How a market-leading e-commerce is using GenAI to improve the customer experience

An established company in the corporate-gifts market, running three online platforms, faced challenges in personalizing and scaling its customer support because of the sector’s complexity. To improve first contact and recommend products more accurately, BlueMetrics built a solution based on GenAI. The technology enriched the product-category data, created a structured knowledge base, and deployed a contextual virtual assistant.

Overview

Our client is an established company in the corporate-gifts market, running three online platforms that connect suppliers and buyers. This sector is highly competitive and marked by complex decision-making, driven by the huge variety of products and the specifics of each request.

To stand out in a market that keeps evolving, the client needed a solution that could combine personalization, speed, and accuracy in the customer experience while scaling operations without a proportional rise in costs. Against that backdrop, BlueMetrics was brought in to propose a solution that could deliver meaningful improvements to the customer journey.

“We specialize in building real solutions for real problems,” he notes. “So this kind of challenge fits both our way of working and our range of solutions.”

---

Problem: how do you improve first contact and product-category recommendations?

This client’s operation faced significant challenges that hurt both efficiency and the customer experience. Support was limited to business hours, which left customers without help outside that window. On top of that, the process leaned heavily on each agent’s individual knowledge, so requests were interpreted and routed manually, often causing delays and errors.

Among the technical limits, the category data had little semantic content and no systematic view of the right purposes and events for each category, which made an intelligent recommendation system hard to adopt.

According to Diórgenes Eugênio, Head of GenAI at BlueMetrics, “The real differentiator of the virtual assistant project for this e-commerce is exactly the way we built the knowledge base using LLMs. That approach let us give the assistant far more context. At the start we had very little semantic information about the categories, and we finished the project with an automated pipeline that processes all the content coming from the e-commerce and improves the data semantically and contextually so it can serve as the assistant’s reference source. This project was a first for BlueMetrics when it comes to using LLMs for data enrichment.”

On the commercial side, slow first contact hurt customer satisfaction. The team also got overloaded in seasonal peaks like Christmas and year-end, which made things worse and led to inaccurate routing, rework, and lost business. These bottlenecks created a series of direct impacts on the business:

• Customer frustration with long response times;
• Unintended favoring of certain suppliers;
• Limited business growth because of the manual support model.

Faced with these challenges, the client needed a solution that was scalable, impartial, and available 24/7, one that cut response time and gave everyone fair access to supplier options. So BlueMetrics built an intelligent solution to automate and improve the first-contact process.

---

The solution: GenAI for personalization and scale

To tackle the challenges identified, the client rolled out a robust solution based on AI, built on three main pillars:

Data enrichment

The process starts by handling XML data pulled from the client’s platforms, using Amazon Bedrock LLMs to enrich the product-category descriptions. It also adds relevant context about the right events and purposes for each category, resulting in a rich, highly structured knowledge base that serves as the foundation for the other features.

Intelligent knowledge base

The enriched information is converted into PDF files and stored in a vector database optimized for semantic search. This architecture ensures not only efficient search but also continuous data updates, keeping the information relevant and accurate over time.

Contextual virtual assistant

The assistant is designed to interact naturally with customers, understanding each one’s context and specific needs. Using Retrieval-Augmented Generation (RAG), it offers relevant, impartial recommendations, suggesting product categories in a precise way suited to each situation.

Together, these components produced an effective, innovative solution that let the client improve first contact, reduce operational bottlenecks, and give its customers a more personalized, satisfying buying experience.

---

Results:

Rolling out the AI-based virtual assistant delivered a range of benefits for the client, translating into concrete, immediate financial results.

Operational benefits

• 24/7 support, removing the dependence on business hours;
• Shorter initial wait time for support;
• A standardized recommendation process;
• Unlimited simultaneous support capacity;
• Less manual workload for the team.

Technical benefits

• A semantically enriched knowledge base;
• A scalable, flexible architecture;
• Easy adoption of new LLMs;
• Simpler maintenance of the knowledge base.

Benefits for the customer

• Instant responses to requests;
• More accurate, contextualized recommendations;
• Impartial category suggestions;
• A better buying journey;
• More confident product choices.

---

Technologies used

The solution built for this e-commerce client was designed on AWS technologies, including:

AWS services

OpenSearch

Bedrock

Lambda

CloudWatch

S3

Amplify

Cognito

StepFunction

Languages, libs, and frameworks

Python

Streamlit

Fast API

---

Conclusion

The GenAI-based solution let this e-commerce player scale its operations and significantly improve the customer experience, further cementing its position in the corporate-gifts market. With a robust, scalable, and highly personalized system, the company is now ready to meet growing demand while keeping quality and accuracy as its competitive edge.