Business Background

Kutes Metal is one of Turkey’s leading industrial metal casting and production companies, serving clients across multiple industries. The company handles a large volume of customer calls regarding product feedback, quality issues, and service complaints. With a strong commitment to quality and customer satisfaction, Kutes Metal sought to modernize its complaint handling processes through intelligent automation.

Business Challenge

Kutes Metal faced significant operational challenges in managing customer complaint calls effectively. The primary issues included:

– Manual call handling: All complaint calls were previously processed by human agents, leading to long response times and inconsistent data entry

– High call volume: During peak production periods, the number of incoming complaints exceeded available operator capacity

– Operational inefficiency: Complaint information was manually logged and later transferred into internal systems, creating delays

– Limited availability: Operators were unavailable outside of business hours, leading to missed or late complaint registration

– Quality control: Human inconsistency resulted in incomplete or inaccurate records

Project Objective

Kutes Metal aimed to automate the complaint intake process using AI-driven natural language understanding to provide a human-like voice interaction, automatically capture complaint details, and deliver structured records to responsible teams in real time.

Solution Architecture

Kutes Metal deployed the Callie AI-powered voice assistant hosted on AWS, built around Amazon Bedrock for natural language understanding and generation. The assistant receives phone calls through a SIP trunk, converts speech to text, processes meaning via a custom Bedrock LLM, and replies to customers in real time with synthesized voice. It also records complaint details and sends structured summaries to relevant staff by email.

Solution Workflow

1. Inbound Call Reception

• Customer calls the Kutes Metal complaint hotline.

• The call is routed via SIP trunk to the AI assistant system hosted in AWS.

2. Speech-to-Text (STT)

• Incoming audio is transcribed to text through an integrated speech recognition engine.

• Transcriptions are passed to the LLM pipeline.

3. Understanding & Response Generation with AWS Bedrock

• The transcribed text is sent to a custom foundation model hosted on Amazon Bedrock.

• The model interprets customer intent, extracts key complaint details (product, issue, urgency, contact info), and generates a contextually appropriate, empathetic response.

• Conversation state and context are maintained throughout the call.

4. Voice Synthesis (TTS)

• The assistant converts the generated response text into audio and speaks back to the caller naturally and in real time.

5. Complaint Logging & Notification

• The system automatically logs the entire conversation transcript, identified complaint details, and caller metadata.

• A structured complaint report is generated and emailed to the responsible team using a secure messaging service.

AWS Services Used

• Amazon Bedrock – Custom foundation model hosting for natural language understanding, intent extraction, and empathetic response generation.

• AWS Lambda – Event-driven workflow orchestration for call processing, transcription routing, and notification triggers.

• Amazon S3 – Scalable storage for conversation transcripts, complaint records, and model artifacts.

• Amazon SES – Low-cost, secure email notification service for delivering structured complaint reports to responsible teams.

• SIP Trunk Integration – Inbound call routing to the AI assistant system hosted within AWS infrastructure.

Results and Benefits

The Callie AI-powered voice assistant delivered measurable improvements across Kutes Metal’s complaint handling operations:

Business Outcomes:

• Fully automated complaint handling through natural voice interaction, eliminating manual call processing.

• 24/7 call availability and instant logging, ensuring no complaint is missed regardless of time of day.

• Human-like and empathetic communication style powered by AWS Bedrock, maintaining customer trust and satisfaction.

• Accurate, structured complaint reports delivered instantly via email to responsible teams.

• Improved internal tracking and accountability through automated record-keeping and reporting.

Cost Efficiency:

• ~35% reduction in operational cost, while enabling 24/7 automated service and improving response accuracy.

• No hardware or fixed telephony maintenance required with the cloud-native architecture.

• Scalable compute (Bedrock + Lambda) ensures cost-per-use efficiency.

• Automated complaint documentation reduces manual labor costs significantly.

Lessons Learned

Technical Insights:

• Low-latency orchestration between STT, Bedrock, and TTS pipelines is key for real-time voice interaction.

• Prompt design and LLM fine-tuning were essential to maintain a polite, human-like tone during calls.

• Conversation context tracking within Bedrock sessions improved complaint detail extraction accuracy.

• Secure architecture design ensured all voice and text data remained within AWS infrastructure, maintaining compliance.

Organizational Insights:

• Introducing AI-assisted communication required staff training and user awareness programs.

• Early feedback loops from real callers were invaluable for model tuning and improving response quality.

• Automating complaint reporting improved coordination between quality and operations teams.

This case study demonstrates how AI-powered voice assistants—built on AWS managed services including Amazon Bedrock—can transform traditional complaint handling processes in industrial manufacturing, delivering significant cost savings while enhancing customer experience and operational efficiency.