finance, where accuracy is critical, requires moving beyond pure LLMs. Xero uses a hybrid system combining LLM-driven workflows with programmatic code and deep domain knowledge to ensure control and reliability that LLMs inherently lack.

Pega's CTO advises using the powerful reasoning of LLMs to design processes and marketing offers. However, at runtime, switch to faster, cheaper, and more consistent predictive models. This avoids the unpredictability, cost, and risk of calling expensive LLMs for every live customer interaction.

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.

AI models are becoming commodities; the real, defensible value lies in proprietary data and user context. The correct strategy is for companies to use LLMs to enhance their existing business and data, rather than selling their valuable context to model providers for pennies on the dollar.