AI-Powered Product Discovery for a B2B E-commerce Catalog
Client
The client is a large UK-based B2B distributor of apparel and promotional clothing, serving businesses, resellers, and print and branding companies. The business operates a complex e-commerce platform with a large product catalog spanning multiple brands and product variations, primarily serving logged-in trade customers.
This project focused on the product discovery and customer interaction layer of their e-commerce platform — the point where customers search, compare, and make purchasing decisions across a highly varied product range.
Challenge
The client needed to improve how trade customers discover, understand, and select products within a large and complex catalog.
Product data was rich but fragmented across catalog, pricing, inventory, and supplier systems. Customers frequently needed contextual guidance — comparisons, recommendations, use-case advice — that traditional search and filtering tools could not provide. Ambiguous or conversational queries regularly fell outside what the existing platform could handle.
The result was high effort for users trying to find suitable products, a dependence on support teams for product guidance, and a volume of repetitive product-related inquiries that placed unnecessary load on customer service operations.
Goals / Success Criteria
Primary goals
- Deliver an AI agent capable of acting as an intelligent product expert within the e-commerce experience
- Streamline product discovery and selection for trade customers
- Reduce customer support workload by handling product queries conversationally and accurately
Success validation
- Accuracy and relevance of responses against real-world product queries
- Ability to handle multi-turn, ambiguous, and contextual customer questions
- Stakeholder validation of prototype usability
- Reduction in manual support interactions
Key constraints
- All responses strictly grounded in client product data — no generic LLM outputs
- Business-safe conversational guardrails
- Integration with existing systems, including Optimizely, PIM, and inventory
- PII protection throughout
- UK English only
Our Role
First Line Software acted as the AI development partner for this engagement, working alongside the client, who led product, UX, platform, and e-commerce integration.
First Line Software was responsible for technical scoping, AI architecture design, backend AI development, and system integrations. The team included an AI consultant alongside AI, frontend, and backend engineers. The client provided product, design, QA, and overall project coordination.
The engagement was structured as a fixed-scope discovery and prototype, delivered over two months, with a working AI agent prototype as the defined output.
Approach
The engagement followed a structured five-phase delivery: a discovery and definition sprint, requirements workshop and technical validation, solution design, prototype development, and stakeholder playback and validation.
The discovery sprint — run over the first two weeks — established the requirements, validated the technical approach, and aligned teams across product, AI, and engineering before any build work began.
Two decisions shaped the delivery. First, a prototype-first approach was adopted to validate the architecture and product fit before committing to a full MVP build. Second, the scope was deliberately kept focused on product use cases in this phase to reduce risk and maintain delivery focus. The RAG architecture, grounded entirely in client data, ensured that responses remained accurate and business-safe throughout.
Quality was maintained through structured workshops, acceptance scenarios defined against real user queries, and close cross-functional collaboration across the product, AI, and engineering teams.
Solution
First Line Software delivered an AI-powered conversational product expert embedded directly into the client’s e-commerce experience and designed to act as an intelligent guide for trade customers navigating a complex product catalog.
The solution moves beyond traditional search, replacing it with a context-aware agent that understands natural language, maintains conversational context across multiple turns, and grounds every response in live business data.
Core capabilities delivered include:
- Conversational AI agent with an embedded chat interface
- RAG-based knowledge system drawing on product catalog, pricing, and inventory data
- Context-aware, multi-turn query handling
- Product recommendation and comparison capability
- Follow-up clarification for ambiguous queries
- Live chat handoff when human support is needed
- Integration with product, pricing, and inventory systems
The solution was built on First Line Software’s internal agentic RAG framework, accelerated using reusable AI components, including a shared chat widget and evaluation tooling.
Technology
The technology stack was designed to support accurate, grounded AI responses while integrating cleanly into the client’s existing e-commerce infrastructure.
Core stack
- Agentic RAG framework (Jaime-based)
- API-driven integrations with Optimizely, PIM, and inventory systems
Accelerators and reusable components
- Jaime AI common framework
- Shared chat widget
- Internal evaluation tooling for response quality validation
Results
The prototype was validated by stakeholders and demonstrated the viability of conversational AI as a product discovery layer within a complex B2B ecommerce environment. The solution is not yet in production, with an LLM usage optimization phase underway ahead of a phased go-live for selected customers.
The engagement established a clear foundation for measurable business outcomes, including a reduction in repetitive support interactions, an improved customer experience through guided conversational discovery, and a reusable AI architecture that can scale across the broader e-commerce platform.
The next phase will focus on optimizing LLM usage and preparing the solution for production rollout to an initial segment of trade customers.