Omlet is a leading online pet products retailer with a diverse catalogue spanning automatic chicken coop doors, modular pet housing systems, and outdoor enclosures. Their customer service team handled thousands of daily enquiries requiring accurate, product-specific knowledge. New Gradient built an LLM customer service automation system using retrieval-augmented generation to deliver instant, grounded responses across 94,000 indexed documents.
Thousands of daily queries, scattered product knowledge
E-commerce businesses face enormous pressure managing product-related customer queries. For Omlet, the challenge was particularly acute: thousands of daily enquiries spanning product specifications, assembly guidance, compatibility questions, and troubleshooting requests. Each query required accurate, product-specific knowledge that traditional support processes struggled to deliver consistently.
Customer service teams spent hours researching answers across scattered documentation, previous correspondence, and product manuals. Response times stretched from minutes to hours, customer satisfaction suffered, and operational costs continued to climb. Every query demanded specialist knowledge that was difficult to maintain across a growing support team.
Retrieval-augmented generation with fine-tuned language models
The solution ingests and indexes Omlet's entire knowledge base: product documentation, historical customer correspondence, assembly instructions, FAQs, and technical specifications. This created a comprehensive, searchable repository of 94,000 documents and past responses.
Vector embeddings convert this documentation into a format that enables semantic search tailored to the company's specific product domain, allowing the system to understand the meaning behind customer queries rather than relying on keyword matching alone. When a customer asks about compatibility between products or troubleshooting a specific issue, the retrieval system identifies the most relevant documentation across thousands of sources in milliseconds.
Fine-tuned large language models generate responses that match Omlet's tone and style while maintaining factual accuracy. The models learned from years of successful customer interactions, understanding not just what information to provide but how to communicate it effectively. The RAG architecture ensures responses are grounded in actual product documentation rather than generated from the model's general knowledge, virtually eliminating hallucination.
60% of queries automated, 44% faster response times
60% of product queries are now answered fully automatically, with response times dropping from hours to seconds. Customer service agents receive AI-generated draft responses for the remaining queries that they can review, refine if needed, and send, multiplying their effective capacity while maintaining the human touch for complex or sensitive enquiries.
The 94% accuracy rate means staff can trust the system's suggestions, with time spent answering queries reduced by 44%. For straightforward requests (product dimensions, compatibility checks, standard troubleshooting), the AI handles the heavy lifting entirely. This frees human agents to focus on complex cases that genuinely require their expertise and judgement.
The system continues to learn as new products launch and documentation updates. Customer interactions identify gaps in product information and common points of confusion, creating a feedback loop that improves both the AI system and the underlying documentation. The knowledge base now spans 94,000 indexed documents and grows with each interaction.