🦄 Making great presentations more accessible.
This project aims to enhances multilingual accessibility and discoverability while maintaining the integrity of original content. Detailed transcriptions and keyframes preserve the nuances and technical insights that make each session compelling.
Overview
📖 AWS re:Invent 2025 - Infrastructure for the impossible: Turning public sector barriers into...
In this video, AWS Vice President Dave Levy presents public sector innovations including the $50 billion investment in AI and supercomputing infrastructure for U.S. government, AWS European Sovereign Cloud with €7.8 billion investment, and AI Factories for global deployment. Lawrence Livermore National Laboratory's Greg Herweg discusses their fusion ignition achievement and cloud transformation journey. The U.S. Navy's multi-modal agentic AI solution reduced submarine maintenance processing from weeks to hours at Impact Level 5 classification. Capita CEO Adolfo Hernandez demonstrates their Process to Agent framework achieving 40% reduction in handling time and 25% throughput increase for UK government assessments. AWS announces 39 new Imagine Grant recipients including Jane Goodall Institute receiving the Pathfinder Award and $1 million for digitizing 65 years of primate research using Amazon Bedrock.
; 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
Introduction: Building Infrastructure for the Impossible
Please welcome to the stage, Vice President of Worldwide Public Sector at AWS, Dave Levy.
Right. Good afternoon and welcome to the Public Sector Innovation Talk. Whether you're a first-time re:Invent attendee or a regular, we want to welcome you to this year's re:Invent. Now, you're a builder, and because you push boundaries, your organization's critical missions require the most innovative solutions. And AWS innovates on your behalf. We take your mission, we work backwards, and we go through the things to understand where you need to be.
And we know that you're solving the world's most complex challenges. It's through innovations like AWS GovCloud, which provides government agencies with secure and compliant cloud infrastructure, AI that accelerates critical missions like military operations while meeting the highest security standards, and even multimodal technology that analyzes decades of disparate research to unlock scientific discoveries. We're going to talk about all of those things today, but let's start with an innovation story not too far from here.
In the early 1900s, the Colorado River was a powerful and unpredictable force behind the catastrophic floods that damaged homes and farmland. So in 1931, a group of builders converged along the Nevada-Arizona border to start construction on the Hoover Dam. They completed this engineering marvel and symbol of American resilience in 1936, two years ahead of schedule.
Hoover Dam is a perfect example of an idea that many people didn't fully understand at first. They thought it was impossible, too big of a risk, and too ambitious. It was a massive public works project full of engineering challenges, and it was during the Great Depression. But the team foresaw a better future and went ahead with the project anyway. And look at all the breakthroughs. The dam controls flooding, generates electricity in three states, irrigates millions of acres of farmland, and provides a reliable supply of domestic water.
Yes, big projects come with big risk and uncertainty, but doing hard things and conquering the seemingly impossible is what sets the stage for innovation. That's why AWS has always delivered and continues to deliver the foundational infrastructure and technical capabilities for the public sector's most critical initiatives.
AWS GovCloud: A Pioneering Achievement in Secure Cloud Infrastructure
So let's dig in. Our public sector innovation story started 14 years ago when we launched GovCloud, becoming the first cloud provider with a purpose-built infrastructure for controlled unclassified workloads. GovCloud was our Hoover Dam moment. It was a massive project, but we listened to our customers and worked backwards from their needs. And remember, this was 2011 when cloud was only a few years old, particularly in the public sector. So the idea of a purpose-built cloud for government was something many people didn't understand, and they certainly didn't think it would work. And we built it anyway.
And we faced unprecedented technical hurdles, including the need to build a physically isolated, fully compliant infrastructure. And so this meant things like custom hardware security modules, specialized network segmentation protocols, and new identity management systems that could verify U.S. citizenship status for every AWS operator. Now the technical complexity increased exponentially when we built AWS Top Secret Region in 2014, and the Secret Region in 2017.
Consider these engineering challenges. How do you provide cloud-scale computing while implementing controls for handling classified information at multiple security levels? Our solution involved new cryptographic modules, air-gapped networks with specialized interconnects, and custom hardware security modules that could handle compartmentalized workloads. October's launch of the AWS Secret-West Region is evidence of continued innovation in government regions. The technical achievement here was creating an active-active architecture that could maintain synchronization of classified workloads across continental distances while still meeting strict security and latency requirements. Behind the curtain, AWS engineers solved complex problems around network routing, load balancing, and failover mechanisms.
So why is this important? Because processing classified information closer to analysts and mission programs helps accelerate the development of advanced defense and intelligence capabilities, and most importantly, it ensures continuous operations to keep our country safe. What makes these achievements particularly compelling is that we engineered solutions while working within strict regulatory frameworks. These technical advancements met government requirements, they fundamentally transformed how agencies operate, and they literally ushered in a new era of innovation. GovCloud is just one example of the culture of innovation that drives our work in the public sector and with our customers all around the world.
The Next Hoover Dam: $50 Billion Investment in AI and Supercomputing
So what now? What's the next impossible innovation that we can bring to the public sector? When we talk to customers, two challenges keep popping up. The first is supercomputing. Supercomputers are needed for mass-scale simulations in cyber, energy research, and scientific modeling. They can take years to build, and they can be complicated to use. The second challenge is capacity. Customers want to train large language models and run those models in everything from defense programs to weather forecasting, but building enough capacity on their own is difficult and time-consuming.
So here's what we're doing about it. Last week, we announced our next Hoover Dam. It begins next year, and it's an investment of up to $50 billion in AI and supercomputing infrastructure in the United States that's purpose-built for government use cases and security requirements. The commitment is the largest federal technology investment in Amazon history, and it demonstrates our mission to be a leader in government cloud and AI. What does this mean for federal agencies? Almost 1.3 gigawatts of capacity. For context, the Hoover Dam produces about 2 gigawatts. For some of you movie buffs out there, if you remember Doc Brown in Back to the Future, this is more than the 1.21 gigawatts he needed to travel back in time.
While we can't promise time travel, this infrastructure will accelerate your missions in ways that seem like science fiction today. By providing on-demand access to that much computing power, government customers can run sophisticated AI workloads and process huge data sets for their critical applications, things like autonomous systems development, energy innovation, and genomic data processing. This is going to revolutionize high-performance computing for the United States government.
Global Expansion: European Sovereign Cloud and AI Factories
As a global company, we have international customers who face many of the same challenges, which is why we're undertaking another massive infrastructure project for Europe. The AWS European Sovereign Cloud will be the first fully featured, independently operated sovereign cloud backed by strong technical controls, sovereign assurances, and legal protections designed to meet the needs of European governments and enterprises.
Only personnel who are EU residents and working from EU locations will have control of day-to-day operations. Backed by an investment of €7.8 billion, the AWS European Sovereign Cloud is set to launch its first region in the state of Brandenburg, Germany very soon. I'm looking forward to the advancements that the European public sector organizations will drive while meeting their unique sovereignty needs.
I'm confident that in a few years, we're going to look back favorably on these AI supercomputing and sovereignty investments. Like the way we view GovCloud today, we'll see these commitments as projects that led to unprecedented breakthroughs for our customers. But breakthroughs aren't exclusive to the US and Europe. Yesterday, we launched AI Factories, which bring purpose-built AI infrastructure directly into the customers' data centers. You provide the data center and power, and AWS handles everything else.
Our AI Factories can eliminate build effort and maintenance. By leveraging two decades of AWS cloud expertise and operational excellence, customers get access to the latest technology, GPUs, petabyte-scale networking, and comprehensive AI services like Amazon Bedrock and SageMaker, all while helping to meet security and sovereignty requirements. Factories will help accelerate time to value for governments around the world. A couple of weeks ago, Saudi Arabia's HUMAIN named AWS its preferred AI partner globally as it builds an AI Zone in the kingdom.
Mission-Critical Innovation: US Navy Submarine Maintenance and Maverick Mobile Lab
This expanded partnership is part of our $5 billion investment in AI infrastructure, AWS services, training, and talent development to advance Saudi Arabia's mission to be a global leader in AI. Now let's come back to the US for an example of the type of innovation that's unlocked when customers have purpose-built infrastructure. The US Navy Commander of Submarine Forces faced a significant challenge with vessel maintenance. The submarine maintenance work required sailors to manually navigate multiple systems and hundreds of documents, and this critical process took weeks and increased the risk of errors that could impact operational readiness.
Remember, this is the Navy. There are a bunch of security classification requirements that limit what types of AI models and tools they can use. But the Navy worked with us at AWS to develop a multi-modal agentic AI solution that has transformed the submarine maintenance process. Their solution accelerates maintenance work by using multiple agents to automatically generate official forms and search across numerous submarine-specific databases. It operates in AWS GovCloud with Impact Level 5 classification, and it's the first generative AI deployment at this security level for the Navy.
IL 5 is the Department of Defense's highest security level for unclassified mission-critical data and requires extensive controls that few AI systems can meet. Maintenance processing that took weeks now takes hours, helping the Navy keep its fleet mission ready. Now the Navy's success shows what's possible when we bring the right infrastructure to mission-critical environments, but innovation doesn't stop at the data center. We're bringing it directly to customers in the field.
I'm thrilled to introduce Maverick, a mobile innovation lab for autonomous systems that AWS built in collaboration with our partners, including Anduril and Gambit. We transformed a standard vehicle into a rolling laboratory that allows AWS to work side by side with customers to rapidly prototype and develop solutions in mission environments. For example, we recently used Maverick to demonstrate conversational drone control. Operators typed simple commands like search the Northeast quadrant, and Amazon Bedrock agents translated that natural language into an autonomous drone mission executed at the edge.
Maverick runs on AWS Outpost server and integrates with partner hardware and software. With Anduril's Lattice system providing real-time visualization, Gambit's autonomy software for mission execution, and Amazon Leo satellite connectivity,
we're proving how integrated partner technologies accelerate innovation. This cloud environment on wheels enables us to develop breakthrough solutions with customers for search and rescue, disaster response, and defense operations.
Lawrence Livermore National Laboratory: Transforming Scientific Discovery Through Cloud and AI
And innovation doesn't just stop at the edge. It extends to our country's most advanced scientific facilities. Lawrence Livermore National Laboratory achieved something in 2022 that scientists had been trying to do for 70 years: fusion ignition. They've since repeated it several more times, and AI played a critical role in making the impossible possible. So let's meet someone who's leading that transformation. I want you to help me welcome to the stage Greg Herweg from Lawrence Livermore.
Greg, thanks for joining us. Happy to be here. First, I want to say I'm super jealous because working at Lawrence Livermore in the National Ignition Facility is probably such a great treat. I would say not the least of which because that's where they filmed a bunch of Star Trek. They filmed a reboot of Star Trek in the mid-2000s. Exciting place to be and you've seen it. We gotta get you out there again. I do. I want to stand next to the warp core again one more time.
So, hey, let's dig in. When I look at what you've accomplished, like fusion ignition, I mean just remarkable. Coming from a family who's involved in nuclear physics, I only know what fusion ignition is, but actually accomplishing it is remarkable. So you're accomplishing fusion ignition, this massive cloud AI and AI transformation. And it strikes me that Lawrence Livermore is really setting the blueprint for how national labs can leverage cloud and AI. So Greg, as you look ahead, what's next? I mean, what are you guys doing and what advice do you give to the other government organizations here thinking about this kind of transformation?
Well, we leaned in early on cloud. We've got about 10,000 people at our site. We've got over 2,000 PhD scientists. And we do a variety of things. We've got a lot of operational stuff that happens just to run our small city, and we have a lot of science that happens, like you said, with NIF. And we leaned in the cloud early. I'll say that, you know, we got a lot of tough characters, right? Scientists like to see things. They want to prove. They can be skeptics.
There were a lot of advantages to the cloud. I'll say one of the things that we looked out to was creativity. We have a lot of builders, a lot of developers, a lot of creative minds there. And while we have 2,000 PhD scientists and a lot of supporting cast, not everybody's a computer scientist. So things with AWS and the cloud give a much broader palette for creativity. That was one of the things we looked for, and I think that's one of the things that attracted us because the more creative we can be, the more science we can do.
But you know, there were skeptics to begin with. Like I said, it's a tough crowd. And one of the things we had to do first was get people's heads around it. And when I'd first go around and talk about cloud, people were like, yeah, we're a government installation. That is never going to happen. And you know, hey, you're a little crazy. And then we worked through with some, and your team helped us along. We got through, then a few people finally said, well, okay, maybe. And then we got to the point with unclassified, which was, well, why aren't you done yet? I'm like, well, we could have gone faster if you didn't have the skeptics, but you know, you gotta prove it to them.
So with classified right now, which, you know, we've started doing, it was about 10 times worse, 10 times more skeptics. And when we started with that, it was, you know, a replay of the same thing. You're not gonna be able to do that. Stop talking about it. In fact, it was like 10 times worse. People told me you're really making people upset and making people angry. This is not going to happen. Stop talking about it. But I persisted along with help from your team, and we're actually at a place now which is, well, we need to, of course it's great for the mission. It's necessary for the mission. Why aren't you done yet? Why aren't you moving faster? So I'd say on this, it's about partnership and about leaning in and doing new things together.
That's amazing. Let's move fast. Look, there's nothing more difficult than trying to convince 2,000 PhDs of something. I want to talk a little bit more at a high level, and you've got an incredible mission there from nuclear deterrence to fusion energy, biosecurity research. But when I think about what you do, the phrase infrastructure for the impossible kind of comes to mind.
And so when you achieved that fusion ignition in 2022, something people thought was impossible, you did it, and you're running one of the world's most advanced supercomputing facilities. Let's talk a little bit about how cloud's enabling that at that level. Well, I think for that it was about convincing people, which we talked about before backstage. There's always going to be skeptics, not going to convince everybody, but it's about the creativity, it's about other things that cloud brings, agility and whatnot. While we tried to weave our way through what was possible, what we were going to do, we actually had to get through a few of the skeptics.
I decided to not try to convince everybody with PowerPoint slides but to kind of show them. What we've done to guide our way through that was run pilots at small scale, prove that things worked, and move on from there. One of the things I've pushed on from the beginning was what I call green blinky lights, which my team has made fun of me over time. Your team has made fun of me over time with green blinky lights, but I'll say I've made progress because it's catching on and I now see some of your team and my team actually using the phraseology, so I'm happy to see that. I think we've made progress because earlier on I would see that a couple of the barriers would be cloud isn't going to be fast enough, it's not going to be secure enough. Those things don't come up now, right?
A few other things come up, cost, loss of control, things like that, which helps with a strong partnership. It'd be great, I know that some of the AI vendors did this dollar for a year. You were a few minutes ago talking about supercomputer. If we could get supercomputer for a dollar for a year, that would be great. You can work on that for me. I think one of our challenges right now to evolve this and get to the next step is what I'll say hybrid. We have a unique environment, a big site, a square mile, hundreds of buildings, new buildings, old buildings, a variety of folks.
For a lot of good reasons, we will probably need to have some stuff on site for the foreseeable future. So I think one of the next nuts to crack is this hybrid thing. How do we take advantage of all these great AI things that are out there, all the new buildings you're going to build for us and all the facilities there, and sort of match that with some of our on-premises resources and how to really do that well. I think that's a nut that hasn't been cracked, so I think hybrid is the thing and I think good partnership will help on that.
That's fantastic. A couple of takeaways from this part of the discussion. One, we probably just need to build an agent that can go out and do green blinky lights for everybody. That will probably work well on Capitol Hill to sell the project. But also, just the ability to work at this scale and integrate with all the things that you're doing at Lawrence Livermore National Laboratory, it sounds like it's just a tremendous opportunity. What I hear in that is that you're dealing with massive computational challenges, you've got decades and decades of operational data. We talked about working at the classified level, so we've got some of the most sensitive national security work in the country, and you're moving a large part of the infrastructure to AWS, so it's not a small thing.
What drove that decision? You've talked about convincing some of these brilliant scientists that it's the right move for the lab, but what were some of the original drivers that got you there? I think some of the creativity stuff, I think some of the agility. One of the things that is important for us, again, we do operations on one side, we run a small city, you've got to keep that going. For the research side, it's about efficiency but efficiency in a different way. It's about being able to go through this discovery science loop more quickly, more quickly, more quickly, because that brings more science.
On the operations side, if we, the more efficient we can be there and spend less time managing our own stuff, building our own servers, things like that, then we can put that into science. On the science side, the more we can actually let scientists do what they want to do well, then they can go through that discovery loop much more quickly. It's more science, more discovery for the country and particularly with all the national security type things of utmost importance.
Yeah, that's great. So let's pivot a little bit. You've got another big move going on and that's deploying AI across the entire lab. So today you're working across mission areas and model developers. You're expediting pilots, doing data gathering across these all environments. Take me through that thought process. How does your lab go from kind of maybe AI could help to 10,000 people who are using it today to where you're going to go with AI at Livermore?
Well, you know, as you know, and everyone in the audience probably knows, things are moving really fast, faster than I've ever seen, and I've been doing this for a little while. So one of the things we've actually done at Livermore was, I'll say, a great year of experimentation. One of the easy things we thought we could do, low hanging fruit, was to get multiple LLMs in the hands of all our scientists. Now it created a little bit of a wrinkle for us because we have to do some of these accreditation things, FedRAMP, get approval from a lot of people. So there's a few wrappers we had to put around a few of those, but we're trying to put LLMs in the hands of everybody so they can experiment and see what's a good fit for their job.
And then we're also trying to basically upskill the whole workforce, our 10,000 staff roughly. We've got a whole program going on where we're trying to upskill, which is we're doing on-demand training, live training. We've had a lot of partner days. You guys have participated in some of those. We've brought in a whole host of speakers to get everybody energized and thinking about it as part of their job. And it's one of the things actually going back to when we started cloud, people would ask me, well, we've got all this other stuff going on, you want to do this cloud stuff, how many people will you need? Like five or ten or twenty, you know? Like, no, no, no, everyone. Everyone that's involved in IT needs to be a cloud person. And I tell them the same thing now. They ask, well, how many people are going to have to do this AI stuff? Everyone. It affects everyone. And in our laboratory, which we do a lot of operation stuff I've talked about before, it doesn't matter whether it's finance or facilities or the legal team, everybody's got to do it. So we're trying to upskill everybody.
One of the things that I'm seeing is people are reimagining how they could do their jobs at the laboratory, no matter which side they're on, whether it's science or operations, which is great. They're reimagining how they could do it. But on the downside, and this is where some of the challenge comes in, you can't just plug in an LLM and they magically reimagine how they do it. There's a lot of process reengineering. So one of the things we're actually doing right now is trying to get people's heads wrapped around the thing you want to do to make your environment better is some process reengineering. And it happens on our facility side, it also happens on our deep science side as well. They're trying to sort of reimagine the process, and there's some elbow grease involved. So I think that's part of our challenge there, is just getting people's heads wrapped around elbow grease.
Yeah, I think that's great, and you're giving them the tools, the models, the computational power, fantastic. So Livermore and AWS made a joint announcement earlier this year about the adoption of AI to support the NIF, one of my favorite places, as I've mentioned earlier. Given the successful deployment of AI-driven troubleshooting and reliability management at NIF, how do you plan to scale or adapt this technology to the other areas and across the lab?
Well, we actually started an endeavor I'll call Smart Lab in our laboratory a little bit before we actually did the joint thing with NIF, but NIF has sort of accelerated that thinking, got more people on board with it. And what the Smart Lab initiative is, is basically take our square mile, which has, believe it or not, about 600 buildings dating back to World War II all the way to modern buildings and everything that would go along with those. And what we really want to do is instrument everything, industrial controls, IoT everywhere, which people also thought was just as crazy as cloud was way, way back when. People are starting to wrap their heads around it. We want to instrument everything, all of our day-to-day operation stuff, as well as our scientific endeavors. We have actually manufacturing on the floor, we have lathes and mills and 3D printers, and we actually want to instrument all of those things, bring data back.
We're working on building a large data repository. We've always had lots of data, but it's been a little bit more siloed from this project to this project. And what we're trying to do is build sort of a data repo that we can have all of our operation, all of our science data, you know, unclassified for the purpose of this, and be able to bring it back, get telemetry, make smart decisions, have AI be a part of that. And I think this is one of the things that's following from the NIF endeavor is, you know, Smart Lab data repo.
One of the things I'll say is a challenge with this is a third to half of our site is classified buildings, classified area, classified projects. And all this sounds really cool until you say, well, I want to put a sensor in a classified room on a thing that makes classified devices. And while probably lots of people in the audience work at places that make really cool devices that people are really, really happy to use, it doesn't exactly translate to us all of the time. We either have to jump through a lot of hoops, which takes us a lot of time, or we just can't use it, so.
Volunteers in the audience here going forward, people that can help us get through that. Think about people that work in air gap environments, classified environments, and I think that's part of the extension of where we're going to go with NF is take what we've learned there, put it in other environments, and it'll make us better, and it's going to help the complex.
That's fantastic. What a remarkable kind of transformation. The incredible work, you know, there's some of us that get to have just a little bit of view of the work you do at Lawrence Livermore and all those folks do, and the amount of breakthroughs and research is just remarkable. Any kind of last tidbits of advice for public sector organizations and public sector teams and leaders that are out there thinking about cloud and AI adoption? Any last little tidbits?
Well, I think, you know, the old adage, crawl, walk, run, start there, get something going. You'll probably have some naysayers. If there's a bone you can throw to the naysayers to get them on board, something's in it with them, that's good. I think the other thing is get some of the people that are just eager to learn, eager to do something new. You don't have to convince everybody at once. Get them on board, and then go do this like build, pilot, run, and get the blinky green lights.
Yeah, for sure. All right, and I also heard the challenge you put to us, the $1 supercomputer. We'll get to work on that. I expect to hear something soon. All right, thank you so much, Greg. That's a great conversation. Thank you.
Healthcare Transformation: Agentic AI for Patient Engagement and Authorization
Wow, it was a great conversation with Greg, and I've had the privilege of visiting Lawrence Livermore a few times. When you see the breakthroughs and the research and the commitment that those individuals have to their mission, it's just really remarkable. We're proud to continue supporting Lawrence Livermore National Lab, and their work is really critical. So, why don't we move on and look at AI adoption elsewhere.
In healthcare and life sciences, customers are interested in implementing AI solutions that go way beyond chatbots and assistance. So once again, AWS is innovating on their behalf. Amazon Connect is our AI-powered customer experience solution designed to make customer interaction more efficient and effective. We recently shared some new agentic AI capabilities for healthcare patient engagement from Connect.
Think about what happens when you try to schedule an appointment with a specialist. How long are you on hold? How many times do you have to repeat the information? US health systems are drowning in patient communication, including millions of calls annually, along with an increasing volume of messages and emails. So when routine interactions like booking an appointment become obstacles, patient loyalty erodes. This puts additional pressure on our healthcare providers who are already stretched thin.
That's why we're making it easier for our healthcare customers to automate routine patient interactions and empower their staff to focus on what matters most, the experience that their patients have with them and delivering exceptional patient care. And our new agentic patient engagement capabilities deliver a personalized patient experience with secure, real-time integration with electronic health records. All of this is up-to-date and accurate information.
Agentic AI handles routine tasks like appointments, scheduling, and patient verification, while the human staff focus on more specialized patient needs. And providers and their patients are already seeing Connect's benefits. UC San Diego Health has reduced patient abandoned calls on average by 29%, with some departments as high as 59%, and call handle times are 21% shorter. Jupiter Medical Center in Florida reported a reduction in their radiology backlog of 60%, bringing appointment wait times down from two weeks to just 24 hours.
So let me share another example of agentic AI's power. Think about the last time you or a loved one needed an insurance approval for a medical procedure or a test. The wait can be stressful, and it's important to you as a family and as a patient. And providers don't like this process either.
It's heavily manual, error-prone, time-consuming, and Cohere Health annually processes more than 12 million authorization requests for more than 600,000 doctors. They're using AWS to fundamentally change this process. Using Amazon Bedrock AgentCore, Cohere Health has reported a reduction in prior authorization review times by 30 to 40% while improving clinical decision accuracy.
Cohere Health's solution helps payers analyze both structured and unstructured data, such as clinical records, patient notes, and faxes to quickly identify evidence and validate the medical necessity of a required treatment. For patients, this means quicker access to the care that they need and better healthcare outcomes. It's what agentic AI can do for healthcare, and imagine what's possible for government.
Capita's Ground Zero Moment: Revolutionizing Government Services with Agentic AI
Picture intelligence analysts who can query across all their data sources in natural language. Think about policy researchers who can generate comprehensive reports from thousands of documents in minutes. Visualize citizen service representatives who have instant access to all the relevant information they need to resolve complex cases. This is the future of government, and at the forefront of citizen and public service transformation are customers like Capita, who's working with some of the world's largest government agencies.
Capita is taking traditional business processes and transforming them into intelligent, responsive agents that fundamentally change how citizens interact with government services. Let's find out how. Please welcome to the stage Chief Executive Officer of Capita, Adolfo Hernandez. Thank you. Good to be here. Good afternoon. It's certainly great to be here, and I want to thank a good old friend, former colleague, and a partner of ours, Dave, for the opportunity to be here and give us a chance to share with you what we've been doing together with AWS in this very fascinating and really exciting space.
For those of you who do not know Capita, we've been operating for nearly four decades. We are headquartered in the UK, we're listed in the London Stock Exchange, and we've had the privilege to work with nearly 500 different government bodies and institutions over the years. We serve over 22 million citizens and customers every year, and as a result of everything that we do, we have over 100 million interactions per year. That's given us a foundation and a set of experiences that gives us the opportunity to ask, where do we take this in the times of agentic AI?
Because the reality is that the BPO industry is fundamentally changing. Anybody who's been looking at the BPO industry for the last quarter of a century would recognize that this was about grabbing the old legacy processes, getting the old legacy teams, getting the old legacy systems, and just managing it in the best possible way. It did deliver results, but it was a very old way of solving problems.
Of late, in the last few years, we've seen an injection of technology into this area. We have seen now the arrival of bots, we've seen more analytics, we've seen some more automation, and we have seen a jump in productivity that's definitely there today. We've all experienced it as users, as citizens. We've probably experienced it a little bit less as users of public sector services, but we're starting to see that there is definitely an acceleration, and this is absolutely great.
But with the arrival of agentic AI, we're seeing a fundamental transformation. I am really, really excited about this because I think it's like agentic AI was designed to fully transform the BPO industry. I like to define it as this is our ground zero moment because absolutely everything that we do can be captured, can be improved, can be streamlined, can be trained, and eventually we've got the capability to have agents operating independently.
Within guardrails, with evidence control, with traceability, with audit, and then having humans working alongside those agents, fundamentally orchestrating the experience, delivering the, shall we call it, unhappy path experience, bringing a number of services that the agents can't do alone.
But if you think about the future of public services or the future of the services that we provide in call centers, in telecoms, in utilities, in local government, the potential to fully transform that is huge. So we have embarked on a journey to work with partners like AWS to fundamentally transform and digitify everything that we do.
We are going to do it in a pragmatic way. It needs to be systematic, it needs to be real, it needs to be grounded in experience, and it needs to be enriched by the learnings of the likes of AWS and everything that they learn about deploying really complex services at scale. So we've created this AI Catalyst Stack that, as you can see, is based on that nearly 40 years of experience understanding and mastering business processes. We can check against 100 million transactions every year. We can also work with the likes of AWS to give us the best and the most interesting features on how we deploy and architect the scalable services on the cloud.
And then ultimately we've also got the capability to inform how we do this through the process to agent framework. So how do you take an old process, how do you understand the process, how do you capture it, how do you optimize it, and eventually, how do you transform that into a fully digitified process? So as we go and do that, we're starting to integrate technologies, right? And we're starting to integrate a lot of the technologies that we know about. We've got Amazon Bedrock, and a number of others.
And actually I was really excited about some of the announcements we saw yesterday from Matt because I think some of the new things that we're seeing now on evidence and evaluation and what we're seeing in policy, it's only going to be more and more relevant in general in the enterprise, but specifically in regulated industries like the public sector. So I think having this type of architecture where you go modular and you start with the citizen, you start with the consumer at the beginning, and then you try to understand through that business process observability layer what is actually happening. Where are the clicks going? Where is the time going? Where is the frustration going? Where is the friction?
And then moving that all the way down to the application layer, which is really when some of the capabilities of AWS really, really help here and the integration, and then work all the way down to integration into legacy systems. It becomes very modular. We can replace or change or insert different components based on what we do for a particular case in any agency, and it gives us the ability to walk backwards from every customer problem, but we do that in a structured manner because then we're building IP, we're building APIs, we're building interfaces. And also the training that we're able to give to our 35,000 employees around the world remains consistent.
Now, when we get to go and do this, you might be saying, okay, is there a real life example? And I think this is where it really gets interesting. We work with a lot of government agencies, and there is a government agency that we work for in the UK where we serve and we process a lot of evaluations, assessments, in the millions, and obviously that has an impact on hundreds of thousands of lives every year. And then we started to look at this and we said, okay, what's not working with the manual process?
And then it's obviously the workflow. The workflow is complex in itself. You look at the huge amount of evidence when it comes to assessments, right? It's processing a lot of documentation, different formats, some of it is still handwritten, coming from different sources, different administrations, and then you've got that service delivery bottleneck that every organization is dealing with. This is happening at a time with any government, including the UK government, where they have a lot of constraint in terms of resources, there's a lot of pressure on backlogs, and there is a huge opportunity to improve.
So in what I would define a truly Amazonian style, we launched a sprint to work together with AWS and with BCG to start understanding the process, what was involved to get from A to B, where was the elapsed time, where was the time being wasted, where were there potentials for acceleration at the process level. Then we started capturing those, documenting those, and then ultimately mapping them to agents. So we look at, you know, well how do we deal with evidence, right? What can we do to process that evidence and understand it, process it and upload it, created an agent in that space.
The next one is how do we go and manage the case itself? How do we get the right case to the right person? They all do intelligent routing, so how is that going to happen? The next one is how do we assist the process of the assessment? How do we get transcription, for example, and make all of that available so there is less handling time and less processing time as a result of that?
Then there's the final decision making. So how do we go check against policies? Do we need to have everybody be a super specialist and have to make everything on their own, or can we have some assisted decision making? And then finally, the ever so important in regulated industries, how do we ensure that this is being done according to the quality expectations and metrics that citizens and the government agencies expect? So what is really good is at the end of this, now we have agentic AI being deployed to work independently but together with humans working alongside that whole process, making sure they stay close to the unhappy path, making sure that they do governance, and making sure that they actually are able to enforce the right decision.
And the results in just under 12 weeks speak for themselves. Probably the most important one is a 40% reduction in handling time. If you sort of process that through, you're like wow, that's a big reduction. That has an impact on everything. It has an impact on the amount of cases that you can process, so the throughput goes up by 25%. Then we've also seen improvements in quality. When you have improvements in quality, you have less issues, you have less escalations, you have less exceptions to go and manage, but you also get the benefit of a happier citizen who's going through a more predictable process.
You have a happier set of professionals who are doing the assessments because agentic AI is helping them deal with the really complex situations and really complex data volumes and the really complex policies, and they can really put their emphasis on what really matters. So what's really important here is that to do this is more than just technology. I think the technology that hyperscalers are bringing to market is spectacular. I think we saw some of the stuff that Marc Garman was talking about yesterday, which was brilliant, but we need a lot more than this.
We need to really understand and start with the process. We really need to get the colleagues, the employees, and people trained to really make sure that they understand that. And to scale this up, you can't just do it organically. You need to create inside the organization a factory or some capability to do that. We have called that the Catalyst Lab. The Catalyst Lab in Capita, what it does is sort of identify and prioritize which business processes, which opportunities like this lend themselves to be identified, and what would be the right design, how we would go about it, what are the right guardrails, how do we implement it, how do we audit it, and we actually manage all of that process too.
And there is a little bit of thinking that needs to be involved in what is that P2A, that Process to Agent framework, which is really important, but it's a set of steps that allows us to move from that linear process that I described at the beginning and get us to the other side. So I think everybody could be doing this. I think we just started it and we're very, very excited about what we can do, what we are already doing, and we've learned so much about this one that our Catalyst Lab is using this one now as a blueprint to deliver the same services to other government agencies that need some form of assessment or case management, and we're really excited about scaling that and taking it to market with partners like AWS who are really making great investments and supporting this area of identification of business processes.
So as I said at the beginning, this is ground zero for agentic AI, and I think BPO has probably been a little bit of a boring space for the last few years. I think you've got to sort of fasten your seatbelts. It's going to be extremely exciting. Everything that we're going to be able to reconfigure in how we provide services to consumers and to citizens. I'm delighted to have a partnership with AWS to make that happen. Thank you. Please welcome Dave Levy back to the stage.
AWS Imagine Grant: Empowering Nonprofits with Technology
Thank you so much Adolfo. Thanks for having us. All right, thank you. Thank you Adolfo. I'm so excited about what Capita is doing with P2A and what it means for the future of government service delivery. I think what I heard there is really exciting: happier citizens, happier professionals.
It's really a future that we can all start to embrace. Throughout the session, you've heard about pioneering work by AWS with our customers. When we apply advanced technologies to the nonprofit space, we're able to address chronic and systemic societal challenges.
AWS is committed to helping address nonprofit needs to advance their missions, and one way we do this is through the Imagine Grant, which provides financial and technical support to nonprofits pursuing innovative technology projects. Since 2018, we've awarded $21 million in Imagine grants. And today, I'm pleased to announce 39 new recipients who make up the largest ever cohort of Imagine Grant winners from across three continents. They're using generative AI and cloud technologies to transform their missions across healthcare, education, humanitarian services, and scientific research. So congratulations to all these amazing organizations. Please give them a round of applause.
Jane Goodall Institute: Unlocking Six Decades of Primate Research with Generative AI
I want to highlight one of the winners. Nine months ago, I had the opportunity, really of a lifetime, to meet and interview the late Dr. Jane Goodall. She left an indelible mark on our world through her contributions to science, wildlife, and society. Her work and legacy continue through the Jane Goodall Institute. JGI is using AWS to unlock six decades of groundbreaking primate research.
They're building a generative AI platform that's digitizing and analyzing an extraordinary archive that's more than 65 years old. It contains hundreds of thousands of handwritten notes in multiple languages, thousands of hours of video, and countless photos and maps from the legendary Gombe region and research site in Tanzania. Using Amazon Bedrock with Claude models, JGI is translating complex field notes that mix English, Swahili, and local dialects. Their system can even identify chimpanzees in historical footage and connect those observations with handwritten notes from the field.
This breakthrough application of generative AI is why I'm proud to announce that the Jane Goodall Institute has received the 2025 AWS Imagine Grant Pathfinder Award and an additional $1 million commitment from our Generative AI Innovation Fund. Meeting Dr. Goodall and contributing to her groundbreaking work was really a distant possibility when I started my career, like really, really distant. But everything you've seen and learned today aren't distant possibilities. They are current realities that are transforming missions right now.
Conclusion: Building the Future of Public Sector Innovation
When those builders started on the Hoover Dam, they faced seemingly impossible challenges, but they had a vision, and they built the infrastructure to make it real. Today, you face equally complex challenges: securing nations, healing the sick, protecting our planet, serving every citizen. And just like the builders of the past, you all, the builders of the present, need infrastructure that matches the scale of your ambition. That's what we've built.
GovCloud was a Hoover Dam moment, a foundational breakthrough that seemed impossible until it wasn't. AWS applies that kind of thinking to help customers all over the world solve their most complex challenges. We listen to your constraints, work backwards from your mission, and innovate on your behalf. And the breakthroughs keep coming: up to $50 billion in AI and high performance computing infrastructure, a sovereign cloud for Europe, the launch of AI factories. The future holds boundless opportunity for your organizations to make an impact, and AWS provides the most advanced, secure, and comprehensive infrastructure for shaping that future.
From the classified edge to the mission cloud, from foundation models to agentic AI, we innovate so that you can stay focused on your mission. Thank you for being here. Thank you for turning barriers into breakthroughs. Thank you for building on AWS.
; This article is entirely auto-generated using Amazon Bedrock.
Top comments (0)