Transcript: Governance and innovation aren’t competing forces: Together they drive AI success
AI governance can accelerate innovation and turn ambitious transformation into lasting value.
Discover how organisations can balance AI innovation with effective governance to unlock greater business value. PwC’s Leigh Bates and BT’s Tom Meakin explore why human oversight, change management and continuous learning are critical as agentic AI develops—and how businesses can improve today’s workflows while reimagining them for the future.
Transcript
Vik Iyer, CIO, Marketing Services:
Striking a balance between autonomous systems driven by agentic AI and the need for strong supervision is one of the key challenges facing global enterprises. In episode three of the podcast, we'll discuss how IT leaders can navigate this dilemma. We'll also look at driving innovation and getting the culture right to ensure AI delivers business value.
Joining me are Leigh Bates and Tom Meakin. Welcome to you both. I would like to start by asking: how are businesses balancing the need for innovation with the governance imperatives that are clearly critical?
Tom Meakin, Chief Strategy and Change Officer, BT:
Thanks, Vik.
Leigh Bates, UK AI Transformation Leader, PwC:
Well thanks Vik. Great to be here with Tom and yourself today. I think your point about innovation and speed versus risk and governance is actually a key theme I hear all the time. But for me, I don't really think organizations should view innovation and governance as competing forces. I really think when this is done well, governance should actually help you innovate faster. Right. And I'm sure we all recognize that we're moving from AI that gives us an answer based on a prompt.
To AI systems that can now plan, they can reason, they can make decisions and take actions on our behalf. So the opportunity with agentic AI is a big opportunity ahead of us. And this changes the governance question for me, because now we're no longer really concerned about is the model producing an accurate answer. It now becomes much more about, you know, what is this system allowed to do, under what conditions, what degree of autonomy?
When does a human need to intervene and what role do they actually play? Now, at one end, AI might recommend an action. A human approves it, and that's what we tend to call human in the loop. And as you move further along this spectrum, AI can then act within defined boundaries. So a human supervising and stepping in only when needed, and that's what we tend to call human on the loop. So I think for me, we need to be much more specific when we say we need
Human in the loop is a key governance control. Because the better question to ask is really what are we asking the human to do in this workflow and what role do we expect them to play and how do we continuously monitor AI and human performance together to get to the outcomes we can trust. Now clearly the answers will depend on the risk, the consequence, maybe the ability to reverse the decision as well. I taking a very simple example an agent scheduling a meeting is very different to one moving money around or
Interacting with a customer or making a material decision. And that's what I tend to think about AI governance as this point around trust by design. This means thinking about autonomy, accountability, controls, human oversight, as you design the system rather than trying to figure this out at the end. So you really have to think about governing the whole system rather than the model and the and the use case itself. Now Tom, it'd be good to get your perspectives on this, right? In terms of how you see this in your organization and
Leigh:
Whether you agree with what I’ve just said.
Tom:
Thanks, Lee. And it's great to be with you and with Vik. I see it in exactly the same way you do. Innovation and governance aren't polar opposites from one another. They actually go hand in hand. For us in our business, we combine those two things in different ways depending on the use case that we're deploying AI in. For example, when we're thinking about using AI in our mobile networks, we have
Quite a lot of governance. It also goes hand in hand with change management. So for example, AI is helping us load balance across our mobile network to ensure that we are responding to changes in customer demand. You know, as there's a peak of traffic in some locations in city centers during the day, you know, we configure the network so that we're always giving our customers the best possible speeds. That configuration increasingly is being recommended and done by autonomous agents.
It's something we've been experimenting with for near enough a year. And as we've got more and more and more confident, we've been able to move humans further back in the process, still keeping them in the loop, given the criticality of this use case for our business performance, but getting clearer on the capabilities that the AI agents can offer and building trust in them.
And also trust in our mobile network engineers who take pride in ensuring that the network delivers day in day out for customers. That's pretty different from use cases in other parts of our business, say in our middle office functions like customer research. We do a huge amount of customer research to ensure that we're understanding the needs of consumers and of businesses large and small. AI has allowed us to move from traditional research based on
Fielding surveys to building synthetic audiences based on our own customers' data and then querying those with much greater frequency and in real time. This increases the speed to insight and allows us to do a better job of serving our customers as their needs change. This is obviously still an important use case, but where there is a very different risk profile than in optimizing the mobile network.
Tom:
And in that area, therefore, we're actually giving much greater freedom to our research analysts to build agents, to test them, and then to roll them out. We, broadly speaking, take governance off in the early stages, and we allow our colleagues to build agents within broad guardrails. As soon as they want to roll them out, however, beyond 50 users,
We go through a quick validation approach that allows us to ensure that those agents are working in the way that we need them to work. If they do, and they typically do, then we start rolling them out more broadly across the organization. So two very different use cases, both of which demonstrate to our mind how we are trying to balance both innovation and governance at the same time to deliver the right outcomes for our customers.
Leigh:
I think that's great, great Tom. And I think and great example of proportionate risk actually. Because I think what you're saying is you know, you need proportionate controls to meet the objective of the use case and two great examples about how you're doing that. I think the other point that you make, which I can really resonate with, is actually getting the innovation speed right at the start, the experimentation phase. And then as you start to think about scaling this, you need to ensure you've got the right controls in place that
Allow those agents to operate within the boundaries that you expect it to with the right testing approaches in place. And obviously then you have continuous monitoring of how that's performing in BAU. So this thing doesn't really stop. You continue to kind of monitor and look at the way that these agents are performing in the future and you know you might iterate those controls going forwards as you learn from the experience of it working in production.
Vik:
Yeah, really fascinating to hear about, you know, the sort of governments in practice there, Tom. Thank you very much for that. And one thing I do wonder though is are businesses actually aiming high enough with AI? You know, are we spending too much time making today's processes more efficient when the real opportunity is actually to redesign how the business works? I mean, what do you think about that?
Leigh:
I mean, a great point, Vik. I a very timely point you're making here. I mean Tom just given two great examples about what he's doing in his business and I'm sure there are many others. But you know, look we would all recognise there is a lot of debate right now about the ROI from AI investments, particularly the scale of investment organizations are making and the cost projections actually as you ramp up AI deployments. We know AI agents come out a cost as well.
And I think this is one of the biggest differences for me between organizations that view AI as an efficiency and a productivity play versus those that are genuinely transforming their business with it. Right. Now, a lot of organizations started by saying, well, how can AI help us do what we do already, but do it faster, do it better, do it cheaper, do it with less people, make our process more efficient. And there's clearly value in that. But I do think the bigger opportunity to ask here is.
You know, if we redesign this end-to-end process, this end-to-end customer journey, let's say, knowing what AI can do today, would we design it in the same way? And I think the answer would probably be no. So I see this opportunity really at three distinct levels, right? You can automate a task, you can augment a person with AI, so the co-pilot idea, and you can reinvent how that outcome is delivered all together through process reinvention.
No, I think a lot of the activity today is focused on the first two, but they're an important and a bigger opportunity with a gentic AI is on that third point around reinvention. But it's not the easiest to execute on. I'd be great to get your view on this, Tom, as well, because the ambition is I'm sure many firms would say that's their ambition, but actually to execute on that becomes quite a difficult task.
Tom:
I think that's exactly right, and I love the three ways you've described it, augmenting the process or task, augmenting the person and then reinvention. Our experience is consistent with what you've described, which is people spend most of their time on one and two and not enough time on three. That's understandable. If you're an operator and you're running as we do, know, 20,000 guides in call centers and retail stores up and down the country, you've got a business to run.
So it can be really difficult to figure out how do you reinvent the process while ensuring that you continue to run the business. We respond to that in a really specific way, which is that we have red teams and blue teams. So on big business processes where we know that AI has lots of potential for reinvention, we stand up a blue team that works
On reinventing that process, well, the red team augments it, to use your phrase, makes the incremental improvements in the existing process as it runs today. The blue team are interdisciplinary. So you'll have process engineers, you'll have customer experience and user experience designers, you'll have software engineers, and you'll have people that work with the hyperscalers and
Other partners that we use to deploy AI across our company. And they start from a blank sheet of paper. We're deploying this in our B2C customer service process, as well as in some parts of our software engineering organization on the software development lifecycle. And the results are pretty profound, as you might expect, depending on whether you're working on
Greenfield or Brownfield architecture in software development, for example, we're seeing anywhere between 10 % and 60 % uplift in coding productivity. But in some ways, that's the really easy bit. The more complicated bit is the change management. And this blue team concept means that you can prove that it works in a real environment. You can then begin scaling it.
Tom:
And over time, you can begin to switch more and more and more of the business process from how it's being run today to how you're reimagining it for the future. That is a complicated journey to go on and we're right at the beginning of it, but we're seeing some encouraging early signs.
Leigh:
Yes, that’s great. I love the red-team, blue-team approach, Tom.
I think we all need to think more about. I think the just building on one point you made, and this is kind of relates back to the Blue Team approach. I think the other thing that for me is critical in this is the point around continuous learning loops. And what I mean by that is that as you start to understand you know how AI is operating across the kind of blue team approach of reinvention, we need to understand what's happening, where humans are intervening.
Where the agents escalate, what level of autonomy was right. And then we use that evidence to continuously improve the AI system and the process around it. So it's not a once and done exercise. We're continuously kind of gaining the signals from what's working, what's not, and we continue to think about applying the right guardrails and controls to ensure we get to the right outcome. And that's I think that's an important point that we need to think about as well.
Tom:
I think that's right.
Vik:
Yes, it’s going to be fascinating to see how that emerges and evolves over time, with these processes being genuinely reimagined rather than simply changed. It will be interesting to see how that plays out, because I’m sure it will affect end users and consumers as well. But Tom, more and more companies are starting to see that AI is only successful if employees are fully bought into the transformation. I imagine you’ve gone through that experience as well. What should IT leaders be doing?
In your opinion?
Tom:
Look, let me talk about IT leaders and then let me talk about leaders more broadly. So I think the most important thing that I ask our IT leaders to do is to solve customer or business problems, not solve a technology problem. And so, we don't think about our AI transformation as being driven by the IT or in our words, the digital unit. It is a whole business transformation.
Where the digital teams and our consumer or business-facing teams work in partnership to identify the big customer problems and business processes that AI can help solve and then build an ambitious target state of where they want to take that process and a plan to get there. That ensures that you are creating real ownership
And amongst ultimately the people who are benefiting from this. It also ensures that you are looking at the full suite of changes you need to make, which in many cases go beyond simply deploying an AI tool. They cover things like data cleansing and data governance, as well as things that are fundamentally unrelated to technology, like how you performance manage people or the kind of skills that you hire from.
A real example, we're increasingly using AI tools to help us in the creative production process. What this allows us to do is to serve more personalized communication, adverts, customer updates, to more segments of customers at a lower cost and faster than would have been possible in a non-AI world. In order to be able to do that, yes, we...
Deploy a small number of AI tools. But actually most of the impact is in reinventing a marketing process, changing the way that we work with our media agencies, changing the interface between the creative production teams and the channels in our consumer and our business units. So actually nothing to do with technology. So the first bit of advice I give or the first ask I make
Tom:
To our IT leaders is solve business problems not technology problems. I think then there's a whole suite of other things that leaders need to bear in mind As they're embarking on these kind of transformations across the whole company. Yep first The narrative and the change management is really important. Yep. This is a Technology that is evolving incredibly fast that is creating
A relatively high degree of nervousness amongst colleagues who are, know, some of whom don't understand, some of whom are excited, but have a capability gap, and some of whom are sort of fearful for the impact it might have on their jobs. And this is particularly acute for us. We employ 100,000 people up and down the UK. And so we spend an awful lot of time working with our frontline teams to ensure that the
Stories we're telling around the impact of this technology are constructive and encourage our people to embrace it rather than feed fears. Second, this is about capability building as much as it is about technology deployment or business transformation. Depending on the use case, getting the most out of this technology is going to require a complete rethink.
To the kind of capabilities we need and that we hire through our strategic workforce plan and how our teams are structured. In the future, if you're a software engineer at BT or a customer service leader, your teams are likely going to be a mix of humans and agents. That's hugely exciting for people managers because it's stretching them into new and interesting spaces, but it also requires us to give them and them to learn.
A totally different set of skills. Third, you need to think differently about how you organize. And that has been perhaps one of the most complicated things that we've started to tackle. We have moved to a hub and spoke model for AI deployment across the company, where we've got an AI hub in our IT organization that is responsible for AI architecture
Tom:
For governance and approvals, for token ops, so that we hedge the risk of price inflation, as well as manage usage of some of these tools. And then we have spokes in each of our major functions and business units, where we put deployment teams, whose job it is to help us understand the business process, reinvent it with the people running it, as well as think about incentivization capability building and some of the other enablers around.
So as a broader leadership group, we're really focused on the narrative, on in capabilities and ensuring that we've got the ones that we need in the future, and making some changes to how we organize ourselves to ensure that we can accelerate through this transformation and really take advantage of the benefits that AI has to offer.
Leigh:
Yeah, look, I think all too often we talk about AI as a kind of technology transformation and for me, as you've said Tom, it's people that's at the center of this, right? If we don't get the kind of people and culture aspect of this right, then no matter how good the technology is, we it won't be successful, right? And we should learn from the past around technology transformation programs in the past that have been very much technology driven without involving the business. So I think
Maybe just to build on your points, there's three things for me that kind of really kind of stood out. One is the point around you need to involve the right people in redesigning a workflow, right? The people closest to a process understand that business process any w versus anyone else. And they understand where the friction is, they understand where human judgment really matters, and they probably understand where AI could have greater autonomy and could genuinely help with an outcome, rather than the technology teams that may be designing
The workflow from a from a tech perspective. The second point is skills and not agent skills because we know there's a lot of talk about agentic skills and building markdown files, etc. But this is much more about people skills actually. And what I mean by that is I think people n increasingly need to know how to work with and supervise AI. So you know when should I trust it? When should I challenge? How do I know when something doesn't look right? When do I need to step in?
And I think that's an important point around actually just think about not just accepting the output of AI systems, really giving the right challenge to make sure we're getting the right answers and the right accuracy. And then this final point I mentioned earlier about the learning loop, and this is the kind of link between people and technology. So, you know, every time a human intervenes, it tells you something around the process. Every time, you know, an individual overrides an agent decision.
Need to understand why. Maybe the controls need to change. Maybe the agency constructions need to change. Maybe the decision should main remain with a person rather than agent. So I think all of these signals we need to understand and we need to learn from it. And then we can really think about ways that we can put the right workflow and governance around how we're applying AI into our into our business using all of those things. So I think very much people at the centre of this. So yeah I would definitely agree with everything you've said there as well.
Vik:
Yes, it’s been a fascinating conversation. There is so much change coming into business over the next few years. What key final thoughts would you share as we come to the end?
Tom:
I guess my kind of closing thought, is this isn't a technology challenge or a technology opportunity. It's a people one. And throughout this conversation, what struck me from the things that Lee and I have been talking about is actually most of the dialogue has been about capabilities, role modelling, operating model, change management, narrative. Those are...
Those are enduring challenges for all business leaders, whether you're a CIO or someone else in the company. And that's where we as BT continue to focus. And as we navigate through the next couple of years, where I'm sure the pace of technology change will continue to speed up, that will remain at the forefront of how we use
AI to better support our customers and allow our people to do their best work day in day out.
Leigh:
Yeah, I think look, it's been a great conversation for me. I think this next phase of AI is going to force us really to think differently about transformation, governance, trust, workforce. And we've talked already about moving from AI that advises people towards AI that can act on our behalf through agents. And that means that one of the defining questions for us becomes how much autonomy
Are we prepared to give AI right? And I don't think organizations that succeed will necessarily be the ones that race fastest towards full autonomy. I think they'll be the ones that learn how to increase autonomy, how to increase agentic workflows, but do it responsibly and safely. So they set clear banded activities, set clear guardrails, set clear controls, keep humans involved where judgment really matters, monitor what happens, learn from those interventions and exceptions, and build
Build confidence. And I think that's really critical for me. And maybe just a kind of final point I've been thinking more about, which is I think if we can bring kind of four things together, and they are the first is ambition, and we started this conversation around that. So thinking really about the big opportunities of what business outcomes we can achieve with AI. So having a really clear ambition around that, not at a use case level, at a kind of reinvention level is the first point.
The second is autonomy. How much autonomy are we allowed? Are we prepared to give AI and what constraints and what risk tolerances do we want to set? Accountability is important. This is the kind of people point about ensuring we have clarity on who owns the outcome. And then the last point is around agility. We need to keep learning and improving. As the technology changes, it's not stopping, it's continuing to evolve. And then we think of need to think about how the organization evolves as well.
And those two things need to move in sync to get to greater scale. So I think this governance point really helps us think about moving with confidence. And if we get that right, along those four things I just talked about, I think we can achieve some really great things in the in the next coming months and years going forwards about how this technology will evolve.
Vik:
Yes, it’s a fascinating time, and that brings us to the end of episode three of this podcast, exploring the interface between trust and AI-driven growth. My thanks to Leigh and Tom for coming on the show. Please make sure you catch up with our other episodes as we explore this critical issue. Bye for now.
