iTranslated by AI
re:Invent 2025: 5 Steps to Build, Deploy, and Monetize Agentic SaaS Apps with Launchpad
Introduction
By transcribing various overseas lectures into Japanese articles, we aim to make hidden, valuable information more accessible. The presentation featured in this project, which proceeds with that concept, is this one!
For re:Invent 2025 transcript articles, information is compiled in this Spreadsheet. Please check it out.
📖 re:Invent 2025: AWS re:Invent 2025 - Build an Agentic SaaS App in 5 Steps: From Idea to Revenue (AIM104)
In this video, John Huehn and Steve Bixby from Pega Systems demonstrate building an enterprise-grade agentic SaaS application using Launchpad. In a scenario where the fictional fintech company HomeLend develops a mortgage pre-approval app for banks, they demonstrated five steps from blueprint design leveraging Gen AI, implementation on AWS, AI agent management through workflows, delivery to subscribers, and monetization. They used AWS Bedrock models to automatically generate data models and UIs, and also implemented customizations such as adding human approval for loans exceeding $5 million. Finally, they demonstrated the pre-approval process through a conversational interface on the U+ Bank website, showing how processing time was reduced from days to seconds.
※ This article is automatically generated while maintaining the content of the existing lecture as much as possible. Please note that there may be typos or incorrect information.
Main Content
Beginning the Demonstration of Building an Agentic SaaS Application with Pega Launchpad
Well, hello everyone. My name is John Huehn, and I'm the Chief of Launchpad at Pega Systems. And joining me today is Steve Bixby, who is our Senior Vice President of Product Engineering. And for the next 20 minutes, I'm excited to show you how you can build an enterprise-grade agentic SaaS application that can run complex and regulated business processes and be sold to thousands of B2B customers, all in just five easy steps.
Those steps include designing it with Gen AI-powered blueprints, bringing it to life here on AWS, operating it with enterprise capabilities, delivering it to subscribers, and monetizing it with actual customer usage. Today, we're going to demonstrate how a fintech company we call HomeLend builds and sells agentic applications to banks. These applications dramatically accelerate processing and provide great customer experiences. The scenario is something that many of us have either experienced ourselves or might experience at some point in our lives: mortgage pre-approvals.
Now, of course, banks are very excited about the idea of automating operations, but at the same time, they cannot tolerate AI bringing variability to critical decision-making, like creditworthiness. And that's where many agentic applications struggle. They have legions of agents trying to manage legions of other agents to deliver outcomes, and of course, that naturally creates all sorts of variability. That's something that cannot exist in financial decision-making.
So today, we're going to use workflows to manage our AI agents and deliver controlled, deterministic, and predictable outcomes, which is the kind of outcome that regulated industries like banks actually love. To build our application, we're going to use Pega's AI-powered low-code app development platform called Launchpad. If you've never used Launchpad, what I want you to know is that whether you're building a new application, or you're augmenting an existing legacy application by injecting workflow and Gen AI into it, or you're replacing and replatforming an older technical debt-ridden application to something new and modern that can innovate faster, Launchpad dramatically accelerates time to market, reduces development time, and reduces cost.
Launchpad provides a vibe development experience to help you get your app off the ground with a managed database and a secure and scalable cloud powered by AWS, industry-leading workflow, integration, UX, and reporting capabilities, and pre-built subscriber management, admin capabilities, and configurability. All delivered with a usage-based pricing model that scales with your business. Ultimately, Launchpad gives you everything you need to build and sell enterprise-grade applications without the cost and complexity of managing cloud, security, and, of course, code.
So let's get started. Here's our scenario. HomeLend, a fictitious fintech company that builds and sells apps to banks, has identified a gap in the mortgage pre-approval process that affects people like Sarah. Sarah's dream home just came on the market, and she's super excited. She immediately booked a showing and contacted her bank, submitting all the paperwork necessary to get a mortgage pre-approval and corresponding letter.
But unfortunately, her bank took two days to process her application and send her back a letter. Sadly, in the meantime, someone else bought Sarah's dream home. Imagine how disappointing that would be. HomeLend understands that pain. So they're going to build an agentic SaaS application for banks to reduce that processing time from days to seconds and provide a great experience for people like Sarah.
And it's really an agentic SaaS application like this one, with a conversational interface. So Steve, I'm expecting you to play the role of HomeLend here, and I want to build an app that manages the mortgage pre-approval process and other loan origination related operations. Okay, that sounds like a challenge. Let's do it. Hopefully, the Wi-Fi holds up. We're going to build this thing super quick. Last session of the day, thanks for coming everyone.
So John, can you actually say it again? Okay, Steve, I want to build an app that manages the mortgage pre-approval process and other loan origination related operations. Got it. Fantastic. That's enough for us to get this process started. So let's go.
We're using Launchpad. For those of you who want to check out Launchpad and try it yourself, please come visit us at the Pega booth right over there. You can do all of this yourselves. Check out launchpad.io. Okay, perfect, it's asking what's the name of the application? This is HomeLend, so I'm going to type HomeLend. And there it goes. So it says, I can build a loan origination application that includes mortgage pre-approvals. It's identified the industry, the location, all the information we need to build our application. Obviously, we haven't provided enough context, but it's given us an option to start building. For the sake of time, let's just go. Let's build it.
So now on the screen, it's saying
"Building the application." It's starting from industry templates, customizing workflows, connecting channels, setting up personas, and configuring data and integrations. Done. Alright, fantastic. The chat has moved to the right, and the Launchpad development environment is displayed. We can see that the launch readiness is about one-quarter complete. We'll keep checking on this as we go through the process.
It's identified as HomeLend. A loan origination application in U.S. retail banking. Fantastic. And if you click in the middle of this diagram, you can see the created workflows. It looks like eight workflows have been created. Here they are. And mortgage pre-approval is actually one of them. That's good. The rest are other things. Yes, lots of other things. Good. That's what we asked for.
Let's take a look at the mortgage pre-approval process. So, what Launchpad does is it takes that business process and it breaks it down into a series of stages and steps, organizing that business process. Now you can interact with it, you can add steps, add stages, drag and drop them around, update them. But if we look at it, we're collecting application data, we're verifying information, then we're doing a financial assessment, a pre-approval decision, and then finally moving to a finalization. So, in terms of delivering what Sarah is looking for in the scenario that you described, this actually looks pretty good.
So let's make some changes to it. Actually, before I do that, let me show you the data model here. The data model is also generated. The full UI is generated. It's all there. You'll see it in just a second. Again, you can update and make changes to this as necessary. Applicant name, date of birth, W-2, annual income, everything you need for a mortgage pre-approval process is all there. All generated with generative AI using AWS Bedrock models.
Customizing Workflows and Deploying to Subscribers
So, let's go back to the lifecycle. Let's make some changes, John. What should we change? Well, Steve, if someone is asking for a pre-approval over $5 million, it might be good to have a human eye on it. $5 million, yes, someone should actually take a look at it. You can type this into the chat on the right in this pre-approval decision, but I want to show you a cool feature here. You can directly click on the area of the screen you want to change and do it inline. You can type, "Configure the system so that loan requests over 5 million require human approval." Period. Done. Nice.
That's fantastic. Now a human approval step, or rather, a decision step, has been inserted. This determines whether a human approval step is needed. If you want to know what this actually did behind the scenes, you can open the left panel here. This shows all the assets created to support this application. And down here, there's human approval. This is a decision rule. It generated the expression that if it's private and the loan amount is greater than $5 million. That's great. So let's close this and show you that you can work full screen.
Let's make one more change. Let's add an agent step, or at least a placeholder where we can insert it. Uh, let's integrate an AWS Nova agent to automate employment verification through an approved third-party service. You can add decision steps, or human steps. These would require screens or some kind of interaction, whether it's through self-service or through a contact center agent conversation.
You can add agent invocations, add decisions, add automations, you can do all of these things right here to build this application. Now, because time is very, very limited, let's just say that's enough for now. Let's preview it. Let's see what this application looks like to potential end users.
There you go. It says right here, preview, Demo Bank. It's not real. It's not real. This is a demo. This is how you test the application, and this virtual assistant appears. It says, "Hello, I'm your virtual assistant. How can I help you today?" We tell it, "Hello, I'd like to get a mortgage pre-approval." It says, "Of course, I'd be happy to help. To get started, can you provide your full name and date of birth?" "Of course, my real name, Steven Bixby, June 14, 1983, which is not my real birthday." "I didn't realize you were so young, Steve." "Thank you. What loan amount would you like to be pre-approved for?" "I'd like to request $1 million." "Okay, let's continue."
Do you see what this is doing? It's actually trying to walk through the first step of the process and collect the information it needs. To further visualize that, in this preview portal, you can see how it looks through a self-service chat, but what does it look like in the bank's back office desktop application? This is the application that was built for us, with a dashboard, a chat, and all sorts of things happening here. And you can actually create any of these workflows directly from here, or instances of workflows, including that mortgage pre-approval.
And now you see the fields prompted in the first stage, first step. This includes applicant name, date of birth, email address, phone number, annual income, and W-2 files. This is pretty cool, right? Because we've just built the core logic of the app. And we're showing it in a conversational interface, but you can actually do the same thing on an employee's desktop, and it also works on self-service websites, mobile apps, and mobile apps. So, once you build that logic once in the core, you can deploy it to all channels. That's right.
Now, let's assume this app is perfect. In reality, you'd want to customize a lot of things to make it your own and serve your specific purpose. But from here, let's launch it. What's unique about Launchpad is that it's specifically designed for organizations building SaaS products to sell. So, we're the provider here, and we're going to sell this to subscribers.
When we launch it, we actually take the blueprint of the application and materialize it. This means we perform domain data modeling, workflow, UI screen generation, integration settings, security configuration—all of these things to ensure that we have a deployable app that subscribers can actually use. Now, we see a list of HomeLend's subscribers. There are five banks here. Let's deploy it so that all banks can access it. And we deploy this application.
Production with U+ Bank and Sarah's Successful Mortgage Pre-Approval
Okay, now that we've made it available to all subscribers, we're looking at the subscriber dashboard. As a provider, you can see what's happening with other apps you've built and deployed. You can check usage, revenue, and so on. And here, those banks are displayed, and you can see that the HomeLend subscription is active for all banks.
Since we built this application, let's do a little fantasy here. Let's switch to the website of one of these banks and assume they've already deployed this virtual assistant to their website. So, this is no longer a fictional bank, but actually UBank or U+ Bank's website. This is their website. They've deployed this assistant here, and when you click it, it says, "I'm your U+ Bank virtual assistant. How can I help you today?" It's very similar to what we saw in the development environment preview, but now it's live for actual subscribers.
And this time, let's say I'm Sarah. I'll be Sarah in this use case. "Hello, I'd like to get a mortgage pre-approval." "Okay. Of course, I'd be happy to help. Let's get started. Please enter your full name." Now we'll go through the same process again. This time, I'll say I'm Sarah.
Sarah Wilson, March 18, 1974. Thank you. What loan amount would you like to be approved for? Let's go with $1 million again. Come on, everyone, $1 million for Steve, $1 million for Sarah. $1 million for everyone. And now, as we saw before, it's asking for a W-2. This time, let's actually do it. Let's upload a W-2. We submit the W-2. We've securely received your W-2. We're currently evaluating your financial situation to determine your pre-approved amount. We'll let you know the decision shortly. Here it comes. Fingers crossed. Let's hope Sarah gets approved. It's not over $5 million, so the rule we added doesn't apply. Okay, congratulations. Fantastic. You've been pre-approved. Shall we send it to your email? Yes, please. Check your inbox. Can I help you with anything else? Let's say that's it. Thank you so much for the quick response. Wonderful.
Okay, John, I think we did it. We built the application. I think Sarah will be happy. So let's bring it back. That's great. Thank you so much for sharing all of that. So, let's quickly recap what Steve showed us, going back to the five steps. First, Steve, what you did was leverage vibe development to create a design, leveraging Gen AI to create a blueprint. And you brought it to life as a secure, scalable reality on a pre-built AWS architecture.
You made it run with enterprise-grade functionality and workflows that manage AI agents. And you enabled subscribers like UBank to configure and deploy it to their customers, generating revenue from actual live customer usage from people like Sarah who are using UBank's app deployment. So, in five simple steps, we built an enterprise-grade Agentic SaaS application, drastically reducing processing time and providing a great experience for Sarah. And we automated cumbersome bank workflows so they can provide the experience. Sarah's story can finally have a happy ending.
So, Steve and I don't really have anything more to tell you, but if you have any questions about what we shared today, or if you'd like to learn a bit about the amazing companies that are already building Agentic SaaS applications on Launchpad, like the one Steve just showed you, or if you'd like to learn about our security features, or a complete technical review that accelerates your journey to building applications and listing them on the AWS Marketplace, please come to the Pega booth. Our booth number is 366, right over here. Or, if you want to start Launchpad today, you can start for free on your own. Click the start button and let's begin with simple step one. You can start by designing your app with a Gen AI-powered blueprint.
So, thank you very much for spending your time with us today. And we hope you enjoy the rest of your time here at AWS re:Invent.
※ This article was automatically generated using Amazon Bedrock, maintaining the original video's information as much as possible.
Discussion