Building inventive customer experiences with data, AI, and serverless

Hello, everyone. Welcome to our session "Building Inventive Customer Experience with Data, AI, and Serverless."

I am Pav Sharma, Principal Solution Architect based in the UK. And I'm joined by J Rao, who is Principal Solution Architect based here in the US.

Today, we are here to explore how data, AI, and serverless can be used to not only meet your customer expectations but actually exceed them.

So the agenda for today is:

• What is customer experience and why does it matter? Why is it not just a buzzword but actually a business imperative that can drive outcomes for your business?

• We'll then talk about foundations of building inventive customer experience. We'll talk about four key pillars, right? The framework that you can take back to your organization to build those foundational pillars. On top of which multiple customer experiences can be built.

• We then take a look upon different architecture patterns and solutions that you can take back and apply within your organizations to build some awesome customer experiences.

• We'll also present to you a couple of case studies to inspire you from customers who have actually built these inventive customer experiences and share some awesome stories there.

• And finally, we'll also talk about different AWS programs and workshops that we offer through your account teams and otherwise, which can help you get started or if you are already in that journey, then accelerate it for you.

So what is customer experience and why does it matter?

We are surrounded by some awesome customer experiences. So when you watch your favorite show or favorite streaming service of your choice, like Netflix or Amazon Prime, you are used to being recommended shows that are based on your previous watch history or based on what other people who have similar taste like you are watching.

You are when you shop online in Amazon.com or other ecommerce providers, you are used to a frictionless digital experience and you expect the goods to be delivered within a day or less.

When you are looking to get your mortgage and you contact your bank, you are accustomed to an experience, which is at an interplay of both physical and digital interactions.

Basically, these leading organizations around the world have set the customer experience bar so high that when customers engage with any brand, they carry those expectations with them.

Hence providing a superior customer experience is not really a choice for any business, but it is actually an imperative to compete in modern marketplace and leading organizations around the world actually understand this and that is why when they're building new products and features and services, they keep customer and customer experience at center of it.

So organizations are using data to provide and gain a more holistic understanding of their customer and use that to create contextualized and personalized offers and services. They are actively using data to continuously fine tune the offers and services that they have so that they are most relevant for their customers. And they are super keen to meet the customer on channel of their choice and trying to reduce that friction that comes by say, asking them to download an additional mobile app or create a new login id just to access some service with an overall end goal to help the customer achieve their personalized goals.

Now, to understand this a bit better, let's take an example right here, we are presenting to you one of the most common customer experience that is customer experience of buying a car. I'm sure most of us in this room are accustomed to this.

Now, here we present to you Jack Zach who is looking to buy a car. Now, during the acquisition stage of his journey, Jack Zach is continuously researching different car options online and based on his requirements trying to narrow down to a select few.

At this stage, there are quite a few common customer experience features available for Zach today. These are like advanced filters. So for example, using this, Zach can really narrow down based on his requirements like what is the boot space that he expects in the car? What are the passenger number that he expects to carry? What are the different interior choices that Zach likes?

Now based on this, he can filter down the cars that match the requirements and then use features like high resolution images and videos to basically take an up close and personal look and see whether those cars are as per his taste likings of his and his family or not.

Even today, there are detailed car history reports that are available which Zach can access to see whether a particular car or the model is safe for his family.

Now, even though these features are super important and incredibly helpful using AI and data and serverless technology, there's ample opportunity to further elevate the customer experience for Zach.

So for example, imagine an interface is offered to Zach where he can just describe his requirements in a natural language as if he's talking to a human advisor. And based on that the natural language model can actually process those requirements and then filter down and recommend him a select few cars that actually match his requirements.

The car dealerships can also even take a step further and provide mixed reality headsets like the upcoming Apple's Vision Pro using which customers can interact with their car in three dimensional way like never before.

So from this, I hope you are able to appreciate that even though there are some very good customer experience features available at this stage, there is still ample room to innovate on behalf of customers.

Now, once Zach has made a choice of the car that he wants to buy, he moves on to the next stage and that is a stage of on boarding. This is a stage where we want to help Zach actually buy the car.

Now, some of the common features available at this stage are like automated document uploads. Now, at this stage, Zach can very easily upload documents like for example, his driving license, his bank statements, etc. using which a decision can be taken, whether what sort of finance can be offered to Zach.

There is also integration with credit reference agencies using which soft credit checks can be done and these are actually done today as well.

Right now, there is still opportunities to elevate this customer experience. So for example, using AI now automated identity and document review can be done which can then enable instant finance approval, vehicle readiness updates can be provided to Zach.

So from the point he places an order when the manufacturer completes its manufacturing and actually ships his car to the dealer and it's actually delivered to Zach, all the journey of the car can actually updates about that can be provided to Zach, which he is actually used to from his e-commerce provider.

For example, so the moment you place an order, you are updated every step along the journey till that box arrives at your doorstep. So carrying basically that experience to this car buying experience is now possible with these technologies.

Now that Zach has made a purchase, we move on to the next stage and that is the stage of service. So at this stage, the focus is how to make Zach's car buying experience more enjoyable.

Some of the common experience available at this stage is Zach can book a service appointment over any channel of his choice. He can use mobile app, he can use social media channels, all of that and can also expect to get personalized notification for upcoming service based on the service schedule of his own car.

These are the things which are available today, but there is still awesome opportunity using AI to elevate the customer experience.

Here we are talking about the car manual. I'm sure all of you would have seen that car manual like 400-500 page manual which nobody actually ever opens until there is some issue or some obscure feature or your car is behaving in some way. And you want to understand why it is the case only then you open up that manual right?

Now imagine that manual can be transformed into an intelligent AI assistant to which Zach can just ask questions about any feature he wants. And not only that agent gives him the response about that feature but can also safely enable and disable that feature within his car when Zach is in the car or also when maybe through a mobile app as well.

So as you can see, there is just so much ample opportunity and actually your imagination is just the limit with what current technology can do at each stage.

Now, we are in the final stage. So Zach has been having this car for a few years now. Now he's looking for a next upgrade or something and some of the common experience available is the recommendation of next car based on the current car that Jack has now.

At this stage, there is also ample opportunity to upgrade the customer experience. Now, for example, imagine when Zach's car goes to for a service, imagine if he was to be offered an automatic valuation of his car, a free valuation basically and not just that valuation but also clear pathways to his next highly recommended car so that it is super easy and quick and seamless for Zach to move on from one car to another car, completely integrated experience where finance provider and everything else is operating at the background.

So now we have presented to you all four stages right of the car buying experience and also explain how at every step there is an opportunity to elevate customer experience using AI data and serverless technologies.

Now we believe that this framework is not limited to just car buying experience framework. You can take this framework back to any use case within your own business and try and apply it to think what are the customer experience elevation opportunity at each of these stages within your own business, within your own use cases.

So what is actually a customer experience? In summary, it is a right experience delivered through a right channel. Be it a website, be it a mobile app, be it over email at the right time when the customer is most receptive to listen about it underpinned by core features like personalization, customer 360 etc.

Now we have extensively talked about customer experience, lots of opportunity to elevate that. But why does it really matter? Is it just we want to give that wow to the customer? Does it really have an impact on the business?

So here we are presenting data from Watermark Consulting, which they have gathered over the last 13 years, which clearly showcases that customer experience leaders consistently outperform S&P 500 benchmark and customer experience laggards consistently struggle to keep pace.

Now, most leading companies understand this. Hence when Bain & Company ran a survey for leading companies, as in how important is customer experience for them, 80% of them turned around and said it is a super important priority for them and they are actually working and delivering a superior customer experience.

But interestingly, guess what, when Bain & Company ran same review or survey with the actual end customer of those organizations that how many of them actually believe that they are being delivered, superior customer experience, only 8% of them felt that they are actually being delivered a superior customer experience, right?

So there is clearly a gap between what organizations feel that they are offering in terms of customer experience and then what the actual customer thinks that they are getting in terms of customer experience.

So how do we bridge this gap? Right, we will explain in subsequent slides, some of the frameworks and processes and mechanisms that you can apply to try and bridge that gap for your own business.

As we go along the talk here, we are presenting to you a quick case study from Amazon.com to showcase how consistent innovation on customer experience can lead to sustained business growth.

For example, in 1995 Amazon started collaborative filtering through which customers could discover new products based on the product that they were buying, which enhanced the number of customers who started coming on to the platform, but also increased the basket size or the cart size of each time when order was placed.

In last 5 to 10 years, Amazon has consistently delivered customer experience innovation in form of voice enabled self-service, same day delivery, reorder express lane, etc. These innovations have driven more customer traffic towards the Amazon.com website which has meant that more and more third party sellers have been super interested to sell on Amazon.com, which then has led to enhanced choices for the end customer in terms of products that they can buy from Amazon. But then also at the lowest prices that they can get from it.

As you can see this customer experience innovation, how it has triggered a revolution and a win, win, win for the end customer in terms of products that they can get at the prices that they can get, in terms of third party sellers that at single point they can sell to so many different customers. And also for Amazon in terms of how it led to greater revenue for Amazon.

Now, let's take a look at foundations of building inventive customer experience. Here we present to you four key pillars which we believe are called to build inventive customer experience.

The first of which is design thinking. Design thinking is a process where you try and innately understand your customer review, the different technology choices that are available to solve those problems for the customer and then basically filter down based on the requirements of your business so that the solution that you deliver, make more sense for your business.

The next pillar which most customer experience professional will never think about or don't usually think about is well architected now. Well architected framework as you will know, provides six nonfunctional dimensions through which you can review your applications to understand what are the opportunities to improve your applications so that they are primed to meet any demand from your customer, any sort of customer experience expectation at any scale. And at any time of the day,

The third pillar is customer insights. You would have often heard that data is at center of all the transformation that is happening today, which is true. And customer insights is about exactly that the ability to continuously capture data after getting a consent from your through various channels like mobile app, like social media, like store visits, et cetera, storing them centrally and then continuously analyzing it in real time to make decisions and fine tuning production services. Based on that, basically creating a unified customer profile that acts as foundation to any customer engagement that you do is at heart of any customer experience use case that you can build.

And finally artificial intelligence. I think with Zach's example, we gave you a number of scenarios where we could demonstrate that how applying AI it's not just a buzzword, but it will continuously you could see in all those examples, how it can elevate the experience for Zach.

So these we believe are the four foundational pillars which you should apply when you go back within your organization and think through across all these four dimensions, how you are doing in any customer experience use case.

Now let's take a look at each of these one at a time. So what is design thinking, design thinking is the process through which you are trying to innately understand your user through these three dimensions, right? What does the user really want and who the users really are? So that is the user pillar of it, then you have the business pillar. Now, user may have a number of requirements, but does it make sense to solve all those requirements? Maybe not? So you want to pick and choose which are the problems that you want to solve on behalf of your user, which make most business sense for your business. And finally, technology, there will be a number of technology choices which are the ones that you want to use to solve those business problems.

So design thinking process is a five step process. It starts with empathize when you try and understand your user requirements, you then go on to define them basically so that it is well framed and there are a number of mechanisms available at this stage so you can use them to deliver outputs. The third stage is idea now that you have defined the requirements. This is a stage you pick and choose different technology solutions that you can apply to deliver outcome. Once you have agreed at a solution, you build up a prototype and prototype can take up a form like interactive prototypes, mockups, et cetera, et cetera that you can build to basically finally put to a test stage through which you can apply usability testing a b testing q a and many other tools and generate incredible feedback which you then feed right back into the empathized phase.

So as you can see, this is an incredibly interactive process. It needs to be super agile and continuously evolving so that you are zeroing down to the needs of your customer and business through this well defined process. This is the process which is in the heart of most of the leading product organization across the world.

So now let's take a look at the well well architected framework. The first pillar of this is operational excellence. The operational excellence is really about streamlining processes so that you can deliver outcomes for your user with minimum cognitive load.

The second pillar is security. Now your users have to trust you that security is is job zero for you and you will do everything within your control to safeguard that data. This allows the users to be more open as they engage with your application to share data, which you can then use to create more recommended personalized offers for the user.

The third pillar is reliability. Now your user needs to trust that your applications will work when they need them the most. Now say for example, when you're swiping your credit card at 2 a.m. in in a las vegas casino, you expect it to work and that reliability has been built over the last many years.

The fourth pillar is performance efficiency. Now we all know that faster load times leads to higher revenue for most ecommerce application and it's super critical to have a performance application to ensure that you have a higher engagement with your customers.

Next is cost optimization. Now optimized applications save lot of cost for your applications which you can then in turn pass it back to your end user delivering a superior customer experience. We at aws are super committed to help our customer or optimize their workloads on aws so can so they can make those cost savings which in turn, they can pass it back to their own customers.

And finally, sustainability. Today, every customer is super conscious about the choice that they are making and the impact those choices are are having on the planet and to build that trust within your customer that every decision that you are taking has this sustainability lens built into it ensures that they will be with you for a longer term.

Now, one key point that we would want to add before we move on is most organizations try to, you know, in the interest of getting quickly out to the market do not focus on nonfunctional requirements. And for us as the tech technology champion, it is our duty to ensure that everyone does apply the same focus on these nonfunctional requirements to ensure all customer experiences are delivered.

The next pillar is the customer insights which is delivered through customer data platform. Now, customer data platform is a software solution that collects organizes, analyzes the information that is gathered through various channels and makes it available so that different use cases can use it. This is really about creating that unified customer profile on top of which all the AI can work and provide customized solutions to your end customers.

And finally, the fourth pillar is AI. We have extensively given a number of examples of AI and how it can really elevate the experience. And here we present to you different things like hyper slide features, frictionless execution across touch point, engaging in multimodal interfaces through generative AI and always on experience through virtual agents.

So now at this stage, I would like you to hand it over to Jay to discuss about how to build this inventive experience on AWS. Thanks Bruno.

All right. So we saw what customer experiences and uh why it's important right to build these experiences in your uh in your applications in this section. We'll see how we can build this inventive customer experience on AWS.

So we talked about the different journey stages of a customer life cycle. In this section. We'll talk about the solutions that you can build for each of these journey stages and how we can deliver this inventive customer experience.

So let's talk about the acquire stage first. Let's continue with the car buying process. The example that we talked about in this stage, you as a car manufacturer or dealer are looking to acquire a new customer. Essentially, you are trying to capture data about the customer to know their buying signals, right? So you are there when they are making that decision. In this stage, you are targeting those prospect customer right to identify the requirement, their budget, their interest. And you can use the external data sources, things like social data, social media or market data. And you can also use your internal data to build this unified customer profile, audience profile and understand the requirement you can pass on this, you can build a propensity list using this this you know information or the data that you collected. And you can pass on this information to your marketing or sales team so that they can be targeted in their offers to this customer when they are when they are going in front of them.

So let's talk about how we can build this unified customer views of data and interaction, also known as customer 360 or customer data platform. The architecture works essentially on the three concepts, unification, analytics and activation unification as the name suggests, right? You're bringing the disparate data sources into a single environment, analytics, right? No secret there. You're using analytics and machine learning to generate the insights about the data from the data. And the last is activation. You're putting this insights to work, right? You're putting it to personalization, ad campaigns, marketing offers and so forth.

So the data typically is collected from multiple different data sources. It could be s a application that you're using, it could be logs, it could be a click stream data, right? Number of data that you can uh number of data sources that you can use. And based on the data source uh that you're collecting this data from, you can use different AWS services to bring that data into a singular environment. Also typically a data lake, right? You can also use AWS data exchange, which is a service that provides you to gather and capture the third party data, the market data that I was giving an example of.

You can bring all this data back into a single environment. Typically, you can use lake formation right to build this scalable data lake on Amazon S3 lake formation allows you to build this unified governance so that you can centralize security, access control, audit trail. You can use AWS glue, which is our ETL service, right? You are extracting transforming and and uh loading the data and you can use glues to data proof if you want to visually prepare the data. And lastly, typically, customers also use AWS Lambda for validation uh and and verification of the data.

So the data is ingested, cleansed and normalized and ready for the analytic service to use at this point. And there are a number of analytic services on AWS platform that you can use at this point namely Redshift, which is our flex cloud data warehouse service, right. So if you want to run complex and article query to generate insights from the data that you captured and gathered. You can use that you can use AWS QuickSight, which is our business intelligence service for you to allow you know, to build dashboards and visuals.

One of the key aspects in any customer data platform or customer 360 is identity resolution, right? What do you mean by identity resolution? You're bringing this data from various data sources. Identity resolution allows you to unify and pinpoint this data belongs to this particular user, right?

When you, when you are especially capturing from different channels, different data sources and that's a key point in this customer data platform, you can use deterministic rules or probabilistic rules to resolve this identity of the users which is in the data that you are capturing across different data sources.

You can use EMR, which is our big data service, right? To resolve using deterministic rules or you can use Amazon SageMaker, which is our machine learning service so that you can use ML models and predictive analytics to match the records.

You can expose this unified customer profile using an API to drive your activation, right? So you can drive your personalization activity, you can drive your marketing offers and you can sort of expose this to your internal teams to make use of it.

Feel free to scan this QR code, which has the guidance around how you can build this customer 360 platform in AWS.

Let's talk about some of the things that you should consider to enhance the architecture that we just saw. If you're building this customer 360 or customer data platform on AWS, first think about the data strategy and this is not so much different than any data lake. But since this architecture, you are collecting a lot of data from various different data sources, various different formats, you want to make sure that you have a correct lifecycle and correct storage tiers, right? That you're using to store those data.

And another key point is you're using the efficient data storage formats and that things like Apache Parquet or ORC and that drives two things - one, performance, right? When you have this compressed data ready for querying, performance is the key. And also the cost, as I mentioned, you are capturing a lot of data and you want to make sure that you, you're storing it in the cost optimized way.

As I mentioned, if you're using Glue for your ETL activity, Glue Studio allows you to visually author the scripts to transform the data. So you don't have to generate the scripts from scratch. You can take advantage of these capabilities which will make life much easier when you're transforming the data.

And lastly, AWS Entity Resolution - so the block that you see about the identity resolution can be replaced by this managed service called AWS Entity Resolution, which allows you to match, link and enhance records using either rule-based matching or machine learning models.

So we saw the acquire stage, let's talk about the next stage of the journey - onboard, right? So the purchaser has to typically when they make a decision to buy the car, you'll have to provide some sort of documentation, you know, and might also have to apply for a car loan going back to the same example of Zach.

So this stage of the journey you can call onboard and you can apply this to different verticals. So how can we improve this experience? Customers or cars nowadays are being sold online. I'm not sure if you've captured the recent announcement of Amazon.com and Hyundai partnership that you can purchase Hyundai cars on Amazon.com website starting next year.

So it would be fitting if you are able to do a lot of this paperwork digitally. So that the car either can be delivered to your doorstep just like the packages are getting delivered nowadays, or you just have to drop into the car dealership just to pick up the key and walk out, right? Or if you're purchasing the car in the dealership, it would be really great if you're able to do a lot of the paperwork after an exhausting negotiation session that you might have while purchasing the car.

So a lot of these capabilities can enhance this experience. And what you can do to enhance is there are a couple of ways that you can do, right? One, you can verify the identity as Bruno mentioned earlier and validate the documentation. So a lot of this paperwork activity that happens today that can be done digitally virtually and seamlessly.

You can also use AWS Intelligent Document Processing. So Bruno mentioned about the document uploads that are available today. But what if the loan provider or any other car dealership can use this intelligent document processing capabilities to remove the inefficiency into the application processing that they may be doing today so that they can get the decision like a car loan decision to the user or Zach faster?

Let's talk about a reference architecture, how you can build an onboarding inventive customer experience for the onboarding stage.

Let's talk about the identity verification portion first. So let's say a prospective customer uploads a set of documents using a front end or UI which lands on Amazon S3 and Amazon Lambda, which gets triggered, which will invoke Rekognition API to do image processing, right? So it can compare a selfie with the identity documents such as a license or passport that they uploaded so that identity can be verified - it's the same person.

Once you have done that, the next stage of the process is triggered, which is the application logic. At this stage, you are trying to validate the document that is uploaded and this is where you are integrating with a third party API. There maybe third parties which allow you to authenticate driving licenses and passports. Either you are using an API or you can use AWS Data Exchange to subscribe to a third party data as part of your document validation process.

As I mentioned, the next part of this onboarding journey, you can do intelligent document processing. So the same process applies - Amazon Lambda triggers the Textract service to extract the text from the document. And then you can use Amazon Comprehend, which is a natural language processing service, to extract the key insights out of the text.

At this stage, you can also use generative AI, which a lot of you know, right? You can use generative AI to create summaries of the document so that the loan officer who is processing this document or mortgage packet can make the decision faster and has all the information available to them.

So some considerations if you're building this onboarding architecture - first, if your use case benefits from reducing spoofing attempts, right? If the user is putting a picture of someone else in front of the camera rather than taking a real selfie for that identity verification, you can use Rekognition's face liveness API so that you know it was not spoofed.

If you're using Textract for document processing, you can use the dedicated APIs to process the identity documents. Or Textract also provides a natural language queries ability so that you are extracting key pieces of information rather than processing the entire document if you're only looking for those pieces of information itself.

And lastly, if you're using large language models, you can use frameworks such as LAMA Index or LangChain, which makes the integration of these large language models in your application much simpler.

And lastly, if you're using this process for critical decisions, you should leverage human reviewers, especially for the low confidence predictions, right? So you have those verification or feedback loops already built in as part of your application.

Alright, so moving on to the next stage of the journey, which is service. As the name suggests, right? The customer may reach out if they have questions or wanted to share concerns and you can build some end user facing tools or you can provide your customer service team some capabilities to deliver a transformative experience, to provide this enhanced customer experience.

Your contact center should leverage AI capabilities using Connect Cloud Center Intelligence solution or CCI solution. You can boost your agent productivity. You can also gain conversation insights of what your customers are saying, what sort of feedback they are providing so that it can enhance either the process or you can take that feedback back to the product teams.

And how can you forget the generative AI at this stage, right? You can use generative AI to power your service agents or you can create virtual agents, you can create bots to assist your human agents when they are supporting or helping your customer.

So, continuing with our story of Zach, let's say Zach purchased the car and but he asked questions about certain car features or functionality. In a traditional customer experience, he would have to, like Bruno mentioned, refer to the car manual or maybe just go online and search on the internet which may or may not yield the right answer.

But what if we can provide a virtual agent which is available 24/7 to which has all the knowledge about the car model, the feature functionality and can accurately reference the car manual when it's answering the queries of the user, in this case Zach? Wouldn't that be an awesome customer experience?

So let's take a look at how we can build such an agent. This architecture is broken into two parts. So there's the first part which is the admin part, right? So through the admin UI you will configure the virtual agent, you will provide or upload car manuals and other relative documentation which will land on S3 and a Lambda function will get triggered and the documents will get passed through embedding models which will generate the numeric representation of vectors and get stored into some vector store.

The second portion of this architecture is the actual one where the user interfaces with this chatbot or the agent. The queries will be received by the Lambda function through Amazon Lex chatbot and Lambda can trigger Amazon Comprehend if you're trying to gain some insights from the request itself to see where you want to route the request or do other processing. Or if you want to do a translation, if you are providing a global service and want to translate the query and store it in a unified manner, you can invoke the Translate API.

But most importantly, the query will pass to the embedding agent, the embedding models, so that you are generating an embedding and finding the relevant passage in the knowledge base that you have stored.

So along with the relevant passages that you searched and the original query, all of those are passed to the text generation LM which will generate an accurate answer referring back to where in the car manual, in this case, where it found that answer, right? So it's not making it up. That's an inventive customer experience.

This solution was published as a Q&A bot on AWS. So if you are interested in learning more about it, please scan the QR code.

Let's talk about some considerations for enhancing that architecture. As we mentioned in the design principle pillar, think of the UI and the channel and the design before you implement the technical solution. Lay out the scenarios of how your users will interact with this agent and work backwards so that it can power your application and design manners.

You can also use large language models to personalize this agent so that it speaks in the same tone, it has the same conversation style as your organization, so it provides that cohesive customer experience.

You can also create a data driven framework to select the large language model. I'm sure as you know, there are a number of models out there and not all models might be equivalently good or perfect for your situation. You want to use data to back up the choice of the model.

Look at how your users have been interacting within the chatbot so that you can periodically update the bot using the utterances that you've been receiving, right? So you can evolve your agent with the trends that you're seeing from the customers through their interactions.

And lastly, define clear measures of success. You need to capture how the solution is solving the problem that you started with, how it's affecting the business outcome. Example KPIs could be increased customer retention, an improvement in the Net Promoter Score, but define that success criteria before you implement the solution. That would help establish the impact on the business of how the solution is helping.

Alright, so the last stage of the journey is the continuous growth of the customer. I mean, there are lots of statistics out there about how expensive it is to acquire new customers versus retaining existing ones. And you don't only want to retain the customer, you want to mature them on your platform and service. For most businesses, retaining and increasing share of wallet of customers is a key part of the strategy.

To do that, you would need to capture data to gain a holistic understanding of the user. In our story, let's say Zach bought the car a couple years ago and using Zach's interactions on your platform and your partners' platforms, you can realize that his situation has changed. Using that, you can recommend a bigger car or car with better mileage or lower CO2 emissions to provide a personalized experience for him.

By knowing your consumer, you can send more relevant messages, recommend products and services that are more likely to interest them. So let's take a look at the architecture diagram of how you can create a personalized experience.

In this system, what we are doing is we are dynamically capturing the user data, user activity and we are feeding that insight to large language model to generate a personalized content.

Imagine a scenario, let's say if you are a car dealer who is creating a campaign to market the popular vehicles to users, you have a top 10 popular cars and you want to pick the top three cars for each users based on their interest, based on their previous ownership, and other traits that they are looking for in the car.

For example, in our story, Zach now has a bigger budget and a need for a bigger car. We could use an AI system to determine which of the cars or sets of cars will meet this requirement for him and then generate that email or content and send it to him.

This reference architecture is built on the Customer 360 as we saw in the Acquire stage. Customer 360 provides activity logs based on the interaction or activity that the user has been doing on your platform or your partner's platform.

The user profiling engine is deployed on SageMaker. It takes the user activity data and figures out the interest, customer interest which is stored in the Feature Store. Feature Store is a tool which stores, manages, and enhances the features for a machine learning model. Features could be the categories of cars the user is interested in, the price range, and similar things. Those are the features for the machine learning model.

The top 10 vehicles, the promotional list of vehicles that you are trying to promote as a dealer, is stored in the Metadata Store.

An AI agent is key, at the heart of this architecture. It coordinates the effort across a number of different components. The AI agent takes the user ID as an input, fetches the customer profile data, customer interest from the Feature Store, it goes and fetches the promotional item from the Metadata Store and then feeds that into the Large Language Model which picks the top three recommendations and creates a personalized email, a personalized body that you can use to send to the user through Pinpoint.

You can also provide a direct interface to the users where they can come and type what they're looking for. For example, Zach could come to this chat board or interface and type: "Hey, I'm looking for a car with 6-4 people capacity, with 350 miles of range, which has low CO2 emission, which has rear seat air bags, rear seat car anchors, all within $45,000."

You can see it's a complex requirement and it cannot easily be solved using rule-based chat bots or engines. But with generative AI, you can break this requirement down easily and then recommend the top three cars or top cars which meet this requirement.

The consideration again is that since the AI agent is the heart of the architecture, using Lama Index, LangChain or AutoChain, all those frameworks are critical to implement this architecture. Leverage the prompt strategy - prompting strategy for any generative AI application is key. There is research on engagement guided prompting. Essentially you are providing the customer preference or customer interest as part of the prompt, so that the AI model, the large language model, knows that it's an AI agent assisting in email campaign crafting and recommending top cars or personalized recommendations for the cars to the user.

Lastly, there are a lot of choices out there in terms of the LMs and may not all models may be equally applicable or useful to your situation. So use a data driven model to select the model that is applicable or performs best from cost and performance point of view.

We talked about different solutions you can build for each of the customer stages. Let's talk about a couple of customer case studies.

The first one is Lunar Energy, a UK based company providing home energy management solutions - essentially a way for customers to manage their energy assets like electric vehicles, batteries, solar panels, etc. It also allows them to sell the energy back to the grid provider.

The platform they have built is called GridShare which uses the customers' energy needs and predicts the consumption and production estimates for the energy needed for each customer every day, for each household, for each device. They use this information to create a personalized charging plan for every customer. Because Grid can co-optimize between the customer's bill and the revenue it generates from selling that energy back to the grid provider, it makes sure the end customer gets the best value possible.

In 2022, a customer in Japan saved an additional 14% on their energy bill. This is an idea of a customer experience - a platform which is scalable to millions of devices, provides personalized recommendations and control plans which result in cost savings for the customer.

Another customer story is a US based healthcare firm. They provide claims and benefits administration software. Using Amazon's Working Backward principle, they landed on improving the customer experience for their claims adjudication engine.

They cut the processing time for claims applications using an event driven service system powered by AWS Lambda along with AI services. As you can see, they cut the processing time significantly which is a major competitive advantage in their industry.

They are using a similar architecture as we saw for intelligent document processing - Amazon Textract to extract text, Comprehend or large language models to classify, and then do some post processing to extract insights or create summaries to power chatbots.

They are using human-in-the-loop for low confidence predictions in their testing. This successfully processed 60,000 claims, which used to take hours, in a matter of minutes.

The key impact is customers are getting quick processing of their claims - they are getting reimbursed faster compared to before.

Now Bruno will talk about how AWS can help and workshops you can take to enhance your journey to build these solutions.