Published: Nov 14, 2024

innovating customer support: NCS generative AI enhancements for the Singapore Ministry of Manpower’s contact centre

Generative AI is a fast-evolving technology that can deliver many transformational benefits to businesses and organisations. It can help to reinvent customer experiences, increase employee productivity and optimise business processes.

However, one of the challenges faced by many organisations when adopting generative AI is that they have existing infrastructure that may not be easily upgraded with generative AI capabilities. Upgrading often requires substantial additional investment, significant implementation effort and long time-to-value.

In this post, we describe how NCS used AWS cloud services complemented with ins8.ai, NCS’ hyperlocal speech recognition solution, to add generative AI features to the Singapore Ministry of Manpower's Contact Centre. The use of generative AI capabilities has helped to unlock benefits to streamline Contact Centre operations and improve customer experiences.

The Ministry of Manpower (MOM) Contact Centre is a key touchpoint for the businesses, human resource professionals, employees and the general public as it manages queries on manpower and employment related issues.

The queries received from the customers can be multi-faceted or cross-cutting. Addressing such queries with precision requires the agents to possess a deep understanding of the subject matter and ability to analyse the needs of the customer. When determining the needs of the customer, the agents would reference past records from the caller and documents from internal systems to determine suitable advice. Referencing multiple information sources would sometimes stretch out the duration of the call. After-call work, which includes call summarisation and other documentation, impacts the total amount of time taken by agents to handle each call, especially for complex cases.

To help MOM’s Contact Centre reap the benefits of generative AI, NCS developed an approach to augment the Contact Centre with generative AI features that addresses the challenges without the need for an expensive rip and replace strategy.

As illustrated in the architecture diagram below, new modules were developed to augment and integrate with the existing Contact Centre solution.

Figure 1. Architecture for NCS’ GenAssist solution deployed for MOM

When engaging callers on complex queries, agents need to have in-depth understanding of the relevant subject matter to provide accurate advice and good customer experience. Onboarding new agents to the level required for the job requires a considerable amount of time and resources to train them well.

NCS pioneered the use of an onboarding chatbot, which accelerates the onboarding process while complementing trainer effort.  Agents can use the Chatbot to assist them in looking up answers to complex customer queries during training. The GenAssist Onboarding Chatbot is implemented using the Retrieval Augmented Generation (RAG) approach with LangChain, Amazon Kendra and Anthropic’s Claude Instant model in Amazon Bedrock. Internal knowledge documents are stored in Amazon Simple Storage Service and indexed by Amazon Kendra. The solution is able to generate responses to complex queries within seconds with an accuracy of 93%.

Conversation audio for each call between the caller and Contact Centre agent is captured by the GenAssist audio service. Callers often use colloquial accents in their conversations with Contact Centre agents, which can be challenging for most commercial speech-to-text products to transcribe accurately. Ins8.ai is NCS’ hyperlocal speech recognition engine that performs hyperlocal voice transcription at scale with high accuracy.  The audio service sends the audio stream to ins8.ai to transcribe call conversations in real-time with 95% accuracy.

The Summariser Service retrieves the call transcript from ins8.ai, preprocesses it and sends it together with an optimised prompt to the Large Language Model (LLM) hosted on Amazon Bedrock to perform the summarisation task. In this case, Anthropic’s Claude Instant model was selected as it strikes the balance between speed, accuracy, and cost-effectiveness.

While there are many available Large Language Model (LLM) evaluation metrics, different evaluation metrics have different advantages and limitations. For the first release, NCS used a human-in-the-loop approach to evaluate the generated summary. Using this approach, the Claude Instant model achieved a 99% accuracy score. NCS continues to research the most suitable text summarisation evaluation metric so that this process can be automated in the future.

The Contact Centre agent uses the GenAssist Summariser plugin in the Agent Desktop Client to retrieve the generated summary, review it for accuracy, and make necessary edits if required before logging the summary into the CRM system. With the generative AI powered summarisation tool, agents can now perform this task twice as fast.

Data Protection and Privacy

To ensure data protection and privacy, Amazon Bedrock neither stores nor logs customers’ prompts and completions. Furthermore, Amazon Bedrock does not use customers’ prompts and completions to train any AWS models or distribute them to third parties. Model providers also do not have access to Amazon Bedrock logs or to customers’ prompts and completions.

Benefits

With the new generative AI enabled Contact Centre, MOM achieved key improvements across various critical metrics of the Contact Centre. The average call handing time (AHT) was reduced by 12%. The average after-call-work (ACW) was reduced by more than 50%.

Harnessing the AI tools to streamline the tasks of agents simplified their responsibilities and empowered the agents to fully focus on connecting with each caller, fostering a deeper understanding of their concerns and alleviating anxiety while effectively resolving queries.

This change has also cultivated a more positive work environment and contributed to heightened job satisfaction among Contact Centre agents, along with an improvement in overall productivity of 6%.

Roadmap

NCS is looking at further improving the solution with roadmap features. For example, using real-time transcription to detect and proactively push context-relevant information and content to the agent in real-time that can help to reduce call handling time.

Generative AI is a fast-moving technology that brings new advancements at a rapid pace. It holds a lot of promise but is also capable of delivering real-world business benefits today.

In this post, NCS has shown that with some creativity in augmenting existing infrastructure with new capabilities, enterprises and businesses can already take advantage of generative AI to help enhance customer experiences, increase employee’s productivity and optimise business processes without a costly infrastructure overhaul.

If you are looking to modernise your Contact Centre with generative AI capabilities, NCS has extensive experience providing consulting, integration, as well as managed service desk and Contact Centre services.

By Royston Bok, Senior Director, AI Innovations, NCS Group
By Kok Soon Toh, Enterprise Architect, NCS Group
By Henry Soh, Lead IT Architect and Generative AI Specialist, NCS Group
By Voon Wong Wong, Principal Partner Solutions Architect, AWS
By Vincent Oh, Senior Specialist Solutions Architect (Generative AI), AWS