Major Global Retailer
Generative AI Delivers Speed and Scalability for Online Retailer
Project Summary
A Global Retailer's Challenge
The good news? Our client was growing at a phenomenal rate.
The bad news? This high growth put tremendous pressure on the marketing team to keep up with multi-language product descriptions for its extensive new online product launches.
The EDJE Solution
We built a fine-tuned, multi-modal, scalable Generative Artificial Intelligence (AI) Retrieval-Augmented Generation (RAG) solution with the ability to quickly and accurately generate multi-language product descriptions that are true to the company's iconic brand voice.
For more detail, please read on.
Introduction
The client is a leading, global, omnichannel specialty retailer of apparel and accessories for men, women and kids through five renowned brands.
Founded in 1892, the company operates online as well as in 780+ stores across North America, Europe, Asia and the Middle East.
Challenge
The company's online product offerings were expanding rapidly and the volume of product descriptions — all in multiple languages — became overwhelming. In spite of the copywriters' efforts to keep pace with the increasing workload and shorter timelines they couldn't keep up. This bottleneck was characterized by 3 key issues:
- Inefficiency—The traditional copywriting process was slow, making it virtually impossible to keep pace with the flood of new product rollouts.
- Inconsistency—With multiple copywriters crafting product descriptions, consistent and cohesive content and brand tone were difficult to maintain.
- Volume—The existing workflow was bursting at the seams — it simply couldn't be expanded to accommodate the growing number of products being introduced.
The company needed a way to generate product descriptions that would be faster, scalable, and true to their brand — a challenge tailor-made for EDJE's expertise
The Project Objectives
- Immediate and significant decrease in the hours needed for product description development.
- Accurate product descriptions that require minimal edits by human copywriters.
- Immediate and considerable cost savings on translation services.
Solution
The client’s Data Science Team initially experimented with AI, fine-tuning language models based solely on metadata. The results were not impressive — they needed something better. Leading EDJE agreed with them that AI was the right way to go, but proposed a different solution — a multi-modal Generative AI approach that would blend metadata with product imagery. In other words, EDJE envisioned a system that would mimic the way a skilled copywriter interprets visual and contextual cues to draft product descriptions.
The Technology Behind the Magic
In the end, we didn’t just throw AI at the problem — we engineered a holistic solution tailored to meet the company’s unique needs using:
- Multi-modal AI—We fine-tuned a text/vision model (PaliGemma) using the company’s extensive product catalog, ensuring that AI-generated names and descriptions were consistent with the brand voice.
- Vector Search—By implementing a k-nearest neighbor (KNN) search, we enabled the model to find similar products and craft product descriptions that were consistent with those in the company’s existing catalog.
- Enhanced AI-generated Descriptions—Rather than regurgitating keywords, our solution drew from the closest product matches to generate accurate descriptions that felt both natural and on-brand.
To bring the vision to life, we leveraged
Hugging Face Transformers / Google PaliGemma / Python / Azure Databricks / Azure OpenAI GPT-4 Vision.
Let's Talk
What we built for this global retailer is just one way EDJE can solve a business challenge. If your company has a business opportunity that can be solved with technology, let’s talk. EDJE’s AI-powered solutions are designed to make your processes faster, smarter, and more impactful.
