🇲🇾 🎧 Customer serviceFrom a real public post
TNB's SARA assistant resolves 15,000 queries a month
Within 10 months of launch, SARA resolves about 15,000 queries a month through self-service with a 91% positive sentiment score (AWS case study).
Done withAmazon BedrockAmazon Bedrock + Claude 3.5 (with Uberfusion)
Reported outcome15,000 queries/month self-served; 91% positive sentimentSource: Vendor case study
1The problem
Tenaga Nasional serves over 10.5 million accounts in several languages, and its older keyword FAQ bot could not handle natural questions.
2Agent setup
SARA, a generative AI assistant built on Amazon Bedrock with Claude, a knowledge base and guardrails, answers FAQs, updates customer details and takes outage reports, handing over to Careline agents when needed.
3Result
Within 10 months of launch, SARA resolves about 15,000 queries a month through self-service with a 91% positive sentiment score (AWS case study).
4Copy-paste prompt
Using only this FAQ and outage procedure [paste], answer customers in BM or English. For an outage, collect account number, address and landmark, then confirm the report and give the reference format.