Lotte Department Store Builds 'AI Biz Center', a Partner Proposal AI Agent Based on Amazon Bedrock

Lotte Department Store has built 'AI Biz Center', a generative AI agent service that supports the entire process from partner store entry proposals to connecting with the responsible departments. This service is based on Amazon Bedrock and integrates information search, proposal writing, and department matching into a single conversational flow.

Service Composition and Architecture

Users can handle tasks such as store entry proposals and marketing partnerships through chatbots on the web and mobile apps. The service utilizes the Anthropic Claude Haiku 4.5 model to classify intent and analyze multimodal data such as PDFs, images, and links to automatically extract proposal information. It combines Retrieval-Augmented Generation (RAG) with Amazon OpenSearch Service to recommend candidate departments, and uses Amazon Bedrock Guardrails to block harmful content and prompt attacks.

Key Technical Challenges and Solutions

To resolve false positives from AWS WAF during file uploads, the upload path was redesigned to use direct uploads via Amazon S3 Presigned POST. Additionally, to improve the accuracy of department matching, a multi-stage search pipeline was applied, combining BM25 keyword search and k-NN vector search with LLM-based normalization. This improved matching accuracy from 25.58% when using hybrid search alone to 83.72%, a 58.14 percentage point improvement.

Implementation Results

In the two months following the service launch, the number of proposal submissions increased by 160% compared to the same period last year. Approximately 25% of proposal customers used the AI proposal writing feature, confirming that natural language-based workflows have become established in actual operations. Internally, repetitive information verification and transmission tasks have decreased, creating an environment where Merchandisers (MDs) can focus on discovering new brands and reviewing proposals.