What challenges did Infosys face before AI implementation?

Support agents needed to locate relevant past call transcripts and provide guidance to meter technicians. 60-70% of issues were repetitive, average handling time for top categories was over 5 minutes, and hiring and training more agents was costly and not scalable.

What was the impact of the AI solution?

The enterprise AI assistant now handles 70% of previously human-managed calls. Average handling time for the top 10 issue categories dropped from over 5 minutes to under 2 minutes, a 60% improvement. Issues requiring human intervention fell from 30-40% to 20% within 6 months. Customer satisfaction scores increased by 30%.

Source

How Infosys Topaz leverages Amazon Bedrock to transform technical help desk operations

Use case typeKnowledge managementUpdated Jun 13, 2026

Infosys built an AI-powered technical help desk for a large energy supplier to ingest past and new call transcripts, summarize conversations, classify relevance, retrieve similar issues, and help support agents resolve meter technician calls faster. The system uses Amazon Bedrock with Claude Sonnet, Amazon S3, AWS Step Functions, AWS Lambda, Amazon OpenSearch Serverless, Amazon Titan Text Embeddings, AWS Secrets Manager, AWS KMS, AWS IAM, DynamoDB, and a Streamlit frontend. It implements role-based access control, asynchronous transcript processing, caching, FAQ retrieval, and a RAG-style knowledge base for agent assistance.

Organization: Infosys
Industry: Energy & Utilities
Location: India
Published: August 2025

Reported outcomes

70%

quantified impactOther quantified impact

5 minutestime2 minutestime−60%time40%quantified impact20%quantified impact

Strategic outcomes

Scale & capacityHandled previously human-managed callsSpeed & agilityReduced issue handling timeCustomer experience & trustImproved customer satisfactionScale & capacityReduced human intervention needs

Primary read

Use case focus

Showing 3 of 3

1Customer Support
2Knowledge Management
3RAG

Support agents needed to locate relevant past call transcripts and provide guidance to meter technicians.
60-70% of issues were repetitive, average handling time for top categories was over 5 minutes, and hiring and training more agents was costly and not scalable.

Build a knowledge base from call transcripts stored in S3.
Use Claude Sonnet on Amazon Bedrock to summarize and classify transcripts as relevant or irrelevant, then generate embeddings with Amazon Titan Text Embeddings and store them in Amazon OpenSearch Serverless for retrieval.
Use Step Functions and Lambda to orchestrate near-real-time ingestion, implement role-based access control, maintain metrics in DynamoDB, and expose the assistant through Streamlit.

Technologies

Amazon Bedrock Anthropic Claude Sonnet Amazon S3 AWS Step Functions AWS Lambda Amazon OpenSearch Serverless Amazon Titan Text Embeddings V2 AWS Secrets Manager AWS KMS AWS IAM DynamoDB Streamlit

The enterprise AI assistant now handles 70% of previously human-managed calls.
Average handling time for the top 10 issue categories dropped from over 5 minutes to under 2 minutes, a 60% improvement.
Issues requiring human intervention fell from 30-40% to 20% within 6 months.
Customer satisfaction scores increased by 30%.

Architecture

A generative AI technical help desk architecture ingests call transcripts from Amazon S3, uses AWS Step Functions and AWS Lambda for orchestration, applies Anthropic Claude Sonnet on Amazon Bedrock to summarize and classify transcript relevance, stores embeddings in Amazon OpenSearch Serverless using Amazon Titan Text Embeddings, secures data with AWS Secrets Manager, AWS KMS, and AWS IAM, tracks user and query data in DynamoDB, and presents the assistant in Streamlit with role-based access control and caching.

Sources & evidence1

Groundedness: 5/5Type: Blog PostPublished: Aug 21, 2025Publisher: Amazon Web ServicesEvidence: VendorConfidence: Medium

Primary source

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases

Customer evidence