AWS re:Invent 2025 - Pioneering Agentic AI Transformation: CSL VMware & SAP modernization (MAM346)

🦄 Making great presentations more accessible.
This project enhances multilingual accessibility and discoverability while preserving the original content. Detailed transcriptions and keyframes capture the nuances and technical insights that convey the full value of each session.

Note: A comprehensive list of re:Invent 2025 transcribed articles is available in this Spreadsheet!

Overview

📖 AWS re:Invent 2025 - Pioneering Agentic AI Transformation: CSL VMware & SAP modernization (MAM346)

In this video, CSL, a global biopharmaceutical company, shares their cloud migration journey using AWS Transform and Amazon Q Business to accelerate data center modernization. Erik Hong, Global Head of Enterprise Systems and Processes, and Hariharan Govindharajan, Senior AI/ML Architect at AWS, demonstrate how agentic AI reduced wave planning from 610 to 60 hours—a 10x improvement. By ingesting 4.6 million records and 17,000 pages of documentation, they achieved 12x faster application discovery and 5x improved wave planning velocity. The session details their SAP RISE migration, digital core foundation, and lessons learned in establishing partner ecosystems for regulated environments.

; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.

Main Part

The Emotional Foundation of Human Motivation and CSL's Mission

Hello everyone, and welcome. Thank you for attending. If I could get you to put on your headset, this is a silent session, and you won't be able to hear without them, so please put your headsets on. Can you all hear me now? Can I get a thumbs up? Excellent, thank you.

In 1986, Dr. Joseph LeDoux published a book titled The Emotional Brain. In this book, it highlights the importance of emotions, and in particular unconscious emotions, as a crucial and important motivator of human action. The book delves into the protective function of the amygdala and how it processes these emotions before it sends silent signals to the rest of the reactive limbic system. This is the part of the brain that allows us to choose our response from emotional stimuli, and this is what Dr. Viktor Frankl so famously observed that we have the power over.

Now, psychiatrist William Dodson believes that our motivations are based off of our interest-based nervous system and emotional arousal. All this to say that we as human beings are highly motivated based off of our interests and our emotional connectedness to things. Though no matter how emotionally connected we are to something or hyper-interested we are in something, there are still limits to how much data that we can process.

Hi, my name is Victor Feinman, and I am a Senior Solutions Architect for AWS. I work with healthcare and life sciences customers, and I'm the global solutions architect for CSL. It's an honor and privilege to be up here on stage to introduce you to CSL.

CSL is a top global biopharmaceutical company. They work to deliver an enduring impact to their patients and public health. Since 1916, CSL has been pioneers in the end-to-end production of plasma-derived therapeutics, from the procedural and technical innovations in the donor plasma collection process to the novel manufacturing techniques to increase the plasma to therapy yield.

Defying big pharma's business model of catering to the masses, CSL focuses on serious, complex, and rare diseases such as hemophilia, immune deficiencies, influenza, and iron deficiency anemia. In my time working with CSL as their global solutions architect, I've been humbled by the many stories that I've heard from my coworkers at AWS, my colleagues at CSL, and the patients about how they attributed CSL to saving and improving their lives.

And not just their lives, but also the lives of their families. Making a profound impact on patients and their families' lives and their quality of life, and sometimes in some cases the life itself, is the edict behind everything that CSL does, which is why this session on pioneering agentic AI for transformation is so important.

CSL's Global Impact and the AWS Accelerate to Cloud Program

So now that we know a little bit about CSL, let's take a look at some of the numbers. 8.8 million—this is the number of patients that have received plasma-derived therapeutics or recombinant treatments. Over 101 million influenza vaccination doses were administered in fiscal year 2025, which has estimated that it prevented over 15 million illnesses, over 200,000 hospitalizations, and over 30,000 deaths.

And then 6 million—this is the number of the amount of economic impact that CSL has within, on average, within each of its

over 300 global donor plasma donation centers, driving patient payments, donor payments, staff salaries, and local spending. So now that we know a little bit about CSL, I want to validate some of the challenges that you may be facing with modernizing your landscape. First off, the prices for the contracts and licenses that you've been paying for years has only been increasing, which is causing large companies to make major moves to capitalize on a lower cost structure. And we, being the IT industry, seem to be moving at light speed, and it's making it increasingly difficult for our legacy infrastructure to keep up with the demands of the current environment.

A lot of organizations are looking towards SaaS providers to help outsource and to reduce some of that risk. But what that does is it contributes to the complex IT landscape organizations face, their data fragmentation, and data silos, which makes it hard and increases the time it takes to discover and plan migration and modernization efforts. Objectively, you know, these challenges are daunting. It's hard to comprehend how difficult it is to get past these challenges. And our customers are asking us overwhelmingly one simple question: where do we start?

Well, at CSL, their migration and modernization journey started with the Accelerate to Cloud from Data Center Program from AWS. This program is offered to large customers doing mass migrations from data center exits to AWS. And this program, realizing that all customers are unique, takes a very specialized approach in doing a very thorough detailed analysis of your technical, business, and regulatory requirements. And then structuring a unified professional services delivery engagement. And by structuring this engagement with a fixed price delivery, including adoption incentives, as well as potentially purchasing depreciated assets on the books, it seeks to reduce or eliminate some of the risk associated with these major migrations.

Through this program, AWS has helped CSL build a foundation where they can run and migrate their innovative workloads. In a moment, Erik Hong will walk you through a pair of major IT initiatives that CSL is going through to make them agile to deliver strategic business objectives and value to their patients. It is my honor and privilege to introduce you to Erik Hong, the Global Head of Enterprise Systems and Processes. There you go, Erik.

Erik Hong on CSL's Modernization Journey and Digital Core Foundation

Thanks Victor. So thanks everyone. Good morning. So one of the things I want to spend some time talking about is our journey, and one of the things that I think my AWS account team promised me is for every time I say AI or agent I'll get some credits for some storage and compute. So as Victor was saying, our story is not very unique in the sense of where we spend our time, where we spend our time as a technology organization in terms of what is commonly known as technical debt, managing and operating in the legacy. We view technology and the role of technology within CSL as one that needs to drive life-saving therapies for our patients.

So when we think about the time that we spent, it's really around the wrong part of the iceberg because of the fragmentation that we have in our landscape, the number of data centers that we manage today, and just the general state, which I know a lot of customers and a lot of other companies are dealing with.

As Victor mentioned, this was not where we wanted to spend our time as an organization. We knew we needed to be in a different spot and really start to marshal ourselves very differently. So one of the things that we started last year was our modernization program, and this is a multi-year journey where we thought about what we want to solve for. Like many modernization programs, it's anchored on our landscape and how we want to transform our landscape, as well as how we want to improve our processes. But the core and critical aspect is around our people, how we want to change and fundamentally change the skills of our resources and our talent, and start to really drive a different level of innovation and outcomes forward.

These are things that we anchored on from a principle perspective, and we really view this first and foremost as a change management journey. So this isn't about technology anymore. This is about how we need to bring change into our organization and think differently in terms of what we do today.

As we think about this journey, it all starts with a foundation. It all starts with what you typically think about as a landing zone in the context of the cloud, but for us our digital core serves as a very different purpose. It serves as a more foundational capability that we're enabling for our enterprise. This is really around the combination of your landing zone and the AWS cloud platform and a lot of the capabilities that we know AWS provides today, but how do we wrap that and package that in a manner where we bring our data natively into it, as well as we bring the right level of platform services and the whole governance and control plane around it. So ultimately what we're trying to do is enable that platform that will serve as the basis to bring our current and our future workloads forward.

This is really where we spent quite a bit of time making sure we got this platform right and spending time with AWS and our partners to really think about how do we need to make sure that we've learned from the experience of others so that we could take this forward. And the critical thing, as Victor was mentioning in terms of our focus on our patients, CSL has a workforce of around 30,000 employees across the globe. We run about seven plus manufacturing sites. We have hundreds of different plasma donation centers, so our key focus is around our employee talent and what we do to make sure that our teams are appropriately engaging and think about technology in the modern context of today so that we can then enable this flywheel effect to deliver the value and the constant value.

As we think about that platform side and we think about the technology program that we've established, one of the key things is CSL has grown both organically and through acquisition. As Victor mentioned, we have a vaccine business which is our Seqirus business. We have our core plasma business, as well as we have our iron business through CSL Vifor. So these are things that sit with us today and they are fragmented. They have their own way of working, their own set of processes, their own set of systems and everything. And it was one of the things that we think about as we think about the therapies that we're driving, how can we gain leverage as a single enterprise and what do we need to do to bring forward that ability.

So we started to really think about what are we going to do with our ERP. ERP, if you're not familiar with it, is enterprise resource planning. This is a core backbone in many organizations, no differently than life sciences and biotech for ourselves, and it's the end-to-end process that delivers and enables the pharmaceuticals that we deliver. So as we think about that end-to-end process, as I mentioned previously, each of our business units, each different product line, they had different sets of processes, different sets of data, different sets of technology for each of these key core processes. So then it leads us into that iceberg challenge to say how do we drive value and what do we need to do and how do we need to think differently.

As we thought about that technology program, one of the core pinnings, we leverage SAP today and we run three global instances today in our environment. So we started to think about, well, okay, if we're establishing a foundation, what do we need to do as a first basis of that foundation. And for us that was migration into SAP RISE, which is hosted by AWS.

This was a key thing for us. It sits on-premises today, and we had to start to bring this and move this environment into the cloud. For us, as we talked about and as I mentioned, this is an extension of our digital core, the SAP environment, because it's very integral to how we view our overall platform. So for us, we need to be able to create this integrated environment from an end-to-end perspective. This is where we start to bring those types of workloads forward.

This enables us to then start to think differently about how we want to solve those core processes and some of the capabilities that we know are accelerating on the cloud side. SAP has their accelerated journey from an AI perspective. AWS is bringing different tools and capabilities for us. It was about how do we bring a marriage between this ecosystem so then we can leverage and create that flywheel effect of value. But you know, this all makes sense on paper, right? As I always say, the easiest thing to do from an enterprise architecture perspective is to draw boxes, right, the end state points of things.

But as I mentioned in the beginning slide, we had a number of challenges. Our CMDB was not accurate. We had challenges in terms of what's where, what's running in which data center, what's it connected and integrated to. So all these things brought us to this need to really say, do we truly understand our environment, our landscape at the level of detail that will allow us to actually bring this safely into the cloud so we don't disrupt our business in a significant way? These challenges that we were realizing required us to make sure we thought a little bit differently in terms of how we leverage capabilities that exist in the market today. And this is where we start to really leverage our partnership with AWS to enable some of those things forward. So I'm going to hand off to Hari right now. He's going to spend some time and take you through some of those tools and our journey from the technical perspective.

Introducing AWS Transform: The Agentic AI Migration Framework

Thank you, Eric. What you've heard from Eric is an incredible transformation story, but I want to pull back the curtain and show you the technical innovation that made it possible. My name is Hariharan Govindharajan. I'm a Senior AI/ML Architect at AWS. I had the privilege of leading the AWS Transform Agentic AI migration framework that accelerated migration by 10x time at CSL.

Eric's challenges aren't unique to CSL. Every organization faces the same fundamental challenges when modernizing at scale. The conventional migration approach involves three phases: assess, mobilize, migrate, and modernize. The assess phase is where we create a business case and we analyze the savings and we project the savings and then get started for migration. And then comes the mobilize phase, where you actually plan your servers from the data center for migration and the applications, and you discover those applications for getting ready for migration and planning. And the third phase is where we do actual migrate and modernize.

But if you see these three phases, and if you have to repeat these three phases for your entire set of VMs, it is a very labor-intensive and error-prone process. And typically, these conventional migrations involve teams from multiple geographies and regions. So that actually introduces even collaboration bottlenecks. So with the modern era of artificial intelligence, we thought of designing a first-of-its-kind agentic AI migration approach that would accelerate this conventional migration approach and have an agentic AI way of doing it.

When we looked at the agentic AI portfolio at AWS, we have three layers. The bottom layer here is for model builders, and that provides the compute for the AI/ML models to run, and that is mainly used for model hosting.

The second layer here is for AI engineers where they can go and build AI applications using completely serverless services like Amazon Bedrock, Agent Core, and they also have the SDK for agents like Amazon Bedrock Agents. On top of that layer, you have the enterprise users using no-code, low-code solutions like Amazon Quick Suite, Amazon Q Business, and AWS Transform. For the use case of accelerating migration using Agentic AI, we thought we could use AWS Transform in combination with Amazon Q Business to actually build the Agentic AI migration framework.

I'm really excited to talk to you about AWS Transform, the first agentic AI service to migrate VMware workloads at scale. AWS Transform has agentic AI capabilities which can automate wave planning and server mapping. It does this through specialized AI agents which analyze your complex VMware environments and orchestrate the dependency-aware migration waves. It also has the ability to convert the network configuration into AWS constructs. At the end, it actually enables the seamless cutover with business continuity.

If you look at the VMware migration using AWS Transform, it involves four main phases. The first one is discovery and data ingestion, where AWS Transform goes and discovers the data that is in your inventory of servers and ingests that data into the platform. Then comes the migration planning, where AWS Transform helps you to plan the migration waves, and the migration waves are nothing but dependent services grouped together that can be migrated in an atomic fashion. The third phase is the network conversion and migration phase, where AWS Transform helps you to actually convert your network constructs into AWS network constructs. The fourth and final phase is where AWS Transform can help you to do the actual VM migration and deployment.

With this, we're going to see the high level of how agentic AI-based migration approach works. The first thing you have to go through is the assess phase, where you will do the inventory discovery and you will use tools like Migration Readiness Assessment to actually create a business case for the migration and project the cost savings and ROI. This is really going to set you up for the business case in migration.

Then comes the mobilize phase. As I explained to you before, the inventory discovery and wave planning takes place. This is where exactly we used the agentic AI capabilities of AWS Transform to do the inventory discovery and agentic AI migration planning. Then we used the migrate and modernize phase to actually migrate the servers from data center to AWS that accelerated transformation through agentic AI.

If you look at the initial Agentic AI migration wave plan, we were able to actually run a small pilot and initially create 40 waves for 316 servers in just a day. We used ModelizeIT to extract and ingest a humongous amount of data into AWS Transform, and that included 555,000 VMware software data records, 3.96 million VMware network traffic connection records, and 190,000 VMware network inference records, along with 4,000 plus VMware instance details. This is a huge amount of data that is impossible for a human to process and come up with a wave plan in just a day. AWS Transform helped us in not only collecting this data but also understanding this data and rationalizing this data to come up with the most efficient wave plan.

Amazon Q Business Integration and Transformative Results: 10x Migration Acceleration

After the transformation, we do the wave migration. This created what we called a generative AI-based wave plan. When we created the generative AI-based wave plan,

we had a bunch of servers that can be migrated together in an atomic fashion, but then we needed a tool to actually discover those servers so that we can see what application those servers belong to and how those applications can be migrated as-is to AWS. So we used what we call Amazon Q Business.

I'm really excited to talk to you about Amazon Q Business and how it solved the challenge of application and server discovery in our agentic AI migration framework. Amazon Q Business is a new type of generative AI-powered assistant designed specifically to transform how we work with no-code solutions. Businesses of all sizes and across all industries are unlocking the transformational value of Amazon Q Business to achieve new levels of productivity and innovation. Given the ability to save the LLM prompts and create GenAI apps in minutes, it gives you great potential to actually do application discovery for all the servers in the GenAI-based wave plan. Amazon Q Business actually comes with a variety of data source integrations, such as SharePoint, Salesforce, Amazon S3, and more.

What we did actually in our framework is we created three Q Business GenAI applications. The first application is the server to application mapping. This is where we were able to actually identify all the applications from the servers that are there in the GenAI wave plan. Once we identified the applications from the servers, we fed the input into the second Q Business app, which is the migration evaluator app. The migration evaluator app has the ability to evaluate an application and a set of servers to be eligible for migration according to CSL standards of AWS. Then once the application is ready for migration, we fed the data into the application discovery app, which actually helped to discover all the low-level details of the applications like user authentication mechanism, RTO, RPO, and so on.

We were able to build these three apps by actually indexing 17,000 pages of infrastructure requirement documents of all these thousands of applications into Amazon Q Business, and that Amazon Q Business automated analysis and detailed discovery was completed in minutes instead of hours and created these three apps for us. With that, I'm excited to talk about AWS Transform Agentic AI migration framework, which we use to pioneer agentic AI migration with AWS Transform. As I said before, we were able to ingest 4.6 million records of ModelizeIT records into AWS Transform and were able to generate a GenAI-based wave plan, which actually accelerated 10 times our wave planning velocity.

We intentionally incorporated human in the loop as part of our framework, and hence any agentic AI or GenAI output is actually being reviewed by a human to make sure it is accurate and satisfactory. So as part of the next step after the GenAI wave plan is generated, we had our portfolio team work with portfolio architects to review the GenAI wave plan and make sure it is accurate as per the requirements. And then we created the reviewed GenAI wave plan.

Once we have the reviewed GenAI wave plan, we have the atomic number of servers that can go together in migration. So now it's time for discovery of all those servers in the wave plan. That's where we launched Q Business with Kendra Index and ingested 17,000 pages of IR documents into the Kendra Index of Amazon Q Business. And then we created the three apps: server to application mapping app, migration evaluator app, and then initial discovery app for the applications.

So these three apps, as I said before, helped us to actually find the applications of those servers, evaluate the application for migration, and then create an initial app discovery report.

So once the initial app discovery report is created, we again had the human in the loop as a process to actually review those initial app discovery reports with the app owners, and the app owners confirmed that the report generated by the generative AI was actually accurate. Then we created what we call the final wave plan and final app discovery report for all the servers that are there in that wave plan. Once we created those final reports, we ingested them again back to the Kendra index of Amazon Q Business and then created what we call the server intake form app.

The server intake form is an input to what we call Cloud Migration Factory, which is an open source tool developed by AWS to migrate servers from data center to AWS. The server intake form created the CSV file that is fed as an input to the Cloud Migration Factory. Again, if you notice, we have the migration team reviewing that CSV file input before it can go into the Cloud Migration Factory. Then the migration team ingested it into the Cloud Migration Factory, which actually helped to migrate servers from data center to AWS using AWS Application Migration Service. So this is how we built the transformative agentic AI framework to accelerate the migration.

Now it's time to see what are the results, the transformative results that we have achieved with this agentic AI migration approach. The first result that we saw is that there is a 10x improvement in the data center wave planning, meaning what would have taken 610 hours for the CSL team to plan the waves in data centers has only taken 60 hours when they used AWS Transform. That is a 10x improvement in data center wave planning. Because we were able to save a lot of effort in the data center wave planning, we were also able to improve our wave planning velocity, meaning we were able to actually plan for five to ten waves per week, whereas we were doing actually only two waves per week when we did this through the manual wave planning process. That's a 5x improvement in initial wave planning velocity.

When it comes to application discovery, it was taking us 60 minutes for analyzing and discovering each and every app. Now when you consider this to actually thousands of applications that are there at scale in CSL, it's going to be a huge number. So we were able to bring down that 60 minutes per app discovery time to just 5 minutes, and that has produced a 12x improvement in the initial app discovery. And if you look at the number of applications that CSL has, it is a huge set of applications, and that saved almost 1,000 hours of effort in total for us with this agentic AI approach.

The last one here is the migration evaluation time. Now the migration evaluation is a process which took a lot of time because you need to just go and analyze each and every app on the OS version and on the SQL Server version and other versions to make sure that it is all not having any technical debt to go into the migration phase. If it has a technical debt, it has to go to the modernization phase. So because of this deep analysis that is required, it was taking like 20 minutes per application to actually even evaluate that application for migration, and people were spending a lot of time in this. So we were able to actually bring this 20 minutes down to 2 minutes, and that's a 10x improvement in the migration evaluation time per application. That actually produced a lot of savings on the time and effort for us when we use this agentic AI migration framework. With that, I'll hand over to Erik to talk about how CSL realizes the benefits of agentic AI in the overall enterprise. Thank you.

Realizing Benefits and Key Lessons Learned from Agentic AI Transformation

Thanks, Hari. So sometimes we actually get the benefit from kind of being a little bit late and slower to the cloud party, right? You know, I think if we did this a year ago or so, the slide that Hari was talking about before would have been all done with people, with resources and labor, right? So as we think about now our journey forward and Hari spent some time talking about the improvements, you know, this is tangible value for us, right, from the perspective of the manual labor, reducing that effort to kind of bring forward. This is allowing us to think differently in terms of how we need to sequence our journey into the cloud.

By drawing different relationships and connectivity instead of trying to manually figure this out, leveraging some of these AI models and tool sets allows us a different level of unlock and allows us to think about this at a different level of pace and velocity. Our journey isn't done, so as we spent time on the analysis side, now this continues into how are we going to architect and redesign some of these platforms and some of these solutions and capabilities. Ultimately, we have a continued journey in front of us over the next set of years that will continue to leverage what we've established from a foundational perspective.

Much of the things that Hari has talked about is getting embedded back into what we're terming our digital core, that core foundational platform for us, so that we could take advantage of all this information, take advantage of all this data and all this learning. It's not just a one and done activity. Rather, this continues to get built within our overall knowledge base on our delivery cycle.

As a closing point, one of the things we just want to talk through is some key top line lessons learned. One of the things to start with is clarifying roles and responsibilities is something you really need to spend time on, and you need to be able to challenge yourself differently. The roles and responsibilities that you have in your organization today and the people that think they have and currently own certain decision rights may not be the same individuals that will own the same decision rights moving forward. As you look to transform your landscape, there is a different set of responsibilities that start to occur. I think a lot of organizations tend to try to fit the future into a static current state, and if you do that, that's where you're introducing recipes of stagnation and not achieving the velocity and the agility that you need to really achieve.

The second point is, as Hari talked through, this was all new tech for CSL. Being in the industry that we're in, we tend to take things in a very risk averse manner. We are a highly regulated environment. A lot of things require us to go through validated processes, so our processes weren't built for new tech and weren't built for the speed to bring these tools in at the pace that some of our partners wanted to introduce them. It required us to think about what processes actually need to change and what are the set of processes that need to exist to embrace and bring this technology in. It's not something we got right day one. It's something that we're still learning and adopting and growing on, but that's something you really have to think about. What is the speed that you have to be able to operate at to bring this stuff in?

The key is you have to establish and have a true partner ecosystem. There was a lot of tech here that Hari talked about. This was not tech that we had teams and teams of people that have implemented any number of times, so we leaned heavily on our partners and some critical partners, specifically with AWS as well as with another partner, Accenture, to really help bring this technology into our organization and be the ones that could actually steward us. That required us to develop a level of trust that I would say the organization wasn't really necessarily comfortable with, abdicating certain decisions to get back to the first point. Who has the decision when we need to bring new technology in, and who do we have to trust? How do you have to start to trust your partner ecosystem in a different manner so that they're not just given a set of requirements and you're going to assume that they're going to execute it to the manner that it needs to be? Again, you don't want to hold back your partners because we did not want to bring our partners in and say do it the way CSL does it, because that's not what we were looking for. We were looking for partners that could say tell us and teach us how we need to do this, because this is not something we have scale and experience at. Creating that right ecosystem and that right framework is a key critical thing.

That's the closing point. Again, thank you for spending time with us this morning. On behalf of Victor, Hari, and myself, we appreciate you giving us a little bit of time, and if anyone has any questions, more than happy to answer them as we end this session. Thank you.

---

; This article is entirely auto-generated using Amazon Bedrock.