Building Responsible AI for Public Services With AWS

[00:00:00] Your agents aren't producing accurate answers because they don't have a complete semantic understanding of your data. And Denodo is solving this and solving it through semantic consistency. Through semantic consistency, your agents can start making accurate predictions in real time. So see what else Denodo can do by visiting denodo.com to learn more. But now let me introduce you to today's guest.

[00:00:32] How can public services move quickly with AI when every decision might affect a student, a patient or indeed a citizen? Well, in today's episode of Tech Talks Daily, I'm quite excited to be speaking with Holly Ellis. She's the AWS Director for UK, International Organisations and German Public Sector Technology.

[00:00:58] But today we're going to talk about responsible adoption across government, healthcare, education and non-profit organisations. And she will explain why speed and safety can co-exist. And expand on how leaders should be judging the consequences of every AI use case and where human review still belongs.

[00:01:23] And we'll also hear how universities handled clearing demand with cloud contact centres. How the Ministry of Justice used AI to modernise a 15-year-old application. And why skills, governance and digital sovereignty. How all these things can now shape every major technology decision.

[00:01:44] So if your organisation is trying to turn an AI experiment into a dependable service, I think you've come to the right place. But enough from me. Let me introduce you to Holly right now. So thank you for joining me on the show today. Can you tell everyone listening a little about who you are and what you do? Of course, it's great to be here with you today, Neil. I lead the UK, German and international organisations Solution Architects for Public Sector team.

[00:02:14] This team works with customers in education, not-for-profit, local and central government and healthcare. And our role is to support public sector organisations to leverage cloud and AI technology to enhance system services, drive operational efficiencies and lower costs. And the Solution Architects, they work back from a customer's mission and help them to architect solutions using AWS services and our partner solutions. I've worked at Amazon for almost seven years. And previous to joining, I worked in the public sector, both in local and central government.

[00:02:44] Awesome. Well, thank you so much for sitting down with me. There's a lot I want to talk with you about today, especially I always, before a guest comes on, I always do a little research and try and find out a little more about their origin story. And Holly, you've experienced public sector transformation from both sides, having worked inside government and now with AWS. So I've got to ask, what do technology companies often misunderstand about how governments actually adopt technology?

[00:03:10] And what do governments sometimes maybe underestimate about the pace at which technology is changing? It feels like you've got quite a unique vantage point here. Yeah, of course. And I don't think it's necessarily a misunderstanding as such, but there are definitely considerations for organisations like AWS when working with public sector. So we hire builders who are both technically deep and have a good understanding of the industry they're supporting.

[00:03:34] So, for example, the solution architects across the team I lead have a depth of understanding of healthcare, education, all the services central government departments are delivering. And we have a wonderful mantra that we use, which is we meet our customers where they are. So this comes with an understanding from us that our customers are managing a whole system to bring about change and adopt technology. And that system could include complex organisation structures, governance, procurement, process, regulatory requirements, culture, and of course, people.

[00:04:04] And specifically having the skills required to build and adopt technology. The Unlocking the UK's AI Potential with Strand Partners report that we published in April this year brings the skills challenge into sharp focus. So half of UK public sector organisations cite AI and digital skills shortages as the main challenge for AI adoption, an increase from 46% in the prior year. Now, this aligns to the increase in AI adoption from 52% to 64%.

[00:04:31] The more we adopt, the more skills we need, the more depth of AI services, the deeper the skills we need. And it's both a challenge in terms of volume of skilled people that are needed and the growing depth of knowledge. And at AWS, we're committed to creating the conditions for UK organisations to succeed through our investment in skills training and support programmes. AWS has launched a number of learning skills programmes in the UK to help address the digital skills gap and enable organisations to take full advantage of cloud.

[00:04:59] An example of this is the Skills to Jobs Tech Alliance. So I'm going to launch the Skills to Tech Job Alliance in the UK in April 2025 to help 100,000 people gain AI skills by 2030. To the second part of your question, what does government sometimes underestimate about the pace at which technology is changing? I don't think that the pace required is underestimated by government.

[00:05:23] But putting that pace into practice is a challenging shift, particularly as things are moving at such an unprecedented rate. The decision making processes and governance that may have worked in the past are less likely to align to the pace we're seeing today. For example, a decision made just a short time ago could be answered differently today. And this is why flexibility is required across mechanisms, things like governance processes and enterprise architectures in place.

[00:05:49] In my nearly 28 years working in the technology industry, I've not observed the need to be so agile in decision making and governance structures. We do see some excellent examples of adoption at pace. And this year's university clearing process is a really good example. So during this year's A-level clearing, several universities, including University of Manchester and the Teesside University, ran their contact centres on AWS using Amazon Connect, working with our partners like Sirius HQ,

[00:06:17] to go from first engagement to live in under three months. The platform scaled seamlessly on results day. One institution handled up to three times the core volume of its legacy system, equating to a four-figure number of student places confirmed in a single day. And beyond voice, universities offered WhatsApp as an additional channel so students could reach them however they preferred, while marketing teams used built-in AI features to optimise ad placements in real time and drive more enquiries.

[00:06:45] So the result, every student call answered, more enquiries converted into confirmed offers, and universities confident their technology could keep pace with one of the most stressful and high-stakes moments in a person's life. And I love how you've led here with a measurable difference, return on investment. And those figures take priority over the technology itself. And it's so refreshing to hear that approach.

[00:07:09] And we do often hear constantly about governments needing to move from AI pilots into production and getting that measurable value. But public services can't necessarily adopt the Silicon Valley mantra of moving fast and breaking things. So what does responsible speed look like when AI decisions could affect healthcare, benefits, education, even national security or somebody's interaction with the state?

[00:07:37] It's a different way of approaching things, isn't it, I would imagine? Yeah, and speed and safety aren't mutually exclusive. And we offer three dimensions, I think, useful considerations. So there's the technology choice. Will you be architecting an entirely new service or solution? Is there an AWS partner solution that will meet your needs? Will you choose a managed service or not? These decisions will impact your pace. The second thing is thinking about capacity and the skills you have to deliver. Can you deliver this within existing teams, using existing skills?

[00:08:07] Or would bringing in a partner help you to realise the benefits more quickly? And last but not least, not being afraid to experiment appropriately. So I'll offer you another mantra. We think in terms of thinking big, starting small and scaling fast. And this way of working makes sure you're focused on the big mission outcome, but starting in such a way that the impact of failure is contained and safe. And once you've got confidence, then you can scale. And I think the clearing example works well here,

[00:08:33] not least because it impacts people's lives and is a particularly stressful time for young people. The universities we work with chose Amazon Connect, our cloud-based contact centre service, which includes AI out of the box, including AI agents. The functionality it provides in working with a partner enable the universities to bring about major change in a short period of time. At no point within this is safety compromised. And that is a mantra I think we can all get behind, a much better mantra than moving fast and breaking things.

[00:09:03] And when I was doing a little research on you, I was also reading that you've said that as AI becomes mission critical, security becomes both a leadership responsibility and an engineering discipline. So what changes when AI moves from just helping an employee draft something like an email or a calendar invite to, as to becoming embedded inside an operational public service? And assuming when you get to that place, what kind of safeguards or guardrails need to exist before that happens?

[00:09:32] Yeah, and I think this is really important because both drafting something and an operational service can have a consequence. So safeguards need to be in place for both to varying degrees, depending on what those consequences are. So both could require human in the loop, for example. A genetic AI may require observability metrics, system level guardrails, for example, least privilege agent scoping. What both leaders and engineers need to do is understand and evaluate the consequences

[00:09:59] and make safeguarding decisions based on that consequence. So leaders have a role to play in defining the safeguards across the whole organisation, as engineers have a responsibility for safeguarding what they're building. And the Ministry of Justice is a good example. So they manage over 350 prisons, they employ over 100,000 staff, and use AWS Transform to modernise a 15-year-old .NET legacy, but mission-critical application that powers the Ministry of Justice's response to parliamentary and public correspondence.

[00:10:29] The first AWS Transform pass processed 24,000 lines of code in nine hours. A second pass took two hours and improved code quality. Human in the Loop review was the quality gate throughout, taking approximately 20 hours of developer time versus an estimate, an estimate of nine months of manual modernisation. So there is still an investment of time. And in that particular scenario, human in the loop guardrails, with the result being a huge time-saving overall.

[00:11:00] And I've been fortunate to attend 15 different tech conferences this year, from Egypt to Vegas. And predictably, this year is all around agents, agentic AI. It's quite telling that it wasn't really a focus last year, but it's exploded this year. And as a result, there's this enormous excitement that we're seeing around all things AI agents, especially on taking on increasingly complex or repetitive work.

[00:11:25] So where do you see agents genuinely helping public sector employees today? And how should departments decide which decisions should be dedicated to AI, and which should require that human approval, and which should remain entirely human? It's like three different parts there. But tell me a little more about that and the best way of going around that. Yeah, of course. If we just take the Ministry of Justice example, and then look across the civil service, as of February 2024,

[00:11:53] there were 28,000 digital and data specialists working across the UK. And AWS Transform is an agentic AI migration modernisation service. So using Transform has assisted engineers modernise an application more quickly, in turn allowing them to turn their attention to other priorities. So you can see there the role of agentic AI in really helping public sector employees across a broader scale than the example that I've offered.

[00:12:19] And the Amazon Connect example that we spoke about in relation to clearing services, it facilitates AI-assisted conversations and uses AI agents to complete actions during customer interaction. So this removes a lot of manual work, saving public sector employees' time. Deciding the degree of humour and interaction in both examples comes back to my point around understanding consequences, and of course, building confidence in the service that you've built. And last December, I was fortunate to go to the AWS re-invent,

[00:12:48] and myself and all tech reporters were driven to the middle of nowhere, and we blew up some old servers to get over that message that tech debt needs to be removed as quickly as possible. And of course, when we talk about public sector organisations, in particular, they often have decades of legacy technology, fragmented data, procurement constraints, and understandably cautious governance. So is AI creating an opportunity to modernise

[00:13:16] some of those underlying infrastructure stories that we're talking about here, or is there a danger of putting sophisticated AI on top of problems that governments haven't got around to fixing yet? What are you coming across here? Yeah, so the AWS Transform example I gave for Ministry of Justice is a good example of AI creating the opportunity to modernise, but this is also a question of balance. So it is the case, and was before this moment of rapid AI adoption, that the services that you can deliver

[00:13:45] and the quality of the outcomes of those services is highly dependent upon the underlying infrastructure and the quality of your data. But getting this right doesn't have to stop you from benefiting from AI today. So actually, both the examples I gave for clearing and application modernisation are good examples of achieving benefits ahead of transforming a whole organisation. HMRC is a good example of an organisation that's migrating from legacy to cloud, providing them with foundations

[00:14:13] from which they can build AI-driven services and benefit from the flexibility of the cloud. And it was really great to see the UK government publish their challenge book in July of this year. One of the challenges addresses this point specifically. So challenge one, zero legacy public sector. 28% of governments of state is legacy, and the challenge book sets out to address this, specifically calling out the need to accelerate modernisation of higher complex and higher risk systems. But I think the message is it's a question of balance. There are benefits that could be received today,

[00:14:42] whilst also addressing the underlying infrastructure and the data that is required to deliver services that AI will work incredibly well on for the future. And the UK conversation, I'll also include Europe in this, around cloud and AI. I've noticed that it increasingly includes things like sovereignty, resilience and dependence on large technology providers. So from your conversations that you're having with public sector leaders, is what does digital sovereignty,

[00:15:11] what does that really mean in practical terms? And how can governments maybe better balance sovereignty with access to the best technology and global innovation at the same time? Yeah, well, from my experience working with public sector customers across multiple countries, sovereignty means different things to different organisations. And in fact, even within a single organisation, the requirements can vary workload by workload. I say digital sovereignty in practical terms is about control, transparency and optionality. And government can absolutely have all three

[00:15:41] while accessing the best global innovation. At a practical level, our services have been sovereign by design since day one. Customers have full control over where their data is located, how it is stored, how it's transferred and how it's encrypted using keys that we cannot access. That level of control is why organisations handling the most sensitive national workloads choose to build on AWS. And one thing that I really hope comes through at the AWS Public Sector AI Symposium this year

[00:16:10] is what measurable success actually looks like rather than just another shiny demo. And you've already given me so many great examples around this, which kind of tells me that is what you're aiming at too. But beyond announcing another AI initiative, where are you seeing cloud and AI? Where are you seeing it make a tangible difference to all citizens, frontline workers, costs, or the speed and quality of public services? What are you seeing here? Yeah, and measuring success can be challenging. Customers often ask me about how to measure success

[00:16:39] when introducing AI services. And it can be challenging for several reasons because AI use cases are so varied and success will be different. For example, the university clearing, example, improve services to young people. Like that's quite hard to measure in terms of the impact a better experience has had on that young person on that particular day. And it's harder to measure than other metrics, such as the cost savings as a result of migrating to Amazon Connect, the cloud-based contact center. The MOJ example has a measurable time saving

[00:17:07] for engineers, so from 20 hours to 20 hours from a potential of nine months. And at the same time, there's a downstream impact. Engineers are able to more quickly focus on the next modernization or other activities. There's a cumulative benefit, which can be hard to measure. So when I talk to customers, particularly on AI initiatives, I guide them to really spend time identifying the outcomes they will measure up front. And it helps to prioritize those initiatives and to understand if the expected outcomes are actually being achieved and then building upon those initiatives

[00:17:37] have demonstrated the greatest returns. What we do know from the Unlocking the UK's AI Potential Report with Strand Partners is that when organizations use AI to truly redesign workflows, accelerate decision-making and build entirely new products and services, they report average efficiency gains of 68% compared with just 40% among basic users. And from the outside looking in, one of the big things that stands out this year is how you've brought leaders from AWS, OpenAI, Anthropic,

[00:18:07] the UK AI Security Institute and Tony Blair Institute all coming together in London. So what is the conversation you think that government leaders need to be having about AI right now that maybe isn't receiving enough attention? And what is that question that every public sector tech leader that might be listening to our conversation today, what do you hope that they will take back to their organizations? Yeah, I truly think government leaders are doing an amazing job of asking the right questions

[00:18:37] and enabling their organizations to safely adopt AI. What I would reiterate is the need for agility and that's agility in decision-making. And like I said, you know, as we were talking a little earlier, you know, the decision you make, you know, just even a couple of months ago, you might make a different decision today. And agility in the governance, for example, the pace of change we're seeing today requires a high degree of agility beyond what we've experienced previously. And so that's the question I think government leaders

[00:19:06] should be really focused on is how agile is our organization in our decision-making, our governance, and enable them to receive the benefits of the technology that's being offered today. And I think that is a powerful and thought-provoking moment to end on. But before I let you go, there are going to be a lot of people listening here, a lot of people attending the event, a lot of people won't be able to attend. So if anyone wanting to find out more information about some of the conversations that are happening there, some of the coverage that's coming out, where would you like me to point everyone?

[00:19:35] We hope you might join us at the Public Sector AI Symposium on 9th of September in London. So please do sign up through the webpage. For the latest insight and developments about AWS's work with public sector, we'd encourage you to visit our Public Sector blog. And for latest Amazon news, you can visit Amazon News Press Centre. Awesome. I will add links to everything that you mention there. If I go over to techtalksnetwork.com, there'll be a blog post associated with this episode. And you'll also find a useful link section with everything

[00:20:04] that we've talked about today. So I urge people to check that out and also feedback to me. Let me know what you're seeing, what's working, what isn't, and your experiences. But more than anything, Holly, thank you for bringing all this to life today. Really appreciate it. It was great to talk to you. Thank you, Neil. I think Ollie's advice today is to think big, but start small and scale fast. Public services cannot wait for every legacy system to disappear before using AI. But that said,

[00:20:34] each project needs a clear outcome, appropriate safeguards, and an honest view of its consequences. And the examples from University Clearing and the Ministry of Justice all show how progress can be measured in things like answered calls, confirmed places, and developer time that was returned to other work. And these things also show why the human role remains part of the design. So a big thank you

[00:21:04] to Holly for explaining how AWS works with public sector teams and why flexible governance matters more than ever as the pace of technological change continues to ramp up. And remember, you can find further information through the AWS public sector blog, Amazon News Press Center. But over to you, how is your organisation balancing responsible AI adoption? Especially with that pressure to deliver better services

[00:21:32] sooner rather than later. Well, I encourage all of you to go over to techtalksnetwork.com. Over there, you'll find eight podcasts that I host. There's over 4,000 interviews. That is an event calendar where maybe you can meet me for a hot coffee or a cold beer on the show floor at an event near you. You can work with me, send me an audio message and most importantly for today's episode, look for the blog post that will accompany this episode

[00:22:01] and you'll find links to everything we talked about and maybe a few Easter eggs in there for you as well. So techtalksnetwork.com. But that is it for today. I'll be back again tomorrow with another guest and I look forward to getting to speak with you again now. Thanks for listening. Bye for now. Bye. Bye. Thank you.