The Challenge: Manual Feedback Analysis at Scale
Imagine being a Product Manager at one of the world's largest shipping companies, responsible for analyzing hundreds of customer comments every two weeks. You're spending hours, sometimes entire days, manually reading through CSV exports, categorizing sentiment, and identifying themes. This was the reality for Hapag-Lloyd's Digital Customer Experience team until they discovered the power of AI-driven automation.
Hapag-Lloyd, with their fleet of 313 container ships serving over 600 ports worldwide, processes feedback from hundreds of thousands of monthly users across their web and mobile applications. Their manual approach to feedback analysis was valuable but unsustainable—limiting their ability to respond quickly to customer needs and scale their insights process.
The AI-Powered Solution: A Complete Workflow Transformation
The company's engineering team, split between Hamburg and Gdańsk, developed a comprehensive solution using Amazon Bedrock, Elasticsearch, and open-source frameworks like LangChain and LangGraph. This isn't just another AI implementation—it's a complete reimagining of how customer feedback can drive product decisions.
Key Components of Their Architecture
Daily Automated Processing: Instead of biweekly manual exports, AWS Lambda functions now fetch and process new feedback entries daily. Amazon Bedrock automatically classifies sentiment (positive, negative, mixed, or neutral) for each comment, streamlining the entire analysis pipeline.
Real-Time Interactive Exploration: Stakeholders access live feedback insights through OpenSearch Dashboards, providing bird's-eye views of user sentiment, ratings, and trends. Product Managers can now drill down from high-level visualizations to specific features or individual comments, enabling targeted root cause analysis.
AI-Powered Internal Chatbot: Perhaps most impressively, they built an internal chatbot that queries their feedback database using natural language. Product managers can now ask questions like "What pain points do customers mention most often?" and receive instant, context-rich answers.
Automated Insights Reports: Every two weeks, the system automatically generates comprehensive reports with key metrics, highlights, and sentiment breakdowns, feeding directly into sprint planning and roadmap discussions.
The Technical Foundation: Smart Orchestration with LangChain and LangGraph
What makes this solution particularly interesting for prompt engineers and AI practitioners is their thoughtful approach to orchestration. They use LangChain for data processing workflows—handling sentiment analysis, embedding generation, and system integration—while leveraging LangGraph for their multi-agent chatbot architecture.
Their choice of Claude Sonnet 4.6 via Amazon Bedrock was strategic: the model excels in multi-turn conversational exchanges and agentic workflows, making it perfect for their internal chatbot that requires reliable performance across various interaction patterns.
The team also implemented Bedrock Guardrails to ensure responses align with brand standards and compliance requirements—a crucial consideration for enterprise AI deployments.
Results: From Reactive to Proactive
The transformation is remarkable. What once took hours or days of manual work now happens automatically in the background. Product Managers can focus on strategy and innovation rather than operational analysis. The system enables faster decision-making and more responsive product development.
Most importantly, this solution demonstrates how AI can amplify human expertise rather than replace it. The engineering team didn't just automate their old process—they reimagined what customer feedback analysis could become in an AI-native organization.
Key Takeaways for AI Practitioners
This case study offers several valuable insights for anyone building AI-powered business solutions:
- Think Beyond Automation: Don't just digitize existing processes—reimagine what's possible with AI
- Choose the Right Models: Model selection should align with specific use cases (Claude Sonnet 4.6 for conversational workflows in this case)
- Implement Proper Orchestration: Use frameworks like LangChain and LangGraph to build maintainable, scalable AI workflows
- Plan for Enterprise Requirements: Include security, compliance, and brand alignment from the start with tools like Bedrock Guardrails
Hapag-Lloyd's journey from manual feedback analysis to AI-native operations shows how thoughtful implementation of generative AI can transform business processes while enhancing human decision-making capabilities.
This case study is based on the original article by Aamna Najmi, published on the AWS Machine Learning Blog. The full technical implementation details and architecture diagrams are available in the original article.