00:40 – Origins of the survey: understanding how ASX boards are actually governing AI, beyond the hype
02:03 – Overall sentiment: comfort with AI varies widely between “leader” and “laggard” boards
03:56 – AI is mostly framed as a cost/efficiency play, not yet a revenue driver; ROI still hard to pin down
05:35 – Red flag #1: misalignment between CEO and board on AI risk and opportunity
08:22 – Red flag #2: disagreement on workforce impact, discussed against the backdrop of CBA’s AI-related redundancies
09:52 – “AI washing” — redundancies mislabelled as AI-driven; most surveyed companies are actually still hiring AI talent
12:14 – The J-curve: near-term costs (e.g., running parallel legacy/AI systems) before efficiency gains show up
15:19 – Red flag #3: scattergun AI project lists (e.g., “300 use cases”) signal a lack of strategic focus
16:19 – How boards are handling AI risk oversight, and the rise of “shadow AI” / BYOAI
19:12 – Finding the balance between enabling experimentation and controlling shadow AI use
21:18 – Why “human in the loop” is a weaker safeguard than it sounds
23:27 – Cognitive offloading and skill decline — an airline-industry/autopilot parallel
27:27 – “Super users” and the risk of AI widening productivity gaps within workforces
31:39 – Looking ahead: token cost management as a board-level issue
33:15 – The missing piece: few boards have defined guardrails, no-go zones, or a “kill switch” for AI
35:01 – A recent AI incident is raised as a case for why kill-switch thinking matters
36:17 – The regulatory divide between the US and Europe, and what it means for multinational AI rollouts
37:16 – Existential disruption risk: Fidelity’s 11-factor framework, and the Kodak vs. Fujifilm lesson for adapting vs. defending
40:23 – What’s next for the survey: cost metrics, guardrails, and expanding coverage across the ASX 200
43:18 – Closing: Sam and Sue Lyn on how they personally use AI
Full Transcript of Episode 142
Wouter Klijn 00:00
Welcome to the [i3] podcast. I’m here today with Sam Heithersay, Portfolio Manager Australian Equities, and Sue Lyn Stubbs, Associate Director Sustainable Investing of Fidelity International. And today we’re going to talk about artificial intelligence and disruption at listed companies in Australia. Sue Lyn, Sam, welcome to the show.
Sam Heithersay 00:21
Thank you for having us.
Sue Lyn Stubbs 00:22
Thanks for having us.
Wouter Klijn 00:24
Fidelity International conducted a survey of 31 ASX-listed companies during the 2025 AGM season in Australia. Can you tell me a little bit about the survey? What was the catalyst? What started it? And what were you trying to find out?
Sam Heithersay 00:40
Well, the main reason for the survey was to understand how boards were handling AI in practice. Through reporting season last year, we heard a lot of rhetoric on the broad thematic of AI and the opportunities that it presented, but we didn’t hear much on the risks, the governance oversight it demanded, or how companies were thinking about the workforce impact, so we used our access to company management, to boards, to conduct a more bottom-up systematic survey to really tell whether the strategy was aligned with governance. And we conducted the survey last year, but as the AI sell-off really gathered pace, it threw up a lot of great opportunities for investing, and so we combined it with an AI framework that our analyst team did from the bottom up, assessing each company and its risk of AI disruption, and we ended up using this governance analysis as a bit of a leading indicator of the AI maturity of a company, and gave us a lot of confidence when a company was thinking more deeply about these issues, that they’re more likely to adapt their business model, not just defend against AI disruption, and we think ultimately that might make for a better, longer-term investment. So, I think the use case evolved from just trying to cut through to noise to to setting a baseline for how we can track the AI maturities of the of the companies we invest in.
Wouter Klijn 02:03
Yeah, and so before we delve into sort of the details of the survey and the answers, at sort of a very high level, did you get a sense that boards and companies were comfortable with the topic, or did you get a sense of uneasiness about not knowing quite where this all is going?
Sue Lyn Stubbs 02:18
So across the survey, we really asked a broad spectrum of companies, both from small cap to large cap, across multiple sectors, financial resources, healthcare. We didn’t want to just focus on technology because we felt like a lot of the rhetoric and a lot of the focus was in you know the SaaS space in technology, and so we wanted to really understand across the broader Australian economy what was happening? How were boards thinking about this? How were management teams thinking about this? So, I would say it depended. Some boards were very comfortable speaking to it, and you can see that in kind of our leaders who are really driving, you know, transformation and adoption of AI, versus quite a few of our companies were in that really early stage. Having said that, most boards were comfortable speaking to it to some degree, but really, when we were pressing more around questions of you know strategic value creation or risk and controls, that’s when we saw some of the answers start to be more conservative, maybe starting to show signs of laggards or are not at that maturity level. So I think most boards were comfortable, but I think you could easily see the bifurcation of where we saw some who were very comfortable, who are leading on AI implementation, and and who were coming up the curve, and I think that really showcases how our access to companies and boards allows us to kind of really see that that that differentiation: who’s really leading the charge, who’s mature on AI implementation, and who is is coming up that curve.
Wouter Klijn 03:41
Yeah, I mean, there’s many different angles to AI, but I think the survey still found that AI is primarily framed as sort of an efficiency lever. Were there any conversations that were more looking at innovation and the other part?
Sue Lyn Stubbs 03:56
It was very early days. Probably around 20 per cent of our companies started talking about revenue generation, strategic value creation, customer kind of developments, and what it meant for, you know, really driving revenue. Having said that, though, across the board, it’s really in that productivity kind of cost perspective, and you know what we’ve been really focused on is trying to uncover what is the ROI, what is the investment case, how are they thinking about capex of AI and capital allocation to AI, and unfortunately, it was challenging to kind of get a clear answer from our boards. That was an area where it was a little bit more opaque. Some sort of reference that it’s in our BAU, particularly the tech sector, the the ones that we’re leading. They were saying it’s in our classic R&D and our BAU capital allocation strategies, but that was an area I think we didn’t get great answers on. I think going forward we will get better answers. Definitely looking to the reporting season and seeing how that answer matures through time.
Sam Heithersay 04:51
Yeah, definitely. I think we’d expect the conversation to move away from just pilot studies and broader kind of opera. Operationalization to more of a transformation, more adoption, and ideally more upside cases than just cost out, where they might where a company might be able to change the the value proposition for a customer, but the survey itself was still very early in in the maturity of a lot of these companies and the way they were thinking about AI, so we didn’t necessarily expect a lot, and we’re grateful that we now have this baseline to be able to measure companies as they adapt and evolve their AI proposition. Because you’re right, it was much more about cost out and less so about the revenue generation opportunities.
Wouter Klijn 05:35
And it’s no doubt that AI is quite a revolutionary technology. There’s of course lots of opportunities with it, but what I found quite interesting as well in the survey is that you mentioned a number of red flags, and I was sort of testing this with a couple of the funds out here where they thought of it, and one of them stood out in particular where you sort of identify that the position of the CEO is different than the board, or they have a different view on AI that could potentially lead to some trouble. Can you tell me about that?
Sam Heithersay 06:06
Yes, there were definitely some instances where we interviewed the CEO and the board, and they were misaligned in the way that they were discussing AI, not just from the opportunities, but also from the risks that it might involve, and that was clearly a signal. I think of you know poorer governance around AI. It’s a great leading indicator for us in terms of AI maturity. Ideally, they would all be on the same page, and we did see that inconsistency internally. And for us, we were able to gauge that, having interviewed CEO and boards, gaining access to both, not just through our company one-on-one meetings, but also through the AGM season, where we tend to interact more with at the board level, and it allowed us to move just beyond that kind of polished AI narrative that a lot of companies have to really understanding that internal inconsistency within the company, and we think that’s ultimately going to play a part in how these companies can adapt to AI. If the board and management are different in the way they might assess cost or risk appetite, then it’s unlikely they’re going to be able to succeed rolling out a broader, more cohesive AI strategy. And that’s exactly what this survey was designed to try and pick up where the governance didn’t quite match with the strategy.
Sue Lyn Stubbs 07:24
Yeah, I mean we were seeing some CEOs uber bullish. Like we have 300 use cases, we have 100 page you know a 300 page deck of all our AI opportunities, revenue generation opportunities, and then you speak to the chair and they’re very conservative. Oh, we’re only rolling it out in these certain aspects, only doing these pilot studies, so that’s where we started to get concerned. One from a prioritisation point of view, whether it’s actually going, you know, the spend is actually going to the right products or projects or to the right products for value creation, and then I think the other thing was really around that risk mitigation piece. So it was both a strategic what capital they’re spending, and then how much oversight is there from the management team versus the board as well?
Wouter Klijn 08:07
Yeah, isn’t it the job of the CEO to be super bullish on everything?
Sue Lyn Stubbs 08:10
True, definitely. I think so. I think it was just how far you know across the spectrum, the CEO versus the chair. I think that’s what surprised us, kind of the reaction of AI and what the opportunities were for the for the business?
Wouter Klijn 08:22
Yeah, were there any areas that stood out more than others in terms of disagreements? I mean, disagreements can be small and large, but you know what you mentioned. If the CEO talks about hundreds of projects and the chair is not quite convinced that that’s all happening, that’s obviously a key concern. But were there any other sort of tensions that stood out?
Sue Lyn Stubbs 08:40
I think from our point of view was also around workforce disruption, and I think not saying that management teams were talking about you know massive layoffs or anything, but I think there was more of a cost consciousness from the management team, what it meant from kind of productivity, etc. But then when we were speaking to boards, they were more conservative. Only around 40 per cent or over 40 per cent reported minimal impact over the next 12 to 24 months from a workforce perspective. I do have to caveat this: this was when CBA came out with their 45 AI-related redundancies, and they got a lot of pushbacks from you know union associations and and that impact. And then they rehired those 45 workers. So, I think we do have to caveat the answers by saying that that was the environment we were asking the boards, and I do think some of their answers were more conservative. Having said that, we haven’t seen much impact in the workforce from a data perspective, at least from what we’re seeing in in workforce surveys. So, I think overall that was a little bit of a difference from that productivity cost savings versus a board talking about the workforce shape over the next 12 to 24 months. So that was another area where there was a bit of differentiation.
Wouter Klijn 09:52
Yeah, let’s go into that a little bit because I’m not sure if the survey was before or after Atlassian also made a lot of people redundant and. Even though they denied it was AI, there was clearly there was an angle there. But basically, it does seem that we’re still early in the cycle here, so the impact is not quite as clear yet. To what degree do you think that this was a reaction to sort of that environment where it was just not very good strategy to say you were going to fire lots of people because of AI?
Sam Heithersay 10:20
Yeah, we were very conscious, certainly in the redundancies that were announced afterwards and blamed on AI. We were very conscious of this phenomenon of AI washing, where they were mislabelled as AI redundancies, but in reality, they might have been driven by other factors. And so the survey was very very helpful in in understanding where a company was adapting and could point to AI as a driver of redundancies and where they perhaps couldn’t. Nearly half of the companies, you know, we had surveyed, they were in fact hiring talent. So what we actually found were most of the companies we were speaking to were still very much in that capability building phase. They were they were trying to align their data sets. They were hiring a lot of people to to ensure that was the case. And only now we’re starting to hear more nuanced conversations around the headcount and the token cost that might come with that headcount and where they should be rationalising, you know, token cost as well. But so I think this is very much an early phase of AI adoption where you see this this capability build out. You see some early movers, perhaps mislabeling redundancies as AI. What we’re really looking forward to is is using this as a baseline to be able to assess where we can attribute AI maturity in the way that these companies are thinking about their cost base and where they can’t and so yeah we think that this reporting season in particular we’ll hear a lot more rhetoric we’ll hear a lot of cost out discussion but we’re hoping for a more nuanced conversation about the counterbalancing cost of tokens, and really get a bit of a read on who’s AI mature and deploying it as such, and who’s just using as an excuse.
Wouter Klijn 12:14
I think when you look back in history, you also see that often with the adoption of new technologies, there’s initially a bit of a hiring phase to plug that gaps in skills that companies might have, and you sort of see that in the survey as well. There’s, I think, 45 per cent of companies added AI talent in the last 12 to 24 months at that time. Maybe we’re just too early for redundancies yet. Are we just seeing a lagged effect, or is that still ahead of us? Do you have any sense of what that might look like?
Sue Lyn Stubbs 12:42
Well, we’ve talked about this, how it’s a bit of a J-curve, right? And we’re not going to see that cost efficiency and productivity come through so obviously in the beginning because of this impact. And I think one of the key insights that I got from one of the boards that we spoke to was the fact that also it’s customer appetite for AI, and so one of the companies we were speaking to in the financial sector was highlighting the fact that they actually have to run dual systems-one very much a traditional kind of paper-based, maybe more analogue versus an AI-enabled one, because there was some customer appetite differences, right? So while they are seeing efficiencies and gains from that AI platform, they still have to run the traditional platform. So in the meantime, in the short term, they’re actually having additional costs, almost potentially doubling the costs. Right. So, in some ways, I think it’s going to be a bumpy ride from some capex, from cost savings, productivity gains in the short term. But I think, like we see with the J-curve, you’ll start to see that come through, and that’s probably what we’ll start to see reflected in workforce shapes and outcomes going forward. So I just think it’s hard to draw early conclusions, and I think, like you said, it’s something we’ll start to see over the next six months, 12 months.
Wouter Klijn 13:52
That’s interesting that you mentioned sort of the difference in preferences because you can see that obviously on the client side, but I think also when using AI, in my own experience, you sometimes go in and out with using different type of systems and applications. I mean, when it first came out, there was a lot of applications on the back end of our website, and they were pretty much all useless. So we sort of played around with a few, and now we pretty much don’t use any of them. But in other aspects, in research and that sort of stuff, it’s really helpful. So it might also take a bit of time before we actually see where the efficiencies are and where the best applications are.
Sam Heithersay 14:27
Yeah, I definitely think so. There’s just a proliferation of tools at the moment, and we certainly gauge from the companies we were talking to that they’re all very enthusiastic in the initial phases about rolling out these tools, enabling their workforce. But I think you’re right. Over time, we begin to understand that there should be really frameworks in place. Certainly, governance and risk frameworks that monitor the way these tools are rolled out and the risks that potentially arise if they were rolled out, you know, too broadly without much direction from management top down. So. Yeah, you’re right. We definitely find there’s a a wide spectrum of utilisation of these tools, and ultimately the productivity benefits that the companies talk to, and it again just signals we’re still in very early phases of of this kind of AI rollout for a lot of these companies in the way that might change their business models.
Wouter Klijn 15:19
Yeah. Now one of the other red flags, Sue Lyn, that you mentioned earlier, was if a company has hundreds of projects and randomly applying AI agents, is that sort of a sign that they actually have no idea what they’re doing, or is that more like over enthusiasm? And I ask this in particular because we have spoken in the past with a person called Peter Strikwerda, who was the global head of digitalization and innovation at APG Asset Management in the Netherlands, and so he’s been really focused on how do you build a solid business strategy around implementing AI, and he had that same sort of argument: you shouldn’t just randomly apply AI agents and have all sorts of projects. You need to have a fundamental strategic approach to it, where you embed it in the operational processes, you embed it in decision making, and so that you have a coherent approach to how you’re going to integrate AI. What’s your sense around that? Was there a little bit of like you know maybe these companies got too enthusiastic?
Sue Lyn Stubbs 16:19
I think that’s right, and I think when we start to hear the 100 use cases, the 300 page deck of different ideas, I think that does raise a red flag in our minds around that prioritisation, around that strategic focus of the business, because clearly there are areas that it would be more impactful from a business model perspective, and I think, and it’s really interesting you bring up that integration piece, because we were looking at how AI risks and how AI monitoring should be incorporated across an organisation, and listening and speaking to kind of AI risk experts, they were suggesting that there’s two main forms you can look at and oversee AI risk. One was from your enterprise risk model and integrating it fully, or having a separate AI risk framework, and every single board we spoke to is integrating into their enterprise risk model, and I think that is constructive from an integration point of view. That it’s, it’s you know across the organisation that is embedded, and you’re not creating necessarily a separate system to kind of monitor this. My only concern is that you know AI doesn’t necessarily create new risks, but it definitely amplifies the risks. And so, if you are integrating it across your enterprise risk model, are you potentially not having certain oversight or amplified risk management of certain areas? And I think that is one of our concerns. So I’m not suggesting that it shouldn’t be in your ERM, your enterprise risk model, but potentially there still needs to be additional guardrails or something to that effect, because our concern is that we will see different outcomes, different perception of the risk of AI, and I think that is something that we haven’t really heard from boards or companies at the moment of of how they’re going to navigate that risk. And I think the other one, which we kind of started to touch on, which was shadow AI, around you know employees bring their own tools, and we call it BYOAI, bring your own AI tools. That’s really around you know potential IP loss, compliance risk, data leakage. So our concern is that boards are not also focused on how AI is being used, both in the organisation and outside the organisation, but how it interacts with their IP, with their data.
Wouter Klijn 18:31
Yeah, that shadow AI usage is very interesting because I’ve definitely seen that in the field where some funds they’re very early on in their AI journey and they’re a bit scared of it. That’s my sense, and so they only just approved Copilot, and they maybe use it for taking notes in meetings. But in the meantime, you know, people in the investment team they’re using their own laptops, their own phones because they want to get up to speed. They want to use this stuff, and so if I suppose the board doesn’t give enough rope for them to experiment, then this problem becomes a lot bigger. Do you see anybody do it well? Is there sort of a you know best practice for how to tackle shadow AI?
Sam Heithersay 19:12
Yeah, I think I think that there’s a balance to be struck in the way that you address a risk like AI for all the reasons you highlight on IP loss and data privacy, at that at one point you want to roll you want to roll out enough tools to enable your workforce such that they don’t go home and start using all these AI tools. But on the other hand, you don’t want to roll out too many tools that it becomes less productive at work and I think it’s ultimately up to management and the board to make sure there are guardrails in place to be able to sit in that balance have that happy medium where the tools can engage the curiosity of the employee because we all know that that’s what obviously drive what drives a lot of employees. To their own AI shadow AI tools, they think they can achieve a lot more with the tools that are unrestricted versus those in their workplace. But we also we ultimately want to embrace some of them, but make sure that they are contained and put in sufficient guardrails from management and board down to ensure that it’s still productive and still safe to be used at work. There were the companies that we spoke to were probably that were more advanced were in sectors like software and were clearly some of the leaders, healthcare and financials as well. I think you have an employee base that is already you know more technologically advanced. Let’s say in in just their day to day, and so they were all probably using shadow AI to begin with before some of these companies embraced some of these tools. You know, it’s just amazing how much it’s pervaded our everyday life. I think we are all bringing this. We’re all naturally curious about this, and we all want to embrace these tools. We just want a framework from a company that allows us to do it to realise those productivity benefits. So I don’t think being a bit of a luddite and preventing these tools at work is going to necessarily help most businesses. There has to be a happy medium struck, and it really has to be set from the top down.
Wouter Klijn 21:18
An interesting finding as well was that it doesn’t necessarily reduce the risk if you put a human oversight on it, the human in the loop. Why is that? That seems to be like you know the answer to everything, but obviously it’s not.
Sam Heithersay 21:31
No, human in the loop is it often sounds much stronger than it probably is in in reality, and it’s certainly being used by some regulators as a as a principal core principle, but in our discussions with companies, it becomes very clear that the human ultimately gets lazy over time, and when they see the AI tool deliver you know accuracy of 99.9 per cent it’s very hard for them to capture that that you know point 1 per cent that that actually gets through, so we would like to see more stronger second, third line controls, formal policies in in place, because ultimately the risk of the perception risk, particularly involved with AI, is much greater. By that I mean if if something goes wrong from a claims processing approval for an insurer or a loans processing for a bank. If something goes wrong and we can put it down to human error, it’s much more palatable, let’s say, for a broader society than if we were to say an AI agent got it wrong and pushed through this product. So there’s a perception risk that the companies need to wrap their head around. It’s why we shouldn’t be rolling these tools out on mass, but it’s also why we can’t necessarily rely on human in the loop as a backstop. Because in talking to the companies, it doesn’t seem to be enough.
Wouter Klijn 22:53
Yeah, it’s interesting because there was recent cases at the say the large advisory firms where reports went out that that basically were AI generated and used a lot of cases that didn’t exist, examples that didn’t exist, and they basically had to refund them all their money for this advisory piece of work. You just mentioned people getting lazy. Is that sort of a risk as well that you that that boards identified, or are people cognizant of, you know, abusing these AI tools?
Sue Lyn Stubbs 23:27
I think at the moment we’re not hearing that from boards. We’re not hearing that from management teams. Again, I think it’s too early days, and like Sam said, this was a bit of a baseline. So I think potentially through time we might start to hear that around, you know, cognitive decline, skill management. How do you get juniors coming into the workforce? How do you upskill them, etc. And so I think the whole skill set, expertise dynamic across an organisation will be a key focus area going forward. I think though I actually thought of an interesting parallel. If we think about the airline industry and you think about autopilot, for example, when that came in, I think there were a lot of concerns around cognitive decline, and obviously there were a few tragic incidents in the airline industry where it did happen. Right, you know, some pilots did become complacent, and there were accidents. But I think what’s really constructive is the response that came out of that that approach, you know, there was more mandated hand flying hours. There were scenario training for automation failure. They had crew resource management. So I think it’s a good parallel to look at what potentially AI would need to do. Obviously, probably much quicker in in the instance of AI. But I think you know no one was contemplating removing autopilot, but instead it’s putting the guardrails, putting the necessary training and parameters in place in order to ensure that you don’t have that complacency, that you don’t have that cognitive decline that happens across a workforce. So I think what we need to be clear is what’s going to happen. Where do we need to put incremental resources across a. Organisation not to lose that IP, not to lose that skill set, because of course it is a concern. We haven’t seen it play out yet, but I think that will be definitely an area for conversation going forward.
Wouter Klijn 25:11
There is a lot more sort of attention for what they call cognitive offloading, where basically you know you don’t think too deeply anymore about certain topics because you can let the AI do it, and you know I’m aware this this is probably something for a future survey for some of the questions that may be added on later on, but from that perspective, you can already start to see a little bit in schools and universities where this is creeping up where there was one article that that basically took it to the extreme and said we see students coming out of universities basically illiterate because they don’t they don’t engage with the content they very formally go through the motions of putting their submissions in and when they come out they actually haven’t learned anything what do you think around that is that a key risk
Sam Heithersay 26:01
Yeah, I definitely think it’s a real risk. And from a company context, what we’re talking about more so is just the loss of domain expertise and trust from clients. If we’re seeing that cognitive offloading of the workforce, particularly on tasks that they would they would normally go to that that particular company for, so we have to be very wary of the perception, the loss of trust, and the loss of skill that might come with overuse of these AI tools. I think we have to balance that with the idea that AI is still fantastic for taking away a lot of the drudgery, really, the kind of lower cognition tasks that and so we have to strike this balance where we do want to enable employees and we want to see our companies enable employees to embrace these tools but really draw that line between where a tool might be encroaching upon the their cognition in a way that undermines a whole company’s value proposition, particularly if it’s if it’s service led, if it’s a service led industry. So again, that that to me is just a nuanced conversation that we didn’t have in this early phase of this of this survey, but we would like to have more so with companies, and we will probably judge the companies that can have a mature conversation about this potential risk, which is much more forward-facing risk, and we use that as a signal of their AI maturity for the broader business.
Wouter Klijn 27:27
Yeah, and I think there’s a flip side to that conversation as well, because I think pretty much with every technology throughout history, there has been a cry that skills would be lost, even with the advent of the calculator. And from what I understand, even in old history, with writing becoming more commonplace and people wouldn’t be able to remember anything, so there is a flip side to that. And one of the more interesting examples of that I thought was there might be in the workforce people that are just better able to adopt AI. They’re sort of like super users, and they’re just able to boost their productivity beyond what the average workforce does, and that might create sort of divisions within workforces. Do you have any thoughts around that, where you know people just are so skilled that they might be able to replace you know whole departments just by themselves?
Sue Lyn Stubbs 28:20
We haven’t seen real evidence of that yet. I think we were speaking to one board where they were very conscious of where they were investing in AI, where it made sense across the business, and kind of really targeted pilots’ use cases. I think maybe that was more from a cost perspective rather than a super user perspective, but I do think we potentially will start to see this bifurcation across workforces because of the cost element as well, right? Where do you actually invest in AI use cases that have the biggest bang for buck? Having said that, though, I think across Australia, there’s still we actually were looking at some of the RBA surveys of appetite, trust, and kind of willingness to adopt AI. Australia is still quite low on that curve in terms of trust, particularly against other OECD countries. So I think across the board and boards, Australia is very conservative from an AI adoption perspective. So I think we may not see that super user bifurcation across the workforce come through quickly. It may take time to kind of come to fruition, and I think potentially it does create some certain workforce divides that that could be challenging for a board to navigate in the future. So definitely something we’ll monitor going forward, but we haven’t seen particular signs of that yet?
Sam Heithersay 29:42
Yeah, I do think it’s something that’s obviously always existed. This spectrum of the workforce, where some are just more productive and better than others, where we’ve had discussions with companies, they will highlight, say, a software engineer might be five times more productive than the top tier. Than the bottom tier, but if it’s an augmentable task like software development, then what we find is an AI tool rollout might make them 10 times, 15 times better than the than the lowest tier software engineer. So it might it’s always existed on a spectrum, but that that polarisation might just increase with AI tool adoption, and that will bring up cultural frictions, particularly if you start rewarding that top performer with more advanced tools and models, and the way you allocate costs. But I don’t think that’s necessarily completely different to what companies already manage in their top and bottom performers. It just might increase that that polarisation that we that divide that we necessarily see between the top and bottom.
Sue Lyn Stubbs 30:46
I do think though when we were speaking to our leaders in our in our survey, they were creating cultures that were leaning into AI adoption, kind of celebrating AI, putting incentivization, really championing AI as a an improvement for your productivity, improvement for the workforce, and it was very different from some of the boards that actually shared that they had a culture of fear of AI and across the workforce, and they were actually having to assure their workforce that it it would there wouldn’t be redundancies, etc. So I think it was very clear that across our survey, that the boards that were able to create a culture that was positive around AI adoption were the leaders as well. So I think creating super users, creating too much divide across a workforce, could actually be detrimental to the adoption and rollout of AI as well. So boards have a hard balancing act going forward.
Wouter Klijn 31:39
Yeah, that was interesting because I think there was the example where a company had put in incentives for using AI, and then within the first sort of three months, they ran out of tokens that they had budgeted for. So we’re getting into that stage where the questions change. So you have that baseline now from the 2025 survey. What are some of the areas that you look at maybe adding to the survey token usage and cost might be one of them.
Sam Heithersay 32:06
Yes, definitely token usage and the way that it might offset any potential headcount redundancies is going to be a key topic of discussion. I think it will come up in reporting season the way that companies are balancing token cost against the productivity benefits that they might be talking to, we are likely to see a lot more rhetoric on AI through this reporting season. But I think, given the expenditure that we’ve seen, we’re probably beyond the point of just offering anecdotes on time saved on talking about the number of pilot studies. We now really need to start talking about the return on investment and you know a significant, a more mature discussion of the potential revenue upside that we’re seeing. So that’s where we’re likely to dedicate a lot of the next round of surveys, and I think that all of this is to just gauge company maturity on AI, and in the initial phases, it was much more about experimentation and access. Now we need a real conversation on how costs might play a part in the in the rollout of AI.
Sue Lyn Stubbs 33:15
I think another area that we’re focused on, and potentially I’m being a bit dramatic with this term, but what are boards thinking about a kill switch? Right, you’ve seen what’s happening with AI deployment across organisations. What controls you have? So I’m not suggesting that some of the ASX 200 companies should have kill switches because they’re not, you know, developing AI models. But having said that, one of the key surprises from the survey was the fact that there was minimal articulation of guardrails, no-go areas that the board had identified, even if it was sort of a brainstorming exercise of these are areas that we will not let AI kind of touch or or really run. So, for example, we know it’s still maturing, but there are areas, several high-risk areas such as workforce surveillance, bias, customer treatment, and so I thought this would come through some of our board conversations, but unfortunately it didn’t. And so I think going forward, in addition to cost, we’ll be very focused on what are the guardrails, where are your no-go areas for AI implementation? Do you have the ability to have a kill switch to shut this off if it actually does run a bit rogue in those areas as you start to implement it for revenue generation, for customer-facing use cases, I think that is another key area we will really focus on. I think unfortunately at the moment we’re all just sort of waiting for a high-profile incident where what we saw in cyber, we think will come over to AI and then we’ll start to have regulation. Right, it was a bit of a novelty. Then we had voluntary frameworks. Then there was a high-profile incident, and then we have regulation and fines, etc. And I think, unfortunately, that might be the process for AI as well. With cyber, it took 15 years. My guess with AI, it’s going to be much faster, right? So that’s my concern. Obviously, that won’t come through in the survey, but. That’s all what we’re sort of waiting for.
Wouter Klijn 35:01
Yeah, no, that’s fascinating because you’ve seen recently already with Anthropic’s Mythos 5 model that it went completely rogue and it went out in the internet. And from what I understand, it was setting up fake profiles in in order to try to get customer details and hand over passwords and that sort of information, that would be a good place to have a kill switch where once it gets into your customer base, that you can shut it down. But that seems to be already here. Do you think that that will be a key focus of potential questions in the future around how to protect your client base, your customer base, rather than just IP in the company, but potentially more the interaction with the outside world.
Sam Heithersay 35:44
Yeah, data privacy discussions are always raised in our conversations with companies and cyber security risks. What we find with AI is it just amplifies that existing risk in the system. And you’re right. If you had a rogue AI agent, you know, disclosing that information, then that’s an example of that amplification of what companies already manage, but need to be aware that the game has changed now in the way that these AI tools can really expose any weakness in the data security protocols that they already have in place.
Sue Lyn Stubbs 36:17
What we’re also seeing is this regulatory divide, this AI divide of you know the U.S. being a lot more open and willing to kind of adopt and utilise models versus the European stance, which is far more conservative, really trying to protect customer data, etc. And so I think for some boards, one in particular I was speaking to, financials again, was describing the fact that they’re finding it challenging. How do you roll out one platform, one product when you’re trying to navigate this divide of different regulations, different requirements? It’s actually creating real hurdles for multinational companies. Of what cost do you put here? Do you roll out that model there? Do you actually put that new project or new product there? So it’s a it’s a new minefield of of AI regs, AI requirements, and I think that’s going to introduce new costs as well and new challenges for businesses that I think obviously were there. Obviously, there’s always different regulation across the world, but I think it’s amplified with AI rollout.
Wouter Klijn 37:16
Yeah, did you get a sense from speaking to the boards that I think about that potentially existential level of threat where certain business models might become irrelevant. I mean, software has obviously been an example that has been heavily impacted, but you could see potentially legal firms, advisory firms. I mean, we have listed legal firms. So, you know, do boards ask themselves to the question.
Sam Heithersay 37:41
Yeah, well, it’s incumbent upon us as investors to ask the question first. And so, in addition to this survey work we did, we also went through all the companies that we own and brought a ASX 200 index with a large team of analysts from a using a bottom-up framework of AI disruption risk, we had an 11 characteristics to judge each company. Whether it’s defensive moats, like do they have a safety criticality that might prevent an AI native startup from challenging them? A proprietary data moat, regulatory moat. Is there adoption friction in the way that they can roll out their own AI? Whether it’s workforce unionisation, and through that process, we were able to understand where companies might be able to defend themselves against this existential threat that you highlight of AI disruption. But I think what we’ve also paired, what we’ve also realised, particularly being longer term investors and having the history of looking at previous examples of disruptive technologies is that it’s very hard to actually understand in the moment who can not just defend their existing business model but adapt to change because this this AI is clearly a transformational technology that will force companies to change in order to survive and we’ve had examples of this previously. One we drew on was digital photography in the early ’90s with Kodak and Fujifilm. And if you had run a framework using a backward-looking defensive moat framework like we’ve done, you might come to the conclusion that they were both losers in the way that they couldn’t defend their print photography business against a digital photography business. But if you were to understand which company was more mature in the way they were able to adapt, then you might be able to pick that Fujifilm quickly adapted. It diversified its portfolio. It wasn’t concerned about necessarily cannibalising its print business. It survived where Kodak failed, and I think the parallel example here is true too. We will see all companies are talking about how they’re going to defend their business model at the moment. But what we’re probably more interested in for as longer term investors are those companies that recognise this might be an existential threat and are mature enough in their thinking around AI to adapt their business model and. Perhaps change it entirely, and the best leading indicator we have in this very early stage is governance. We feel like that’s the way that we can assess where a company is thinking very deeply about this risk, and who is more likely to adapt in order to survive. Because you’re right, it will be an existential threat for some companies, and it’s up to us to pick which ones we think can adapt rather than just defend.
Wouter Klijn 40:23
Yeah, and going forward, do you have an idea of what areas might be coming up in the next survey where you’re going to have an extra close look at?
Sue Lyn Stubbs 40:32
Cost is definitely that one of the key ones. As we mentioned, I think you know we’ve already heard from one board how they were looking at revenue per FT. If this was a tech company. They were looking at revenue per FTE, and they were changing their board metrics to better understand cost and AI capex. And so now they’re looking at revenue plus capex per FTE. So that’s a clear signal that boards are already trying to get a better handle on what the investment is, how effective it is, how efficient it is. So I think cost is very much one of them, and I think the other area is really that those guardrails, that risk mitigation piece. If human the loop fails, what are your next defence lines, etc. I don’t have good examples for those. I think it looks different for different organisations, but I think that those are the areas where we’re really wanting to see more information, much more concrete answers to those questions.
Sam Heithersay 41:24
Yeah, as I said, I think we have to move beyond the anecdotes now, and we have to move beyond the just repeating the number of pilots that they might have. We want to see progress return on the capital that’s invested, and we do probably want to see companies moving beyond just thinking about this, as 70 per cent of them did from just an operational efficiency gain. I think the next stage is is augmentation, so not just where you can replace tasks, but where you can enable your existing employees to be more productive. I think after that, that we have to look at more transformation of of end-to-end processes, and really that the final stage in my mind is this complete reinvention that might be required for certain companies if they conclude that AI is enough of a disruptive threat that they have to do, you know, a Fujifilm style diversification. So, we’d like to see more understanding of a better understanding of the potential threat and how companies are going to adapt because the operational cost efficiency answer is probably not going to be enough in the next iteration of this survey.
Sue Lyn Stubbs 42:28
So our goal with the survey is really one to be able to see the companies that we surveyed. What is the progress? You know, a year on, how have they maybe matured in certain areas? What areas have they improved on, etc. But then I think we also want to expand. We want to cover more of the ASX 200. Unfortunately, it would be it would be lovely if we could do it across the ASX 200. But it is more of a detailed company engagement, speaking to the board. So quite manual and labour intensive. But I think as we look, there might be more proxies or data around AI that we can kind of use as to enhance and and complement our survey as well. So that’s something we’ll be looking at. But I think it’ll be great to do a time series and understand how our companies have improved, what are the differences, and then also expand that across more sectors and across more companies as well.
Wouter Klijn 43:18
Yeah, maybe final question to end up with: Are you using AI yourself?
Sam Heithersay 43:23
Yes, definitely. I use it. I use it for work. I use I use it at home. Yeah, using it for parenting hacks. So there’s no part of my life it probably doesn’t touch. And yeah, I think it’s I think it’s very much in the camp that it can. It’s a fantastic, you know, tool. It’s a fantastic technology that we should be embracing. But I guess yeah, doing this this this work highlights that it does come with risks, both personally in repeating my parenting hacks to my wife and professionally at work.
Wouter Klijn 44:00
Fair enough. Sue Lyn?
Sue Lyn Stubbs 44:01
Also using AI a lot and have multiple tools as well. I think it’s such a useful tool. It can be so effective and helpful, but obviously with the right guardrails, with the right risk considerations as well, and just being very clear of what it can and can’t do. Right? What are the limitations of it? But yeah, I love it, and we’ll continue to use it, and it will be very much part of my day-to-day activities all around.
Wouter Klijn 44:27
Excellent. Well, thank you very much for this conversation, and thanks for coming to the offices. It was great to have you. Thank you.
Sue Lyn Stubbs 44:32
Super. Thank you so much.
