In Business with AI - AI adoption

In this episode of In Business with AI, Melbourne Business School Dean Professor Jenny George sits down with Simon Kriss, the founder of Sovereign Australia AI.

In Business with AI

In Business with AI is a podcast presented by Melbourne Business School that explores how artificial intelligence is reshaping business, leadership and the future of work.

Hosted by Dean Jenny George, each episode features conversations with AI experts and CEOs who are implementing AI in unique ways in their organisations.

We unpack what’s working, what’s not working, and how leaders can navigate one of the biggest technological transformations of our time.

Episode 2: AI adoption

 

In this episode of In Business with AI, Melbourne Business School Dean Professor Jenny George sits down with Simon Kriss, the founder of Sovereign Australia AI.

Simon spent 30 years working in customer experience before consulting with brands like Coca-Cola and Apple. He’s the author of The AI Empowered Customer Experience and consults with boards and executives on how to build AI literacy.

In this episode, Simon unpacks the different layers of AI adoption, where he believes Australian organisations need sovereign capability and the risks of data going offshore.

Simon shares how AI is changing customer experience, why customer testimonials matter more in the AI era and his advice to capture high value business opportunities. 

Transcript

Jenny George

Hello, and welcome to In Business With AI. I'm Jenny George, the Dean of Melbourne Business School. And in this podcast, we're all about takeaways that you can put into real life action as a leader within your organisation. 

Today, I'm so excited to be joined by Simon Kriss. Simon is the founder of Sovereign Australia AI and a leading voice on AI adoption. And you've got experience working with some of the biggest names in the world-- Coca-Cola, Apple, many brands that we're all very familiar with and who, no doubt, have got a lot of really interesting thinking around AI. You're also the author of The AI Empowered Customer Experience. We're going to explore some of that with you. But I'd really love to start with your story. So thanks for joining us, Simon.

Simon Kriss 

You're welcome.

Jenny George

You've got a military background I think.

Simon Kriss 

I do. So I left school and didn't know what I wanted to do. And so my mother said, well, why don't you join the military? So I did. I joined the Royal Australian Air Force. I'm trained as an electrical fitter. And I'm amazed in the field of AI of how many ex-electrical engineers and electricians I come across and worked in that field for a number of years and then left, moved into sales and marketing, and spent about 30 years of my career around customer experience. But always with a lens on technology and how technology could enable- because I'm a bit of a part-time nerd. And I got really interested in AI, but it never quite delivered on the promise in customer experience. I could do some regression analysis or something, but not much else. And it was only with the advent of generative AI, even in its early forms, that I saw it could have a radical impact on customer experience. And so I just leaned in further and further and further. And as you know, you start with AI. It's a bit of a black hole you just fall into and keep going.

Jenny George

Yes, definitely a lot of rabbit holes that you can find yourself in. But let's talk a little bit about your transition, I think from probably that quite technical background, the interest in customer experience, and then how these things have come together with AI in the mix. You describe yourself, I think, as a bit of a nerd. How did that nerdy approach to the customer experience been melded in with what's happening in AI?

Simon Kriss 

Yeah, they did have a natural joining point. There's a term that marketers have been using for probably three or four decades, which is mass personalization. And mostly what that's meant is, we're gonna put you in a homogenized group and give you a treatment based on that group. Where AI now truly allows us to do true mass personalization at scale. So if you think about not only what products and services you might be interested in, Jenny, but we might know that you traditionally buy after a voice call. But your service stuff you like to do on chat and on voice you like to be quite serious, but in the chat world, you are happy with acronyms and jokes and shortenings and things like that. And just learning and knowing that much information and being able to process that really gives us a great point of differentiation. Because like most things, customer experience often comes down to trust and trust is built through knowledge and time.
If I go back to when I was a child and I would go to the butcher shop with my mum, you know, you remember that little bell on the door and the sawdust on the floor, that butcher knew my mother so well. He not only knew what she bought and when she bought it, he would even know how much he could flirt with her and get away with it. And if something was wrong with that product, the next week mum would still go back and tell him what was wrong. In today's world, if I'm in Coles and I buy something and it's not quite right, next week I'm in Woolworths. And so it's how do we build that level of trust through knowledge that we weren't able to do before, and AI now allows us to do it. So it really was a natural thing for those two to come together.

Jenny George

And you've now made a shift again, this time towards the whole concept of sovereign AI. That's a huge topic. Before we start talking about you and where you are at, let's maybe zoom out a little bit and think in 2025, about two years after the first chat GPT really went mainstream, we then started having a lot of debate about what is the role for nations in thinking about AI capability. And this became something where different countries introduced different kinds of regulations and different ways of doing that. As you looked at that debate at the national level, what trends could you see? Where does Australia fit in that picture?

Simon Kriss 

It's a really interesting picture. And for me, that picture is blended. I'm blessed to be part of a UN sponsored group called AI for developing nations. And we meet twice a year at UN offices. And looking at some of what the developing nations are trying to do, their expectation that actually they will leapfrog some of the pioneers because of the learning, they're just going to be able to leap forward. Others though, we're still grappling with, do we buy GPUs? Do we put up another 5G tower or do we build a hospital? So some really quite fundamental arguments out there. Australia is, it's an interesting little quirk I find in Australia. We have some of the most amazing AI researchers in this country.
The main person behind China building the DeepSeek model did their masters here in Melbourne. So we're exporting this stuff around the world. And yet we seem reluctant as a country to take on a sovereign position. We seem to be quite happy to just hand it over almost in a digital colonization way to the US and China and other superpowers.
And many other nations have already made sovereign moves. We're actually behind the pace. But it's a huge debate here and often a very binary debate. And the thing about sovereignty and AI is it's anything but binary.

Jenny George

So talk a bit about that. When you say it's not binary, that's because there are multiple different choices that you can make, in perhaps lots of different optional ways. What do you think is the reason that Australians are thinking this is a yes, no, or a binary choice? What aren't they seeing about those options?

Simon Kriss 

So the first thing they're looking at is the model producers, the frontier model producers have spent reportedly billions and billions, actually a lot of that is just overinflated company valuations. But they've spent billions of dollars getting to where they are. We can't keep up. We don't have the GDP of those nations, so on, so on, so so. So a slightly defeatist attitude, somewhat like we took with car manufacturing years ago.But sovereignty occurs on a whole lot of layers. So the very first layer at the bottom of the stack, if you like, is the infrastructure. And this is getting the most debate right now between politicians and media.

Jenny George

So data centers.

Simon Kriss

Data centers. And the GPUs, CPUs, RAM that lives in those data centers. Yes, the actual chips can't be sovereign at this stage because we don't manufacture chips in this country, and probably never will. But we can buy them and we can host them here. Of course, you have the hyperscalers, the Googles, Amazon, Microsofts of the world, encouraging us not to build sovereignty, but to leverage their infrastructure. And the one critical problem I have with that is a piece of legislation in America called the US Cloud Act, which basically gives the American government the right to compel any American company to share the content of any server anywhere in the world that they own or lease. And if they're a service provider, they can't tell the customer that they're accessing the data. So, you and I might have my health records that's hosted in AWS, and the FBI can look at that at will.

And that's going to be a problem for us. As we get into health type data, truly personal religious belief and sexual orientation and all of those types of things, we need to be protecting all of that. So that's that base layer. And there's a lot of discussion going on, and we have a lot of sovereign capability being built out of that layer right now.

Jenny George

So the implications then are we often see companies here in Australia saying a lot of my data hosted here. We have servers sitting in Sydney. That's not enough anymore, because if it's an American company, and many of them are, then the US government has some ways that they can put pressure on those companies to potentially show that data. And we wouldn't even know it.

Simon Kriss 

Correct, that's exactly it. And it's not just where the data's stored. The other thing to think about is where is your data processed? So OpenAI has a limited GPUs in this country. Microsoft have limited GPUs.

Jenny George

Now I'm just gonna stop you quickly there. What are GPUs? We get quite a lot of technical terms and maybe not every person who is watching this podcast quite understands it.

Simon Kriss 

And that's a great call in AI. We fall into using all of these acronyms that we just think everybody else knows.
A GPU is a graphical processing unit. So if I take you back to your laptop, you have a CPU, a central processing unit that tends to do things in a one, two, three, four order. But you also have a graphical processing unit. And these are designed to do complex mathematical calculations at scale. They were first designed for gamers. So for those of you that are listening to the podcast that were gamers back in the day, thank you very much. We built these chips, these graphical processing units to process millions of calculations so that we could make the little soldier on the screen run across the screen left to right. It turns out that AI, because we turn every word into a number, also needs a unit that can process millions of mathematical calculations per second.
And now we've just taken them out of the size that would appear in your laptop, which would probably fit on your fingertip, to a unit that's half the size of the table that we're sitting at. And of course they need extra power and cooling. And so we can't run them in an office environment. So we tend to put them in a data center. And that's why people talk about all of these GPUs being in the data centers.

Jenny George

Simon can we  talk just a little bit more about data centers because they are a little bit controversial for some people, particularly power use and water use. What's your response to that? Is there a different perspective we should be taking here?

Simon Kriss 

I'm a little bit pro data center and I get that they might look ugly in your suburb, if you're in your suburb, and occasionally when they need to fire up their generators, that is a diesel generator that's running. But they're doing a lot of work in renewables. They're building data centers with water treatment plants. But the very much publicized power and data usage, for me, is a little bit of a furphy, to use another good Aussie phrase.

There was a study just done recently by Mandela and they discovered that we use more electricity in Australian shopping centers than we do in Australian data centers. Now that's true of today. And I get that if we stay on the explosive path of data centers that we're being asked to, that will change over time. From a water perspective, we actually use seven times more water in our public swimming pools than we do in our data centers. Now there's a social license to that. Another way to look at it is we use more water in golf courses than we do in our data centers. And the other thing to keep in mind is that the tech is getting better and better and better by the minute.

So now we have GPUs, graphical processing units that are cooled by liquid, much like in your car. It's a closed loop system. And then we just use a little bit of water and air. And we're in Melbourne and that's particularly good for data centers because most of the year, we can cool with air because we've got, if you're in far north Queensland, that may not be the case, you may need a bit more water. But even the latest chips are being designed to run at 30 and 35 degrees. So we're actually going to have to cool part of the data center for the human workers, not for the chips. So I think there is an opposing view and I think we need to take a better or more balanced view of data centers.

I think there is a place for a difference between an international data center. So let's say a big international conglomerate decides to come to Australia to build a data center to serve their US clients because we're just a good resilient place. I mean, if you look at everything that you want for data centers, Australia is the perfect target. I think if they're building data centers for that, then they should be contributing to our power generation, our water generation actively. But if it's an Australian data center running Australian software like Xero or Canva or something for Australian customers, they should be treated like any other business.
I don't believe at this stage Anthropic has any GPUs in this country. So every time you use Claude, most of the times when you use ChatGPT, OpenAI, that data, that inference is being processed on a GPU in another country. So although it's only being processed or at rest waiting to be processed for a couple of seconds, that still means the data is leaving this country. So I'm often amused when government departments talk about, oh no, we have all of our data here. And when I ask what AI tool they use, I'm immediately kind of giggling that they don't really have control. So there are alternatives. The alternatives are not at this stage as big as the AWS and Google, but we have NeoCloud providers in this country that are fully Australian owned.

Jenny George

You’re probably not so worried if you're creating a funny cat video or something like that. But of course, if it's data that's got sensitivities and especially Australian government data, there are real concerns about how that might land in another country. 

Simon Kriss 

Correct, and sovereign by the way, is not going to be every process in a business. If all you're doing is providing advice and next steps on how to apply for this credit card or do something, and you haven't even asked the customer's name at this stage, then who cares where that's processed? Because it's public knowledge that's probably living on a website somewhere anyway. But it's the little stuff that, and here's a great example. Imagine that you have a call center here at Melbourne Business School and somebody calls in and you want to use AI to transcribe that call and put a record in your CRM. Sounds like a great use case. If you haven't redacted that transcription, when you pass it off to an AI engine, to an AI model to summarise it down into 50 words, you're handing over name, address, probably date of birth, because you've done ID and validation in the call to make sure that I'm talking with Jenny George. And then there could be health information, the children's school information, all this kind of stuff, is just contained within a phone call. How are we making sure that that data isn't winding up in the wrong place as it easily could? 

Jenny George

So, Simon, you've talked a little bit there about privacy. How should businesses think about privacy and protect themselves?

Simon Kriss 

The very first thing is know who's using AI and how they're using it. Unfortunately, a lot of organisations jump straight to, it's dangerous, so nobody uses it. And all they've done is driven that use underground. We often refer to it as shadow AI. You want to bring it out of the closet, make it okay for people to use tools. Treat them like adults and give them some information.
Definitively, no private or confidential information into a public tool. I think everyone kind of gets that. I think beyond that, it's as you are starting to AI applications. Ask more questions, particularly of the sales team, right? And you will often need to go beyond the sales team to the technical team. To find out where is this AI hosted? How is it being done? What protections are in place end to end? Are they SOC 2 certified? And those, the rigor that you would normally put around any other IT, you need to definitively put around AI.

So that bottom foundational layer is an important layer. But then the layer above that is the models. What models are we using? And where did those models come from? How were they built? What biases been introduced? What was the dataset? Was that data sourced ethically? We know in the case of most language models, it has not been sourced ethically.
What do we or don't we know? And people don't think about the hidden bias in a model, but the biases can actually be quite strong and introduced quite innocuously.
And that's the field that in particular, I have a burning desire to play in. Then you've got an agentic layer. So what tools are we using to build agents and things? Are we using Chinese tools, US tools, or Australian tools to build those?

Then you've got your layer around observability. So control planes and harnesses and all these technical terms. But how are we controlling these agents and making sure they're not going rogue and not doing the things they shouldn't? Are we using Australian software for that? Or are we using a US monitor service for that? Then you've got your application layer. And that's the other layer that government and most businesses really understand. They kind of understand data centers and the application layer. So that application layer would be names that you would know, right? Canva and Heidi Health and Harrison and all of these guys have got beautiful applications there. And then the sixth layer for me is regulation. What is the government's position on this? And what should we be doing? What shouldn't we be doing? And I know that that's a very hot topic right now and there's a joint select committee on AI that has deliverables in the next 100 days and things like that. So they're kind of those six levels. So across that, you need to look at what is your organisational values and how does that intersect with it? What processes must have sovereign control and what, eh, doesn't really matter.
What work are you doing? Therefore, what do you need?
Because a lot of the agentic work and simple chatbot type stuff doesn't need frontier model capability. If you were using it to check all of your code and you're an IT shop or something, sure, maybe you do. But a lot of businesses don't. We're just buying it because we're told it's the best. It's the biggest we should use it.

Jenny George

Yes, but using Fable to write your email is a bit like sledgehammer and the nut example.

Simon Kriss 

That's exactly what it is. And the other problem with models that aren't built here is that we don't, I love, Mr. Marles used the term agency. And I don't see how we can have agency without sovereignty, not true agency. And you mentioned Fable, that is the perfect example. The US government make a decision and anthropic turn it off. And it turned off at 6 p.m. Friday, our time. If you were a bank and you were using that model as part of your payment processing process, and it fell over, most Australian companies haven't yet gotten to the point where alarms go off when those things fail. They do for databases and applications that they've had for years, but because AOS are new, often there's no alarm. So you may have a situation where the US turns off a model and no payments are processed for 48 hours. So it cost us, at our economy, billions of dollars.

The other thing that I want to say about sovereignty is using overseas models on an overseas platform. So open AI on a Google platform. The economic benefit to Australia of us using that model is cents in the dollar. Absolutely minuscule. And so where is the economic benefit for this country?
And what are we prepping our students for? You know, if there's no exciting work here, they're all going to gravitate offshore. And we've already seen that brain drain to a certain degree starting to happen. So I think we need more things that are going to excite our students and keep them here and working on exciting work, rather than them just floating away because we've got, what did someone call it the other day? Managed dependence with the US. I'm sorry, I just can't say managed dependence is working for me.

Jenny George

Let's go back and talk about the model piece. There's a couple of different things I think are really interesting. The models themselves have what are called weights, which is where you're looking at all of the various tiny little pieces of information and deciding how they're best put together. On top of that, you also have a layer, which is how the model then interacts with a human being. And that's quite interesting. And those models have been created often to make humans feel good. And to give them, now of course, give them the answer they want. Part of that's about the weights. So part of that is, are we putting together all this information correctly in a way that human beings find the output useful? But another part of the way the models interact is the style they use. And it's really interesting that of course, these are mostly built by Americans and they've been built with an American style. How much do you think each of those different elements, the weights that are used in those frontier models, where we're iteratively with those frontier models getting to more and more accurate model weights that do a better and better job, but then also the stylistic choices that each of those different frontier models and various other models that come out as well use. Where does sovereignty fit into those equations?

Simon Kriss 

Lovely, nice questions.
This is probably the most engaging conversation I've ever had on sovereignty and AI.
Look, weights are, I keep saying to people, it's much like when you use Excel and you want to change the thickness of the border and they talk about the weight of that border. It's the thickness, right? So, the dog chased the, when you're going to have a really thick line to cat and ball and almost a dotted line to oyster, right? And that's just built up by how many times has the model seen the dog chased the cat? Well, it's seen that a ton more than it's seen the dog chased the oyster. So--

Jenny George

And it's the set of probability distributions that link all of those.

Simon Kriss 

That's right, I've got a really, really strong connection to this one, so it's highly probable that's the right answer and so that's what I'm going to use. And then we tell the model, be helpful, be professional, be courteous, help the user, that type of thing. So, all of that kind of comes together. And so the only way you start to change those weights is to feed it more and more and more and more information. And right now we're almost at a tipping point where if you built a model today and all you did was take the contents of the internet, almost a quarter of what you would consume would be synthetic information that has been output by AI earlier. So you either keep doing that or you start adding more of the content that you want to that model.

Jenny George

So controlling the quality of the content by curating it carefully.

Simon Kriss 

Curating it much more carefully.

Jenny George

And being much more thoughtful about that.

Simon Kriss

Correct.

Jenny George

But of course that's a choice.

Simon Kriss 

It is a choice.

Jenny George

And then that curation may result in information that's more or less helpful depending on what country you're in or what culture you're from and so on.

Simon Kriss 

Yeah, and how it's biased. So, you know, in Open AI and anthropic things like that, Australian data makes up 1 to 1.2%, which is punching above our weight given our tiny population. I'd like to see a model that was seven, eight, 10% Australian data and that the model was biased to that data as it was being built, right? In other words--

Jenny George

And here bias is not a negative thing. Bias just means overweighted.

Simon Kriss 

Yeah, take 80% of your answer from this bit of the data and then use the rest of the data to form the other 20% of the answer type of thing. So when a child asks about what's it like to be in the Navy, it defaults to talking about the Australian Navy, not the US Navy. And all of our US models default to US things because they've simply been trained on far more US data than Australian data. We need to train on world data because we still need to be able to talk about the length of the Nile and where is the Golden Gate Bridge. But we want to have a lot more Australian data and also data that complies with our privacy and e-safety laws and most importantly with our copyright laws. So how are we making sure that Australian copyright holders have been made whole in this process, which the US not only doesn't care about, it is actively lobbying against. And just yesterday, the Department of Justice came out in New York and told the court judges that consuming a book into a model constitutes major transition and therefore copyright shouldn't apply. So we've got this stuff going on that isn't aligned to Australia's values and Australia's laws. The other bit is, we're, Aussies are a weird bunch. We love to take the mickey out of each other.
If you're not being insulted by your friends, you're probably in trouble. And we love, Carl Barron talks about this, we love to talk about what things aren't. So how are you today? I'm not too bad. How much was that? Oh, it wasn't that expensive. How far is it? Oh, it's not too far. So we say what things are not. Americans tend to say exactly what things are. If you take that to a health setting, when my aunt is asked, how are you? And she says, I'm not too bad. That means I'm really sick.
And so trying to bring that culture, that vernacular, those values into a model, for me is an important part of sovereignty. I don't think it's the deciding factor because we can force an offshore model to do that. But it's definitively an argument for it.

Jenny George

Yeah, thanks so much, Simon. So as we think then about all of those various layers, what's the role that the Australian government's playing in thinking about the different considerations for sovereignty and how much Australia should be playing in these different layers?

Simon Kriss 

There's still a bit of literacy needed at the government level, I'm afraid to say. I have the pleasure of talking with ministers and senators on both sides of the house, and literacy is a very large part of those discussions.

If I look at places like the UK, as kind of a potential model to base ourselves from, when they announced the Office of AI, they announced a billion pounds in funding. They subsequently set up a 1.1 billion pound sovereign AI fund to build sovereign capacity. And they've just launched a 200 million pound ‘by UK AI’ strategy for the government.
We're not doing any of that in Australia. I think all that we've invested about $3 million. We've set up the NAIC, the National AI Centre, and Lee and his team do some amazing work there. We've got the AI Safety Institute. Now we've got an Office of AI, but we've also got South Australia doing a Royal Commission and ... So we're doing a lot of discussing, doing a lot of at the risk of offending naval gazing without really making concrete steps in this area. And the steps that are being made are either discussions around data centres and expecting that data centres will somehow magically want to give up some of their compute to sovereign, which I can't see that happening.

Or it's this, how do we go and make sure that we get a share of what the US is doing in a, please sir, can I have some more type of approach?
And both of those are helpful to a certain degree. Yes, we do need to still be plugged into frontier capability but at the same time we need to be developing our capacity. And we have it from a brain power perspective. We invented Google Maps and WiFi and black box flight recorders and pacemakers. And we invented the Hills Hoist and the Gooney bag, and we know how those two work together.

The rest of the world doesn't. We've got the capability, it's just the willingness and the funding. And a lot of the government funding and the investment community in Australia, I want to say high net worth individuals, family officers, and even the VCs, still don't quite understand what sovereignty means in AI and why it's important.
And they're also still tied to an old funding model that says, get one or two million, build a product, go out, get 200 customers, and then we'll put some money in. When you start talking about building a data center, well, you've got a couple of billion dollars upfront. If you're going to build a model, you're talking a hundred million dollars upfront.

These are things AI is kind of changing the investment profile where your seed round is an $80 million seed round because that's just what it takes to get a model going. And then it's off and away to the races. So different governments are doing different things. There's 26 countries outside of Australia where a sovereign foundation language model has either been built and they're using it or is almost finished. 26. And about 85% of those projects had some level of government funding. There are a number of them, and Switzerland is probably the most obvious one to look at, where the government funded the entire thing. It actually got universities, gave them the funding and said, build us a model that we can then give to the people. And it's not just open weights, it's open source.

They've said, here's the weights, here's the data, here's everything that we did.
And that may or may not be an opportunity for the Australian government. But at the moment, the government is definitively saying, well, industry will take care of that and industry is not taking care of it.

Jenny George

And of course, there's two big wise reasons for this. One, you've already talked about business continuity risk, but the second is defence. And that's one that I think we're all maybe at the back of our minds aware of. But the role of AI in essentially defence systems, countries both offensively and defensively, is certainly coming if it's not already well here. Have you heard much about how that thinking is interacting with the government's thinking about the private sector being responsible for sovereign AI? Because it feels like something that other countries are seeing as actually part of a defence spend.

Simon Kriss 

It is in a lot of countries a part of the defence spend. I've seen some evidence of defence spend, in the generative AI field, but more in the traditional AI field. So definitely, embodied AI like drones to either prosecute or defend, assessing whether or not a battalion is fit and ready to take that hill based on internet sensors in their suits, really practical embodied applications. When it comes to identifying threat actors early and having defensive positions and things like that, I think we're still kind of waiting. And then the obvious one of course is, if you wanted to take down a government or a country, take down the data centres, take down their artificial intelligence tools, and they kind of grind to a halt, they go a bit dark. Where are we on that? And I don't know that our defence force here has an answer for that yet.

Jenny George

And obviously from a national defence perspective, the more we embed AI into everyday processes, into the way that payments happen, the way that retailers operate into every business, the more likely it is that that could be a way to attack us as a country. It doesn't have to be a weapons system.

Simon Kriss 

If you think about what SISO refers to as the attack surface area, for a number of the businesses that are listening to this podcast, watching this podcast, they're starting to adopt AI to write monthly reports and do this and do that. If you have an agent, just one simple example of an agent that writes the monthly report, it's rare that the agent gets it right the first time. The agent's going to have 20, 30 goes at writing that report. Now, if it stores every go that it had, the outbound inference and back, you now have 30 times the reporting data being stored somewhere. So your attack surface area has just grown immensely, right? 10 fold. So you open up attacks through connectivity, you open up attacks potentially through models, but you just don't going to protect by having more data hanging around that could be potentially hacked.

Jenny George

You've talked about the layers that are part of any AI ecosystem. Do they all have to be sovereign? Some of them cost a lot more than others. Do we have to invest in everything?

Simon Kriss 

No, I don't think we do. Like I said, some of the processes need to be sovereign, have sovereign capability. But—

Jenny George

So which ones? 

Simon Kriss

Well, ones where you've got definitively got an Australian's personal data or highly confidential data about your business that gives you your competitive edge, your unique value proposition.
Granted, anything to do with defense is going to be right up there. But for the rest of it, you don't necessarily need to be sovereign. Accenture wrote a great report last year where they called out that they believed that 30% of agentic AI workflows would need sovereign capability.

Personally, I think they've over-egged that number a little. I think it's probably more like 20% of total workload. Unless you're a government department, then it might be 40 or 50. If you're a defense, it's probably 80 or 90. Or the intelligence community. So for most Australian businesses, they really only need to think about true sovereignty for a slice of their work. And even then, it's what is that use case? And so what parts of the AI stack are you using? And therefore, what needs to be sovereign? So it's very nuanced and very non-binary.

Jenny George

There's kind of a vertical and a horizontal view here. So what you've been talking about is there's some processes where they need that whole vertical slice, all of the different layers to all be sovereign. But there's a whole heap that doesn't.

Simon Kriss 

Correct.

Jenny George

Are there any of the horizontal layers that don't need to be sovereign? So from application through to all of the agents, the foundation model, the data centers, do we need a sovereign slice that's vertical across all of them?

Simon Kriss 

No.

Jenny George

So which are the horizontal layers that we could outsource?

Simon Kriss 

Let me give you an example. Imagine you're a medium to large company and you're using Salesforce. Arguably one of the world's best CRM systems. And I don't know that we have something in this country that is that advanced. And so you're a large organisation, you want to reap the benefits of using that. Then the application layer isn't sovereign. But you could say to Salesforce, when you get to doing this task, I want you to make sure that you're using a sovereign model that's running in a sovereign data center. Because we need sovereignty around, we need to protect that data being processed or being stored. But everything else that you're doing in Salesforce, yep, don't worry about it.

Jenny George

So it sounds like there's almost a mental checklist of where is the data? Does it have to be private? And where is the, I suppose the capability that we really want to use that sort of sovereign model for? So that's the question you ask about the very particular task. 

Simon Kriss 

Correct, about use case by use case. And look, any decent AI governance model is kind of for me winds up with an X and Y axis of along the bottom is how risky is the AI use case? Because summarising phone calls in a call center, nobody's going to die if that goes wrong, but synthesizing a new drug might be different. But on your other axis is the sensitivity of the data. If it's publicly available data, who cares where it's processed? But if it's PII, I privately identifiable or health data or payment data, no matter how simple the use case, you've still got a high risk involved. And so sovereignty in this sense is about controlling that risk.

Jenny George

Well, let's change tack a little bit and think about how businesses are leveraging AI. And I know you do a lot of work with C-suite leaders, particularly I think on that last applications layer where we know Australia is actually doing a lot and we've got smart people who are working on really interesting AI applications. What do you think leaders should be thinking about when they think about how to create business value?

Simon Kriss 

I think the first thing leaders need to think about is, and it kind of brings my, this is the only time my masters does come in, is I find AI has a liken to fine wine. When you were drinking fine wine with someone, you know, that person that really knows wine and they're talking about acidity length and tandem structure and your sitting there going, tastes like wine.
You don't wanna be the one that, you know, I don't know what you're talking about. Just kind of quietly sit there and drink the wine and pretend like you know. And I'm seeing that a lot in the sphere of AI. Most executives are super busy. They don't have eight or 10 hours on the side of their desk to go off and learn about AI. So they read articles here and articles there, they might play with something at home, but their deep understanding of AI for most leaders that I come across is simply not there. If I ask them to articulate the difference between machine learning, traditional AI and generative AI, they can't do it. And so they're flying blind a little bit. So, you know, I'll often say to a C-suite leader, what would you like to do with AI? The most common response I get is, I don't know what can AI do. So they haven't yet, if I ask them, what do you want your phone system to do? They could answer it because they've been around it for years, they still haven't. So in this country, I think we've still got a job to do and it'd be a great job for Melbourne Business School, a job to do around AI literacy.

Jenny George

Yep got great programs on that!

Simon Kriss 

You do and led by the amazing Dr. Jon Whittle

Jenny George

Yes, in fact our Institute for Digital Innovation and AI is doing so much great work in that space. As well as in our longer programs, yeah.

Simon Kriss 

Awesome, and yeah, and do we need more of it? At every level, like I said, federal senators need to do this stuff, CEOs and CIOs need to do this stuff.
Then that next thing is rather than just jumping on, I was at a conference and saw this tool and we need to buy it. Actually stopping and going back and saying, okay, where are we in the business?
To Jon Whittle's point, what's our vision? Where do we want to go? But look at where is the biggest pain points in our organization? And if you start talking around the organization, you're going to find the pain points, those jobs that are highly repetitive, low value, no one's ever seem to be able to fix them. They're your obvious targets to start with. Go after that sort of stuff.
Rolling out co-pilot is never going to give you big ROI or big wins or anything. It's a lovely tool, and whether that's ChatGPT Enterprise, Claude Enterprise, whatever, it's going to have minimal return on investment because for the most part, it's you asking a question and it returning you an answer. Great, that's Google on steroids. 
You don't really start to see the ROI until they're getting into agentic processes, more automation, but often when you do that, you kind of stop them and go, hang on. You're trying to automate a form because that form is really badly written. Why don't you just rewrite the form? So AI isn't always the answer. Sometimes you just need some good old transformational innovative thinking and then look at what kind of AI you need to kind of steer your journey. So--

Jenny George

And of course, it's not just redesign the form, it's does this whole system potentially change? And might we end up-- Do we even need it? A world in which forms don't exist because we're actually doing things in a completely different way. But that requires more than one organization to change. It actually requires the whole system to change. And when we've seen that in the past with the adoption of the internet or with other really major changes, that's taken quite a long time. What's your view on the timelines here for the simpler changes and then the system changes that would be added?

Simon Kriss 

I think it's going to happen faster than the internet. The internet kind of happened over about 12, 13 years. I think this one's more got a window of five years and we're probably two years into that five years. So there's probably another three years or so that organizations have got. So definitely you shouldn't be sitting on your hands. I had a CEO recently say to me, "Oh, we're going to wait until AI slows down a bit till the dust settles." And I kind of gently indicated to him that he might be looking at the wrong end of the train. He's not looking at the engine coming at him. He's looking at the caboose gone past. So, you need to move. You need to start at least.

Jenny George

Yes.

Simon Kriss

And that starting comes with a whole lot of change management, human change management, cultural change management. Now what a lot of people are trying to do is they're trying to shoehorn AI into their existing business model. We call it exploiting AI, right? How can I use AI to be stronger, faster, cheaper, better?
Those that are really going to win are those organizations that start exploring AI. What new markets can we enter? What new products can we offer? Can we radically change our industry because of AI? And when you look at examples like Honey Insurance that completely changed the insurance industry, are simply giving their customers internet sensors that slid under the fridge in the washing machine. That type of stuff is where I think the organizations are going to have to go.

Jenny George

And your background and customer experience is really melding well with a lot of what's happening. What are the cutting edge things you're seeing in the customer experience meets AI world?

Simon Kriss 

Yeah. Some of it is the obvious fruit, giving people more choices in how they self-serve,
using AI to better understand why are we getting this request for service in the first place? Where's the problem that we need to fix earlier in the mix?
Lowering, in one way, lowering the load on staff because customer experience usually comes down to humans doing the job, be that somebody at reception or the checkout or be that someone in a call center or whatever.

And so we're trying to lower their load by having tools around them that make their life easier. The worker in the call center types, no refund, but generally, AI dresses that up and goes out, sorry, Ms. George, on this occasion because you drank the wine, we can't refund the bottle. You know, those types of things.
But we're also inadvertently heightening the load on humans. So if you think about a traditional customer service environment where seven out of 10 contacts were simple, low-level questions, ‘where is the closest parking to the Melbourne Business School?’ Something like that. It kind of gives people a break. And only two or three of the calls or chats are really high emotive load, cognitive load. 

What we're starting to see is with AI taking the cannon fodder work out, that staff now, 10 out of 10 interactions are high cognitive, higher emotive load work. And so we need to give them more breaks. We need to treat them differently. You know, the measures that we have traditionally used in customer experience, like average contact duration and adherence to schedule, they're all being thrown out because AI is fundamentally changing the work. 

In fact, even the people that we recruit, in the past we would have recruited mostly for IQ. Now we're recruiting for EQ, for LQ, their learning quotient, can they continue to learn? And their RQ, their resilience quotient, can they be resilient enough in this world where not only is the workload heavy, but every customer that calls in knows way more than they used to because they've asked ChatGPT. And they've got all this, you know, be it hallucinated or not, they've got all this information.

Jenny George

Of course, that won't just be CX, that'll be across all of our roles because for all those roles will have an AI augmentation possibility, which means that EQ will have more significance.

Simon Kriss 

Much more significance in the future because current generative AI can feign EQ, but it doesn't truly understand emotion, right? And so that is a very human trait. The other human trait is the ability to detect BS. I'll be nice for your show. You know that when you hear it, the heads on the neck of your back, just something's not right about this. 
We're going to need to lean into that because the AI fakes are getting better and better and better all the time. And so it's going to be those human triggers, that human ability to detect fake that is really going to come into play. So I'm not one of the big believers that AI is coming to take all of our jobs. It's probably coming from mine, but you know, and I'd love a future where we all work three days a week instead of five because AI was, you know, efficient, but the reality is there'll be job change, but I don't think there's going to be massive job loss on the kinds of scale that the media would like us to believe. And if there is, yeah, that's likely to happen in developing nations rather than here because all of that low value, simple, repeatable, low transactional, high transactional work, we've offshored. We pushed that out to lower cost locations. So it will be places like the Philippines and Honduras and South Africa and Fiji that are going to feel job loss at a much higher rate than what the developed nations will.

Jenny George

Simon, one of the things that you do is think about the future. Is there a trend that you think people may not be seeing coming and that you think is really important for people to understand?

Simon Kriss 

It's always scary to use the words future and AI in the same sentence. I love it when people try and make a five-year prediction and I kind of say to them, would you have predicted where we are now three years ago? And the answer is no. Look, the biggest prediction that I don't think people see coming is how rapidly people are going to take on a personal AI assistant.
Many people are already moving away from Google search to using perplexity, using Chat GPT for their search type requirements. And that means all of the companies that are spending millions of dollars rebuilding their website should probably be rethinking that investment because AI doesn't care if you've got pretty pictures. It doesn't care if your background color palette is on brand.

It doesn't care about, in fact, it doesn't care about brand at all. It just goes out to do its job. So we tend to build websites today for humans and make them pretty. We have great menus, structures and things like that.

When you look at that in the future in the land of AI, when you say to your AI assistant, please go and get me an insurance policy for my Camry and off it goes and comes back and says, well, you know, auto in general has the best deal. And you say, but I've always been an RACV customer. And it says, I know that, but auto in general, you're going to tend to go, okay, book that for me. So how do companies, they need to start thinking about how is a company going to build brand in that future state where people aren't trolling through websites? And are they starting to make their data easily accessible for AI?

Jenny George

And of course, this is a really interesting debate because on the one hand, the AI engine doesn't care at all about brand in the logos and colors sense, but cares immensely about brand. In fact, some really interesting data that's coming out that shows that AI goes to the global brand first much more than the functional specifications would suggest. So—

Simon Kriss 

Because it gets more mentions. 

Jenny George

That's right. That's right, so reputation in that brand sense is gonna be critical, it's gonna be huge. And particularly the third party endorsement style of reputation. So it's a very interesting change, I think, for brand managers and customer experience people to think about how you bring those together in that AI enabled world.

Simon Kriss 

Yeah, customer testimonials are going to be worth way more than a TV advert.

Jenny George

And making those easily accessible.

Simon Kriss

Yeah. So from a technical perspective, you'll start hearing phrases like knowledge graphs and vector databases and things, because that's the stuff that AI loves to consume. So business leaders should start talking to their CIO and stuff around what are we doing in this space?

Jenny George

Yeah. Well, thank you so much, Simon, for joining us on the In Business with AI podcast. It's been wonderful having you here.

Simon Kriss 

Thank you, been an absolute honor.

Jenny George

We have a little black book here and we ask each of the people who come on the podcast to leave us with your number one piece of advice
So our listeners can head to the MBS website and we'll keep you up to date with the digital version of our little black book. Well, thank you again, Simon, for joining us.
Please do subscribe to this podcast. We would always love to have more listeners and watchers of our podcast each time that we put that out. Also keep up to date with Melbourne Business School. Our institute, the Digital Innovation and AI has many fantastic events. We'd love to see you here, thanks.

 

"My one bit of advice for leaders would be: Approach with humility - be humble. It's not always the best idea to assume your knowledge of AI is as extensive as it could be. Be curiously humble and ready to learn." - Simon Kriss