Hi everyone. Greetings for the day. Welcome to Con 42, machine Learning 2025. Thanks for taking out the time and joining me in this session. I am Neha Hacker, an associate solutions architect with AWS, and I'm based out of London. I predominantly work with financial services and FinTech customers within the UK to help them design and optimize their architectures on AWS. Today I'm here to talk about how you can transform your customer interactions into actionable insights in near real time, whether it's phone calls or chat conversations. This solution helps you understand what your customers are saying, how they are feeling, and what topics matter the most. Let's walk through a typical day at a bank's call center. It's 9:00 AM the morning starts with an unexpected surgeon calls. Is it a system issue, a new policy causing confusion? Without new real time analytics, no one knows. It is now 11:00 AM Multiple customers are raising concerns. Agents sense a pattern, but there's no way to confirm or quantify the trend and valuable time is lost. It is now 2:00 PM The quality team is manually reviewing yesterday's interactions. Critical customer feedback from this morning won't be analyzed until next evening. It is now 4:00 PM management reviews, daily metrics, showing declining performance, but lacks the insights to understand root causes or take preventative action. The result a day filled with reactive decision making, missed improvement opportunities. Frustrated customers and agents and lost business intelligence. Now let's examine what happens when you can't analyze your customer interactions Every day, thousands of customer interactions occur, but most of this valuable feedback remains unanalyzed and unutilized. This creates significant delays in identifying emerging issues, allowing problems to multiply before we can address them. The result unnecessary costs from handling preventable repeat calls, each representing a customer issue that could have been resolved earlier without analysis of this conversations. We miss crucial insights that could help prevent future issues and improve service. This isn't just about missing data, it's about every frustrated customer, every missed opportunity, and every preventable problem that slips through the cracks because we are flying blind. So let's look at how this solution can help solve this problem. It creates complete visibility. Remember those thousands of unanalyzed conversations, our solution ensures every interaction is captured. And analyzed automatically. No more blind spots, no more missed insights. It gives near real time intelligence. Instead of waiting days to identify issues, you'll know immediately what's happening in your customer conversations. Think of it as giving your team superhuman listening abilities. It does intelligent analysis. Gone are the days of manual reviews. Our AI engine automatically understands what customers are talking about and how they are feeling across every interaction. From audio calls and chat support, it provides predictive action. This isn't just about faster analysis, it's about getting ahead of issues. You'll spot trends as they emerge, not after they have become problems. Transform from reactive to proactive service, and it also helps generate strategic insights. Finally, all of this intelligence feeds in, feeds directly into your decision making. Every customer conversation becomes a source of strategic insight for your business. This isn't just an analytics tool, it's a complete transformation in how you understand and serve your customers. In banking where every customer interaction matters, this level of insight creates a genuine competitive advantage. Now, let's walk through how our solution transforms customer conversations into actionable intelligence. It starts with capturing every customer interaction, whether it's voice calls from your contact center. Or chat conversations with your service team. Every customer touchpoint is automatically collected and securely stored. This ensures we don't miss any valuable customer feedback. Next, our processing engine goes to work. Voice calls are automatically converted to text while chat messages are extracted and stored. This happens in near real time, creating a unified format for all customer interactions. Think of it as translating every customer conversation into a language or analytics can understand. This is where the intelligence happens. Our AI engine analyzes every conversation. What are the customers talking about? How are they feeling? What patterns are emerging? This gives us deep insights into customer needs and concerns as they happen. Finally, we transform this intelligence into action. Through interactive dashboards and reports your team scan spot trends, instantly identify issues early, make data-driven decisions. The beauty of this solution is its simplicity. It's completely automated. Running 24 by seven, taking every customer interaction into business intelligence. No manual processing, no sampling, no delays, just continuous near real time insights from every customer conversation. Think of it as giving your organization a super human, the ability to listen to, understand, and act on every customer interaction simultaneously. Now, let's build this architecture using AWS. We begin with two types of customer interactions, voice calls store does, audio files, and Amazon S3 chat conversations from your support team. For voice calls, we have built an automated pipeline. When a new audio file lands in S3, it triggers a Lambda function. This transcribed Lambda function calls Amazon transcribe. The resulting transcript is sowed in an S3 bucket. Next, we handle both the transcribed calls and chat logs. Another S3 activates our sentiment Lambda function. This function leverages Amazon comprehend, which goes beyond basic text analysis. Comprehend understands banking context automatically identifies key topics like loan applications or account issues. It detects customer sentiment throughout the conversation, helping you spot satisfaction trends. It helps recognize patterns and customer inquiries, helping you identify emerging issues before they escalate. The structured analysis is then stored for further use. Finally, we make all of these actionable insights actionable and accessible. Amazon QuickSight transforms this wealth of data into interactive dashboards that your teams can actually use. It offers machine learning powered anomaly detection to automatically flag unusual patterns and customer interactions. Business users can ask questions in natural language, no technical skills needed for deeper analysis, we can use Athena to run SQL queries directly on the data in the S3 bucket. What's powerful about this architecture is it is automatic and then it is near real time. From the moment a call lands or a chat closes, the analysis begins within minutes. The insights are available in your dashboards. There's no manual processing, no delays, and no sampling. Every interaction is analyzed. This solution gives you a complete, near real time view of your customer interactions. You're not just handling customer service anymore. You're turning every customer conversation into valuable business intelligence. Imagine identifying trends and issues before they become problems, understanding customer sentiment across thousands of interactions and making data-driven decisions based on actual customer conversations. Now let's look at a quick live demo. So I have created this solution on AWS console. You can also create it using infrastructure as code. So this is how AWS console looks like. You'll be able to select the region where you want to deploy your workloads or your solutions. So I have used London region here. And here you'll be able to see my recently visited AWS services. You will also be able to navigate to this by clicking on the links here, or you'll be able to search a service here and go to the service. Or if you click these buttons, you'll be able to navigate to AWS services as well. So now let's look at S3. So I have created an S3 bucket within the bucket. I have created these folders for audio files. This is where I'm gonna upload the audio file of a chat of a conversation along with its metadata. So I will just upload an audio file. Here along with its metadata, JSON file, and this is gonna upload in a fraction of seconds as you can see. Once this is done this event triggers a lambda function. Lambda function will now call Amazon Transcribe, which is a transcription service to convert speech to text. So here, if you see there's a transcription job in progress, but in the interest of time, I'm gonna show you how it looks like once the transcription is done. So here you'll be able to see the employee details, customer details, and the transcript. This information will be stored in another folder of S3 bucket called extracted files. Once this is stored here in this folder again, a lambda is triggered, which calls Amazon comprehend sentiment analysis and topic modeling. So if you see, you'll be able to see that topic modeling job is in progress. So again, in the interest of time, I'm gonna show you how this, how the final output looks like. So it looks like this here. So you have employee Id name, customer ID name, and account type. You have the transcript. And we have done a sentiment analysis on this transcript using comprehend. So it is negative sentiment. Comprehend also gives you a score such as here it is saying it is 99.5% sure that this conversation has a negative sentiment to it. And for topic modeling, again, I have selected 10 topics to be picked up by Amazon comprehend. So you'll be able to see all of those here and the related terms to each topic. And if you scroll down, you'll see the metadata of this file. So all of this is against stored into another bucket of S3 call combined. So you have all of this here. Now I am feeding all of this data into Amazon QuickSight to create visualizations. So this, these visualizations could be created based on your requirement but for this demo, I have created few sample visuals like the one on the screen. The first one is distribution of customer sentiment. So we have gotten 61 customer conversations, either through chat messages or through audio calls. So I'm showing the sentiment across these conversations here. Here you'll be able to see the sentiment distribution across across the account type, such as which account type has a negative sentiment associated to it, or a positive sentiment associated to it. And topic analysis dashboards. I'm gonna show the dominant topic in customer conversations. So if you have. Topics such as account services for which the customers are calling you often, such as how do I open an account, how do I close an account, et cetera. You'll be able to look at these dashboards to maybe create an FAQs document so customers can help themselves rather than calling the agents at the banking call center. This way you are reducing the calls that can be that can be self-served by customers. And here in customer service performance, I have created a visual which shows the performance of an agent. So if an agent has a con consistent negative sentiment while talking to customers, you'll be able to put them on a training program. So you're making the customer, sub customer experience positive. Some other features of QuickSight are you'll be able to connect to QuickSight from all of these data sources listed on my screen here and, you'll be able to create executive summaries and data stories of these dashboards, such as executive summary. I'm gonna click here. So this is how you can do it. And for data stories you'll be able to create or select visual. So I'm gonna write, create Me data story for negative sentiment. Then I'm gonna add visuals. I'm going to select these two click on build. So it's gonna build me a data story based on the visuals that I have selected. It takes a couple of minutes to run, okay? So here you'll be able to change the format styles of the story based on your requirement. And here it gives you a good introduction. A story for each visual, and at the end it also gives you things like these and also a conclusion. So I think it's good for the people who are reporting to the leadership at the end of a quarter or at the end of the month. It's quite handy for them. Now, let me explain why our architecture's, modern design makes it so powerful. It. Our solution is completely event driven, which means when a new call recording occurs or arrives in an S3 bucket, it automatically triggers analysis. When chat conversations end, they immediately get processed. No scheduling, no batching, no delays. Think of it as having an incredibly efficient assembly line that starts working the moment. New information arrives. And being serverless transforms how we deliver the solution. There's no infrastructure for you to manage or maintain. The system automatically scales with your volume. Whether you're processing hundred or thousands of conversations, you only pay for actual processing time. No idle resources. And this architecture delivers three key advantages. Speed analysis starts instantly when conversations end efficiency. Resources are used only when needed Cost effectiveness. You only pay for what you process. And what does it mean to you as a business? There's no capacity planning needed, no infrastructure management, no upfront cost, automatically scaling during busy periods, and cost optimizations built in. Built in these modern approaches ensure you're always ready to handle customer interactions, whether it's a quiet Sunday morning or your busiest hour on a Monday. And security is fundamental to this architecture. Let me show you how we protect your customer data at every level. Data protection. Everything starts with comprehensive data protection. All customer data is stored in S3 and which is automatically encrypted at rest. Every transmission between services is encrypted. We use A-W-S-K-M-S for secure key management give you, giving you complete control over your encryption keys. Access control. We implement certain access controls. Each component operates with specific IAM roles. Lambda functions have precisely defined permissions. No service has more access than it absolutely needs. Think of it as each component having its own security badge. With every specific access, right? Compliance features for banking compliance, transcribe automatically identifies and can redirect sensitive information. Every action is logged in CloudTrail for audit purposes. We maintain continuous compliance monitoring. This ensures you know exactly what's happening with your data at all times and coming to infrastructure security. At the infrastructure level, services run in your VPC network access is tightly controlled. Continuous security monitoring alerts you to any unusual activity, and the beauty of this design is that security isn't an afterthought. It's built into every component and interaction. You get enterprise grade security automatically helping you maintain compliance while protecting your customer's sensitive information. Now let's wrap up what we have covered today. You have seen how this solution automatically analyzes every customer interaction, delivers near real time insights for calls and chats. Securely scales with your needs, turns customer feedback into actionable intelligence. And in the demo, you have seen how this works in practice from conversation to insight in minutes, not taste. And this isn't just about better analytics, it is also about understanding your customers better, solving problems before they escalate, making informed decisions based on. Real data staying ahead in competitive banking environment. The question isn't whether you need these insights, it's how quickly you want to start benefiting from them. Thanks for staying with me till here. I would love to connect with you, so please feel free to connect with me on LinkedIn. Thanks so much. Bye.