With the rapid evolution of large language models into sophisticated shopping assistants, the quality of product recommendations has become the new battleground for e-commerce. To navigate this shift, we’re joined by Zainab Hussain, an e-commerce strategist specializing in customer engagement and operations. Today, we’ll explore how tapping into real-world shopping behavior, rather than just static product data, is setting a new standard for AI-driven commerce. We’ll discuss the intricate process of refining vast product catalogs into personalized, transaction-ready suggestions, the crucial balance between leveraging massive datasets and upholding brand integrity, and how this new layer of intelligence serves both retailers and the next generation of AI platforms.
Your Agentic Commerce Recommendation Service reportedly improves relevancy by up to 60%. What specific “real-world shopping signals” drive this significant lift, and how do they outperform models that only use product descriptions? Please provide a detailed example.
That 60% improvement is really the heart of what we do, and it comes from moving beyond the static, one-dimensional world of product descriptions. A model that only reads descriptions knows a jacket is “waterproof” or “insulated.” Our model, powered by the commerce intelligence from 720 million daily shoppers, knows so much more. It sees which “waterproof jackets” are actually being purchased in a specific region right now, which ones have a low return rate, and what other items, like hiking boots or thermal layers, are frequently bought alongside them. For example, if you ask an AI assistant for a “good camera for travel,” a basic model will just search for those keywords. Our service analyzes real-world signals and sees that while one camera has great specs on paper, a different, slightly cheaper model is being purchased five times more often by people who also bought plane tickets. That’s the difference—it’s not just about what a product is, but how people actually interact with it in the real world.
The service delivers a “curated shortlist” to AI assistants rather than raw catalog data. Could you walk me through the filtering and ranking steps that transform a broad product catalog into these transaction-ready recommendations for an individual consumer?
Certainly. Think of it as an intelligent, multi-stage filtration process. When a consumer’s request comes in, we don’t just dump a list of keyword-matched products on the AI assistant. We start with our vast inventory of 4.5 billion product SKUs and immediately begin applying layers of commerce intelligence. First, we filter for basic relevance and availability for that consumer. Then, the crucial ranking begins. We look at popularity signals—is this product trending up or down? We analyze user intent—has this person shown interest in sustainable brands or premium materials in the past? We even consider complementary signals—do people who buy this also buy that? The result is that instead of the AI assistant receiving a raw data feed of a thousand possible options, it gets a refined, curated shortlist of maybe five to ten products that are truly transaction-ready and hyper-relevant to that specific individual at that specific moment.
Given the scale of 720 million daily shoppers, how do you leverage this massive dataset for personalization while simultaneously respecting retailer data and maintaining brand integrity within an AI assistant’s environment? What specific safeguards are in place?
This is a critical point, and it’s foundational to our approach. The trust of our partners is paramount. While we operate at an immense scale, the intelligence we gather is used to understand patterns and intent, not to expose any single retailer’s proprietary data to another. The system is designed to be a trusted intermediary. When our service provides a recommendation to an AI assistant, it honors the retailer’s brand integrity; the product is presented as the retailer intends. We aren’t creating a new marketplace; we are building an intelligent bridge that connects a consumer’s query within an AI environment directly to a retailer’s inventory in a secure way. The key safeguard is that our model learns from the ecosystem’s collective behavior to improve relevancy for everyone, without ever compromising the data sovereignty of the individual retailers who contribute to it.
As retailers develop their own AI chatbots and large language models evolve into shopping agents, what is the core value proposition of your service for each? How does it specifically benefit a retailer’s proprietary AI versus a third-party shopping assistant?
The core value is the same for both, but the application is slightly different: it’s about providing commerce-grade intelligence that they otherwise wouldn’t have. For a retailer’s own AI chatbot, our service supercharges its capabilities. That chatbot may know its own inventory perfectly, but it doesn’t have the context of the wider market—it doesn’t know what’s trending across the entire internet or what a specific shopper was browsing on other sites just five minutes ago. We provide that crucial, real-time market context to make their on-site assistant dramatically more effective. For a third-party shopping assistant, our service is the engine that makes it truly useful for commerce. Without us, it’s just a conversational tool that can scrape public data; with our service, it becomes a genuine shopping expert that can provide trusted, relevant, and personalized results that lead to actual purchases.
Your service handles both exploratory and product-specific queries. Can you illustrate how it would process a vague request like “I need a durable suitcase for travel” versus a specific one, and how it intelligently suggests complementary items in each scenario?
That’s a great way to highlight the system’s flexibility. For a vague, exploratory query like “I need a durable suitcase for travel,” the service taps into broad intent signals. It will rank suitcases not just by the keyword “durable,” but by analyzing real-world data like low return rates, positive reviews mentioning longevity, and popularity among frequent flyers. It might then suggest complementary items that other travelers often buy, like packing cubes or a portable charger. For a specific query, say for a particular brand and model of suitcase, the service confirms availability and best options, but its real intelligence shines in the next step. It might say, “Great choice. Shoppers who bought that suitcase also frequently purchased this specific brand of travel pillow and this TSA-approved lock,” turning a simple product search into a complete, context-aware shopping experience.
What is your forecast for agentic commerce?
My forecast is that within the next few years, agentic commerce will become the default starting point for most online shopping journeys. We’ll see a major shift away from typing keywords into a search bar and scrolling through endless pages. Instead, consumers will simply have a conversation with a trusted AI assistant, describing their needs in natural language. The real competitive advantage in this new landscape won’t be the AI model itself, but the quality and depth of the commerce data powering its recommendations. The platforms that can access high-fidelity, real-world shopping signals to deliver truly relevant, personalized, and trustworthy results will be the ones that win the consumer’s loyalty. It’s an ecosystem-wide transformation, and it’s happening right now.
