Working time:
2024 – ongoing
Industry:
Retail
The service:
Generative AI
Overview
The client is a leading multinational retail brand with a strong presence in both e-commerce and physical store formats across North America and Europe. Renowned for its broad product range—from consumer electronics to apparel and home goods—the company consistently seeks to redefine customer engagement through innovation.
With a digital-first vision, the client aimed to elevate their shopping experience by embedding intelligence directly into the product discovery journey. Their goal: to help customers effortlessly navigate vast product catalogs and receive highly relevant, conversational recommendations in real time.
To accomplish this, they partnered with our team to design and deploy a Generative AI-powered product recommendation assistant—similar to Amazon’s Rufus—built entirely on AWS infrastructure. The result is a solution that blends natural language understanding with contextual awareness, delivering intuitive and personalized customer experiences that drive both satisfaction and sales.
Challenge
With an expanding online catalog exceeding hundreds of thousands of SKUs and diverse customer profiles spanning multiple geographies, the client faced several strategic and technical challenges:
Friction in Product Discovery
Traditional search and filter systems were not meeting customer expectations. Many users found it difficult to articulate exactly what they were looking for, resulting in high bounce rates and abandoned sessions.
Siloed Customer Insights
The client had rich behavioral and transactional data, but it was fragmented across disparate systems. This disjointed view prevented the delivery of unified, personalized recommendations that could adapt to user context in real time.
Limited In-House ML Capabilities
While the client had experimented with traditional recommendation engines, their internal team lacked the capacity and GenAI-specific expertise to design, train, and deploy a scalable, conversational AI solution at production-grade performance.
Cost-Effective Scalability
Given the seasonality of the retail business, the client needed an architecture that could scale up for peak periods (e.g., Black Friday, holiday season) and scale down gracefully during off-peak times—all without compromising performance or blowing out operational budgets.
Case study
5 min to read
Automating remittance workflows and reducing costs with GenAI and AWS
Solution
We architected a robust, cloud-native Generative AI assistant tailored to product discovery and recommendation—leveraging state-of-the-art AWS services and retail-specific language modeling techniques.
Key components of the solution included:
Foundation Model Fine-Tuning with Bedrock
We utilized Amazon Bedrock to access and fine-tune foundation models optimized for retail language patterns and product taxonomy. These models were adapted using the client’s anonymized product metadata, customer inquiries, and feedback loops to provide domain-specific relevance and tone.
Real-Time Context Engine with Amazon Personalize
To enable dynamic personalization, we integrated Amazon Personalize as the backend engine for real-time user behavior analysis. The system dynamically adjusts recommendations based on session data, clickstream, and past purchases—feeding directly into the GenAI assistant for natural conversation flows.
Scalable Inference Layer with SageMaker and Lambda
Inference workloads were containerized and deployed using Amazon SageMaker endpoints for high-throughput, low-latency response. Serverless orchestration via AWS Lambda allowed for cost-efficient handling of variable traffic loads while maintaining sub-second response times.
Secure Data Integration via AWS Glue and DynamoDB
We built a secure, scalable data integration layer using AWS Glue to transform and prepare product and behavioral datasets for model training. DynamoDB was used as the low-latency data store for rapid retrieval of user profiles, product embeddings, and metadata at scale.
Conversational Interface via Amazon Lex and API Gateway
A user-facing chat interface was implemented using Amazon Lex, integrated with the GenAI backend through API Gateway and Lambda. This allowed customers to ask questions like “What’s a good gift for a 10-year-old who loves science?” or “Show me stylish yet affordable office chairs,” and receive curated product selections instantly.
If you find this case interesting, we recommend taking a closer look at
Results
The GenAI product recommendation solution delivered measurable improvements in both customer engagement and operational efficiency, transforming the client’s digital commerce experience.
+37% Conversion Rate for Engaged Sessions
Users who interacted with the GenAI assistant converted at significantly higher rates compared to traditional search users. The assistant’s conversational and adaptive nature helped users find what they needed faster, reducing friction and increasing cart completions.
4x Faster Product Discovery
Customers reported faster and more satisfying experiences when exploring the catalog. Average time-to-find dropped from 2.5 minutes to under 40 seconds for most product types, especially in broad categories like home goods and personal care.
60% Reduction in Product Returns
Contextual recommendations led to better purchase decisions. Customers received suggestions better aligned with their needs, resulting in significantly fewer post-purchase returns—directly impacting fulfillment and logistics costs.
~50% Lower Inference Costs
By deploying inference endpoints on spot instances via SageMaker and using token-level optimization techniques, we achieved substantial cost savings per API call, allowing the solution to scale efficiently across multiple storefronts and geographies.
Increased Customer Loyalty & Retention
Repeat visits and loyalty program sign-ups saw noticeable uplift after the assistant’s launch. Users appreciated the brand’s investment in innovative, helpful technology—driving long-term retention and net promoter score (NPS) improvements.
