I’d like to share the key customer service use cases that we believe will be transformed and improved by generative AI and the types of business outcomes that can be accelerated with this exciting technology. Our omnichannel cloud contact center service Amazon Connect is built from the ground up to deliver real world results with artificial intelligence and machine learning. Customers have used it to achieve
outcomes such as reduced handle times, improved agent and manager efficiency, and improved customer satisfaction. Amazon Connect’s AI/ML powered capabilities are easy to use for specific use cases such as conversational analytics with Amazon Connect Contact Lens, Agent Assist with Amazon Connect Wisdom and conversational bots with Amazon Lex. We are extremely excited about the potential of generative AI and are investing heavily right now in large language models or LLMs, for key contact
Overview of Amazon’s approach to customer service
center use cases, while ensuring the right level of security, accuracy, observability and removing bias to make sure generative AI outcomes can be both trusted and valuable. There are three major areas, agent assist, manager assist and customer self-service. Where we see an opportunity to use LLMs to build on Amazon Connect’s existing built-in AI/ML capabilities to drive immediate business value. I’m gonna walk
you through an art of the possible demo. The first area of opportunity is to expand and enhance agent assist capabilities to generate real time responses and provide suggested actions to help agents solve customer problems quickly and accurately. To show you an art of the possible with generative AI, in this
example, I’m an agent that’s taking an incoming customer chat in Amazon Connect. Trained ML models listen to the customer and capture their intent. Then generative AI synthesizes and recommends a response and actions for the agent to take based on company content like knowledge articles and internal support documents. This helps agents serve customers quickly and accurately whether it’s helping cut ramp up time for new agents or ensuring experienced agents have answers at their fingertips.
Key Innovations in Amazon’s AI Customer Service
By having a human in the loop here, the risk of delivering incorrect statements to customers is minimized. Agents can choose to edit the recommended response before it gets to the end customer. The exciting potential of LLMs is that they can leverage data from multiple sources like purchase history and the customer’s profile to further personalize generated responses. The second area of opportunity with generative AI is to assist managers with enhanced realtime and post contact analytics. Today, Amazon
AI-driven chatbots and virtual assistants
Connect Contact Lens categorizes customer issues, assesses customer sentiment and intents in real time, transcribes and summarizes customer conversations and can automate the scoring of agent performance. There’s an opportunity to use LLMs to generate concise summarization of long conversations that capture the most important information. To show you what this could look like, let’s take a look at a transcript of a conversation between an agent and a customer about setting up a new credit card. Generative AI can
analyze the transcript and generate a short summary of the conversation including the issue and outcome of the call. This can reduce the time agents and managers spend taking or reviewing notes, and the summaries are also helpful for sharing context when transferring contacts between agents. Generative AI can also help managers review contacts to efficiently identify agent coaching opportunities. Here we’re
looking at another transcript about a customer booking a vacation package. Generative AI automatically provides answers to questions like, did the customer leave the call satisfied as part of a pre-filled evaluation form. Along with rationale for each answer. Generative AI can also be used to identify agent coaching opportunities based on the transcript, saving manager time. For this manager assist use case, there is still a human in the loop to review the generated output. Learning from agent and manager
assisted generative AI use cases will help improve the third area of opportunity. Customer self-service experiences. Generative AI will make it easier to deliver LLM powered recommendations for automated workflow configurations. Understand the complex nuances of a customer’s intent and build natural sounding self-service experiences. Today, customers can use Amazon Lex with Amazon Connect to build self-service capabilities such as chatbots and voice assistance that intelligently route calls, answer questions with natural responses and automate common tasks. With generative AI, instead of a contact
center manager spending time specifying intents, there is an opportunity to offer more natural sounding conversational capabilities that are flexible and able to generalize, handle or process complex scenarios without the need for companies to define every aspect of the conversation. This will help contact centers design and build self-service experiences more quickly. Let me show you what an improved self-service experience could look like. Like in this insurance chatbot example where a customer is asking about
Benefits of AI in Customer Service
specific insurance policies. These three generative AI use cases, agent assist, manager assist, and self-service interconnect. By learning what drives great agent outcomes and layering in analytics for managers, knowledge content can be improved and this can drive improved self-service and agent assisted interactions. In a customer facing environment like a contact center, security, accuracy, observability and removing bias are paramount, especially for the self-service use case I just showed.
Having a human in the loop can help ensure the accuracy of LLM generated content, particularly in sensitive and critical customer interactions. Transcription underpins everything. It’s what feeds degenerative AI models to generate a recommended response for an agent or a self-service response. It’s really good, but can be impacted by external factors like audio quality, different accents, and background noise. By having a human in the loop, an agent or manager can refine the prompt and or output. As we learn how agents augment generated content and we start to deploy LLMs for self-service use cases, it’ll be critical to have the right controls and transparency for companies to ensure safe, accurate responses.
Challenges and Limitations
To mitigate cost concerns, we hear from customers about using LLMs. We believe LLMs will make the most sense for high value use cases that aren’t fully realized with existing AI/ML capabilities. As you investigate how generative AI could be beneficial for your contact center, I encourage you to think about your top use cases for generative AI and identify the key outcomes you’re looking to drive, such as reduced handle time or improved first call resolution. Today, thanks to how open and extensible Amazon
Case Studies and Real-World Applications
Connect is, customers and partners are empowered to easily integrate new technologies like LLMs in their contact centers. If you’re ready to experiment with generative AI in your contact center, talk to your AWS account team about getting started with Amazon Connect and Amazon Bedrock. We hope you’ve enjoyed this demonstration of the art of the possible with generative .
AI for customer service use cases including agent assist, manager assist, and self-service. We are excited about the potential generative AI and how it can continue to enhance Amazon Connect’s built-in AI/ML capabilities.
