 This is Sadhana Nandakumar and I'm a Senior Solutions Architect with Red Hat. Today, we're going to be looking at a use case demonstration of how organizations can improve their customer satisfaction by rethinking their customer engagement strategies. With the digital age, intelligence is everywhere and data drives decisions. By providing personalized real-time service to the customer based on their needs, we will dramatically be able to improve their loyalty. And for the purpose of today's demonstration, we have considered a real-time offer management system. The intent is to provide real-time offers to the customers based on their behavioral and historical profiles. All of the customer events that come in are being acted upon by our decision component, which then determines the right offer to be sent over to the customer in real-time. Just to give a little bit of overview on the use case that we are looking at today, we have the decision model with a historical model represented on the left, which is a combination of two factors. We consider the qualified purchases of the customer in order to understand his spending summary. We also consider the last response to a marketing offer that's being done by the customer. On the right, what we have is the customer segmentation model, which is a machine learning model, which gives us an indication of the likelihood of a customer accepting an offer of a particular type. This considers several factors about the customer to come up with this prediction. The sentiment analysis that we are doing here is going to enable us to fine-tune or focus the right offer for the right customer. And with that overview, let's go ahead and look at the demo in action. What you see in here is the online banking webpage of a customer named Sara. Sara is a platinum card customer. She has a card ending in 4567. And from her spending summary, we can see that she has predominantly made airline purchases in the past. With that overview, let us go ahead and make an airline purchase on behalf of Sara. This is an online booking webpage where Sara is going to make a booking. As you've noticed, Sara is using the card on her file, which is 4567, to complete this purchase. And as this transaction has been completed, an airline transaction event is being put in into the event stream. This event would then be acted upon by our business decision component to come out with a right offer for Sara. So let's go back on to Sara's online banking webpage and let's refresh. When you go back and check the offers, you can see that Sara has been upgraded to an airline premium card. What is important to notice is that the event has been acted upon by a decision component and it was done in real time. So as soon as the transaction purchase was completed, the offer was sent back to Sara. Now we quite understand how the transaction went in and how the offer came out, but what happened in between? What did the decisioning component really do? Let's look at that in a little more detail. What you see in here is the decision model as represented using the decision model notation, which is a graphical way of representing the decision flows within your business logic. This is a standard that's governed by the object management group and is completely platform agnostic. It empowers the business users to create these assets without having to depend on IT. So it reduces those lengthy STC life cycles that you might have to go through if all of this is going to be code changes. All of your rounded rectangles that you see on this diagram are inputs that come in into the decision model. I have color coded the inputs so that we understand the source of this information. Age, income, and customer's class are all customer profile related information that comes from the customer's databases or data stores. The current event is the event in context and the last offer response and qualified purchases are historical data about the customer. All of this in addition to your customer's segmentation, which is a machine learning model that I introduced you to earlier on, is going to go on to determine the offer. So as you can see, the customer segment model is a random forest classifier model. We are pulling in this trained model and we are using it as a part of our decision graph so that we can determine the best offers for the customer. Now, how does this typically work in a production system? You're going to have teams that will be working on creating these assets. You're going to have data scientists creating these models. They're going to be creating a tool of their choice. In this example, I'm using a Jupiter notebook, which the machine learning scientist has used to create these assets. As you can see, the segmentation model considers several inputs like the age, income and classification and it comes out with a segment as a bucketing of zero, one or two, which indicates a low, medium or high likelihood of a customer accepting an offer of a particular time. Now, once this model has been created by the machine learning scientist, has been trained against the production data, we are trying to use that trained model as a part of our evaluation of the role. So coming back here, you can see that all of this information is going to go on to determine the offer. All right, I understand that the decision model kind of governed the offer that was extended for the customer. But one thing that's still missing is I don't quite understand what really happened with that particular transaction that Sara performed. I need to understand to the level of depth so that there's explainability of the decision that just happened. For that representation, I have a simple dashboard that we have represented here, which kind of shows all of the offers that have been extended for the various customers. Let us now quickly filter by the customer ID for Sara so that we can understand what really happened with that one transaction. So I'm going to go back on to the online marketing webpage and get the customer ID from here and let's apply the changes. You can see that Sara has extended this offer to upgrade to an airline premium card. She had a customer classification of platinum. The age distribution and income distribution fell in this particular section. From her historical purchases, we knew that she had performed an airline transaction event predominantly in the past. So finally, you can see that her customer predictive profile also came back high. So what would have been a black box with respect to the prediction results created by your machine learning model is now being exposed via the decision layer. So there's better control on what can be extended for the customer. So this overrides that's put on top of the intelligence that is coming from your historical model and your predictive model that kind of determines the offer for the customer. So now we clearly understand what exactly happened for that particular use case. So when you go back and read the decision model again, we had an airline purchase that was done by Sara. She was a platinum card customer. Her customer segmentation fell in high. Her qualified purchases said she has predominantly made airline purchases in the past. And so she was extended this offer to upgrade to an airline card. So we are able to understand completely end to end what happened with this particular use case. As you might imagine, there's going to be several capabilities required for implementing a solution of this nature. We need an event streaming backbone, the ability to pull in data from different sources, as well as the ability for business users to create these assets. We've brought together the functionality of our entire application development portfolio to implement this solution. Feel free to visit our booth or connect with us to understand more about this architecture and what we can do for you. Thank you.