NCS GenAssist Cuts Singapore MOM Contact Center AHT 12%, ACW Over 50%

Title

NCS GenAssist Cuts Singapore MOM Contact Center AHT 12%, ACW Over 50%

Content

NCS has augmented the Singapore Ministry of Manpower (MOM) Contact Center with its GenAssist generative-AI suite, reducing average call handling time by 12% and after-call work by more than 50% while raising overall agent productivity by 6%. The deployment, built on AWS Bedrock with Anthropic's Claude Instant as the primary LLM, layers three purpose-built modules on top of MOM's existing contact-center stack rather than replacing it. The GenAssist Onboarding Chatbot uses a Retrieval-Augmented Generation (RAG) pattern built with LangChain, Amazon Kendra, and Claude Instant; internal knowledge documents live in Amazon S3 and are indexed by Kendra, and the chatbot returns answers to complex queries in seconds at 93% accuracy. The GenAssist Audio Service streams call audio through ins8.ai, NCS' hyperlocal speech recognition engine, achieving 95% transcription accuracy on the colloquial Singaporean accents that callers use when phoning about manpower and employment matters. The GenAssist Summarizer then feeds the cleaned transcript with an optimized prompt back into Claude Instant, where human-in-the-loop evaluation has measured summarization accuracy at 99% — letting agents complete the same post-call summarization in roughly half the time. The deployment responds to a real operational squeeze: MOM's contact-center agents had to draw on deep domain knowledge across multiple legacy knowledge sources to resolve multi-faceted manpower and employment queries, stretching both call duration and the after-call documentation that follows. NCS — an AWS Advanced Tier Partner with a 13,000-person team across 58 specializations and a launch partner of the AWS Generative AI Partner Innovation Alliance — designed GenAssist to plug into MOM's existing infrastructure without a rip-and-replace migration. Bedrock's data-protection posture — neither storing nor logging customer prompts and completions, and not reusing them to train any AWS or third-party models — was a key compliance fit for a public-sector deployment handling citizen queries. Looking ahead, NCS is working on real-time transcription analysis that proactively pushes context-relevant information to agents during live calls to further compress handling time.