Back office ops · Production

Hapag-Lloyd uses Amazon Bedrock to transform customer feedback into actionable insights

The problem

Hapag-Lloyd's product teams manually analyzed customer feedback by exporting CSV files and hand-categorizing sentiment every two weeks—a process that took hours or days and was too slow and inflexible to scale with the volume of feedback from hundreds of thousands of monthly users.

Workflow diagram · grounded in source
1
Customer submits feedback
trigger
“Our web and mobile applications serve hundreds of thousands of customers each month. Users can leave a rating plus text comments, helping us understand user experience and improve services.”
2
Daily Lambda ingestion to S3
integration
“A AWS Lambda function runs once per day to fetch the new feedback entries from the feedback repository into Amazon S3”
3
Bedrock sentiment classification
ai_action
“We use Amazon Bedrock to classify sentiment (positive, negative, mixed, or neutral) for each open comment, streamlining downstream analysis”
4
Index in OpenSearch
integration
“Processed records are indexed in Amazon OpenSearch Service, serving both as our full-text search engine and vector database”
5
Chatbot answers stakeholder queries
ai_action
“Product managers and support teams can ask natural-language questions, for example, "What pain points do customers mention most often?" and receive instant, context-rich answers”
6
Biweekly insights report
output
“Every two weeks, a second Lambda function aggregates and analyzes the latest feedback trends. It generates a concise report with key metrics, highlights, and sentiment breakdowns. The report is automatically delivered to our Product Mana…”
7
AI reports track post-release reactions
feedback_loop
“This feature was prioritized directly in response to a high volume of negative user feedback highlighting the lack of a preview capability. After its release, AI-driven reports allowed us to track user reactions in detail. Feedback relat…”
Reported outcome

The automated pipeline processes over 15,000 feedback items per month at 95% sentiment classification accuracy, delivering structured summaries in seconds instead of hours and enabling product decisions within days rather than weeks.

Reported metrics
Feedback items processed per monthover 15,000
Sentiment classification accuracy95%
Time to structured summariesin seconds
Decision cycle timewithin days rather than weeks
Show all 5 reported metrics
feedback items processed per monthover 15,000
sentiment classification accuracy95%
time to structured summariesin seconds
decision cycle timewithin days rather than weeks
customer sentiment trendmore positive comments and a noticeable reduction in negative feedback
Reported stack
Amazon BedrockElasticsearchLangChainLangGraphAmazon OpenSearch ServiceAmazon S3AWS CloudFormationBedrock GuardrailsAmazon CloudWatchAWS CloudTrailClaude Sonnet 4.6
Source
https://aws.amazon.com/blogs/machine-learning/how-hapag-lloyd-uses-amazon-bedrock-to-transform-customer-feedback-into-actionable-insights/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The automated pipeline processes over 15,000 feedback items per month at 95% sentiment classification accuracy, delivering structured summaries in seconds instead of hours and enabling product decisions within days ra…

What tools did this team use?

Amazon Bedrock, Elasticsearch, LangChain, LangGraph, Amazon OpenSearch Service, Amazon S3, AWS CloudFormation, Bedrock Guardrails, Amazon CloudWatch, AWS CloudTrail.

What results were reported?

Feedback items processed per month: over 15,000; Sentiment classification accuracy: 95%; Time to structured summaries: in seconds; Decision cycle time: within days rather than weeks (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Customer submits feedback → Daily Lambda ingestion to S3 → Bedrock sentiment classification → Index in OpenSearch → Chatbot answers stakeholder queries → Biweekly insights report → AI reports track post-release reactions.

Related back office ops cases