This is an AWS & Constructor webinar recap post. Watch the full webinar here.
AI buzzwords are flying around in ecommerce, but they’re not always clear. Terms like “foundational models,” “GenAI,” and “agentic AI” can make your eyes glaze over.
In our recent webinar, David Dorf, Global Head of Retail Solutions at AWS, and our own Eli Finkelshteyn, Founder and CEO at Constructor, unpack the lingo and got down to what it all really means for retail teams:
You can catch the webinar replay here. In the meantime, here are some key takeaways from the discussion.
Foundational models are large AI systems trained on massive datasets — text, images, code, and more — that can be adapted for a variety of tasks. They’re the underlying intelligence behind tools like ChatGPT, advanced search engines, and recommendation platforms.
Why they matter in ecommerce
Unlike older keyword or basic vector search, foundational models understand context, not just similarity. A traditional search for “butter” might also return “margarine” because they’re semantically close. But if a user types “Find me a flight from Boston to London,” older systems often failed to understand the directionality, treating “Boston” and “London” interchangeably. Foundational models, built on transformer architectures, interpret the relationship between terms and return results that match intent.
This leap in contextual understanding improves:
The net effect: better alignment between what shoppers mean and what your platform delivers.
While general-purpose foundational models are versatile, they’re not optimized for ecommerce’s unique challenges. That’s where domain-specific models come in. They're trained on your vertical’s data, like product attributes, clickstream behavior, and purchase history.
Example:
A general model predicts the next best word in a sentence. An ecommerce-trained model predicts the next best action. If a shopper buys organic milk and bread, the model can suggest organic strawberries — understanding the purchase pattern rather than just keyword overlap.
Benefits for ecommerce teams:
Because they’re trained on proprietary behavioral data, these models often outperform general AI in driving conversion rates, increasing basket size, and lifting revenue.
Generative AI (GenAI) can create new text, images, video, and code, enabling faster content creation for ecommerce teams. In practice, this means:
ROI in action:
Key takeaway: GenAI doesn’t need to be perfect to be valuable. When targeted at use cases that either improve the customer experience or boost productivity, the payoff can be significant.
The next wave of foundational and generative models is multimodal, capable of handling text, images, voice, and video.
Examples from AWS:
For merchandising teams, multimodal AI enables:
As AI makes it easier to generate multiple variations of product copy, images, or videos, there’s a great opportunity to expand personalization from product recommendations to fully personalized product pages — content and all.
Imagine dynamically swapping product images to reflect the shopper’s demographic or browsing history, or adjusting the tone of descriptions to match their buying motivations. While not mainstream yet, this is where personalization is headed. And early experimentation now will pay dividends later.
If GenAI creates content, agentic AI acts. These systems can reason about a goal, break it into steps, and execute those steps (often with minimal human input).
Pricing: Gathering demand data, competitor pricing, and brand guidelines to recommend optimal prices
Search & Discovery: Interpreting open-ended queries (“I’m camping on Mount Whitney”) and factoring in weather, terrain, and inventory to recommend gear
Product Page Q&A: AI agents that auto-generate FAQs and answer customer-specific questions in real time
Inbound Agents: Bots visiting your site on behalf of shoppers, requiring a strategy for data accessibility and optimization
Outbound Agents: Personal shoppers that source products you don’t carry and fulfill under your brand
MCP connects AI agents to tools and data sources, telling them what they can use and how. For example, in the Mount Whitney scenario, an MCP-enabled agent could:
MCP also allows retailers to deliver real-time, personalized data to inbound agents, potentially tailoring results for each digital visitor, whether human or AI.
Phase 1 — Quick Wins
Phase 2 — Competitive Differentiation
Both experts agreed: Buy unless you have an extremely unique use case. AI innovation is moving too fast for most internal teams to keep up, and switching vendors is easier than unwinding a stalled internal build.
Foundational models have redefined what’s possible in search and product discovery. Domain-specific training pushes performance even further. Generative and multimodal AI expand creative capacity and personalization. And agentic AI promises to transform ecommerce from reactive transactions to proactive, goal-oriented experiences.
The real opportunity lies in moving past buzzwords and starting with the practical, high-ROI use cases that make a difference for your team and your customers today.
Intrigued by AI in ecommerce? Watch the full webinar replay here