Welcome to AI in the Accounting Industry

Rebecca Morgan: Hello, and welcome to your Fireside Chat all about AI and its impact on the accounting industry. Now, would you believe that an aerospace engineer is currently helping tax and accounting professionals implement AI within their firm? Well, it turns out that the precision and systems thinking that's used to get an aeroplane off the ground is also useful in helping professionals launch an AI implementation both strategically and safely, and of course with confidence. Now, AI is that shift in the workforce that we certainly can't ignore, and for that reason, we've actually invited Ethan here, Ethan Glessich from Kognitive, who is actually our aerospace engineer, but recently he's been turning his mind to productivity in professional firms, particularly accounting and tax firms, and of course in the last few years that's involved taking a look at what AI tools are out there and how they're helpful for us. So Ethan has a very practical approach to productivity, and I just want to welcome you to our Fireside Chat, Ethan.

Ethan Glessich: Thank you, Rebecca.

Rebecca Morgan: All right. Now, before I dive on in, I did just want to mention that the NTAA with Kognitive have been working to put together a seminar that's really designed to get into the nuts and bolts of setting up an AI implementation process. But we'll get to that a little bit later at the end of our chat today. A few housekeeping issues that I wanted to address. Please note that if you wanted to download the slides that are going to be used today, if that is something that you might like to have as a bit of backup there, you certainly can download those. There's a specific button for the slides. And the other thing which I'll get you to do right at the end of the session is just fill out that evaluation form. We'd love to hear what everyone has to think about some of the issues that we've discussed today. All right. Well, Ethan, let's get into this. And before we dive into AI specifically, I might come back to the context of your background. And specifically, what led you from being an aerospace engineer through to working with tax and accounting professionals?

From Aerospace Engineering to Accounting Productivity

What led an aerospace engineer to work with accounting firms? [02:07]

Ethan Glessich: Yeah, it was a bit of a journey, moving from, started very specifically in the technological space, working with AI back in 2004, but moved from technology towards people before founding Kognitive in 2012. The challenge or really the opportunity was we started very early and we didn't know what we didn't know, so we entered a business planning competition, and I'd love to say it was strategic that we started working with accountants, but it was more serendipitous than that. So one of the judges in the competition was the manager for CPA Australia Conferences, and we were combining neuroscience and productivity and looking at, "How do we make meaningful improvements?" And she saw some value that would be there for their members. So, we ended up doing — they gave us one presentation at a conference or Congress in Melbourne, and I was a little bit nervous, and a friend of a friend had a small accounting firm and said, "Hey, come in, you can practice on our staff." And both the practice and the event went really well. The one event with CPA turned into 30 plus over the next couple of years, travelling all around Australia, focusing on productivity and neuroscience. And that one little practice run turned into a contract that turned into a much larger accounting firm, which we're still working with to this day. So it was quite by accident. Later on, we realised that accountants had made this connection of time equals money. And if we could help them achieve more time, then that would fall to their write-offs, or their bottom line, or their productivity metrics. So it was a seamless conversation and it just grew from that foundation to today, it's one of the major areas of our business.

Rebecca Morgan: Absolutely. I can hear that you're singing their song for sure.

How did Kognitive's focus shift from productivity to AI? [04:03]

Rebecca Morgan: All right. So obviously you've spent quite a few years in the boardrooms of different accounting firms. So I guess the question now is, we're not just talking about productivity, are we? But rather, you've seen a pivot, or I'd suggest that perhaps Kognitive has done a pivot into looking more into AI tools.

Ethan Glessich: Yeah, it's a good question. So for us, pivot is a strong word, it was more of an evolution. So we internally saw, first thing before we were working with clients with the technology in 2022, like, "Wow, there was this moment of what is this thing?" And we started to play with it. But early on, we really struggled to get this ROI from the tool. It was great to play with, but the return on that investment was very difficult. And little by little, we kept exploring different technologies and it started to grow in our business where we could see, "Oh, this is really useful for us." At the same time, as we were working with clients, there were specific problems that we were working on solving where AI became a tool that could solve it more efficiently than we'd planned.

Rebecca Morgan: Oh, because you were looking for productivity wins.

Ethan Glessich: Right. That's the lens. So it's any tool that surfaces, and AI was that next tool. So a common workflow at the start was we're working with a small accounting firm, this is one of the first initial deployments, about 20 employees, and they had challenges where seniors would be out doing the work, the partners or the seniors meeting with clients, taking all the intel, all the knowledge about the project, but the juniors would then deliver. And there was a workflow process, it was very difficult to do the handover, the quality of the handover, the back and forth, the information communication was a really big bottleneck in the business. And so we could solve that with a little AI assistant that took all the notes, captured all that information, and then effectively created this information for the juniors to work with to do the work.

Rebecca Morgan: And stop the backwards and forwards, or at least minimise that.

Ethan Glessich: Exactly. And so it was one problem, then it was the next problem, then it was the next problem, and little by little, that just became the most efficient way to solve many of these problems in the businesses.

Rebecca Morgan: So it wasn't that you got overtly excited about AI, it was more that you had problems and you could find solutions in these AI tools.

Ethan Glessich: Well, both. So we did get very excited, but that excitement actually dwindled when we couldn't get the ROI. It went from excitement to frustration in the early days. And then it started to drip-feed its way in, and then all of a sudden it became an avalanche in the last few years.

Understanding AI Technology for Professional Services

What exactly is AI — and why is the term so confusing? [06:23]

Rebecca Morgan: All right. Well, let's get everybody on the same page. When we talk about AI, so Artificial Intelligence, what exactly are we talking about here?

Ethan Glessich: Yeah, that's a really interesting question because I know there's a lot of confusion in the marketplace. And I like to simplify AI in one word. And that word is something that you may have even used today with the weather that we've been having. And that is an umbrella. AI is an umbrella. And when I say that, people look at me like, "What are you talking about?" I'm talking about AI being an umbrella term. So just like we have vehicles, and we have trucks, and we have cars, and we have buses, they're different types of vehicles, AI is the same. So there's many different types of AI. In fact, it's been in your spam folder for a very long time, it's in the Netflix recommendation engine, Apple's had machine learning in phones for a decade before their attempt at AI. It's a technology that's very mature, in fact, even in the 1950s is when it first originated, if you can believe it. So why that matters is, if you're reading a study like McKinsey's published a 40% productivity gain in AI, it doesn't necessarily mean they're referring to the technology at the core of ChatGPT, if you will. That technology is a generative type of AI, that's the name, it's generating this content or this information. And at the centre of that is something called an LLM, or a Large Language Model, which you can just call it a brain. It's literally a little brain, like a human brain, it's got artificial neural networks instead of human neural networks. It's literally a brain that is driving a lot of these improvements. Now, that brain is very different to some of the other technology. So it's important that we understand the difference so we can compare apples with apples. And today in this conversation, we're just going to chat more specifically about that brain, if you will, in the context of AI.

Rebecca Morgan: Sure, and that's probably what most people are playing with at the moment.

Key Challenges Facing Accounting Firms

What are the four biggest pain points firms face with AI implementation? [08:12]

Rebecca Morgan: All right. So coming back to you working with firms, looking at productivity wins or looking for productivity wins across Australia, we've got a couple of questions in here to say when you have gone into a firm to implement AI, what are the bigger pain points or the sort of biggest concerns, or pushback, I guess, that...

Ethan Glessich: Yeah. So there's really four main things that surface quite frequently. Maybe your members are finding similar challenges. So the first one is they know AI can deliver some gains, but they're just not sure where to start. And there's a lot of overwhelm out there. There's tools that are evolving so fast. There's an enormous amount of information coming at us. It's the speed of evolution is unbelievable.

Rebecca Morgan: Even I know my own inbox, like there's so many sort of AI sort of design products that are out at the moment.

Ethan Glessich: Right, and that makes it very difficult to navigate from a practical perspective. The other is firms have started to experiment and haven't achieved the ROI. So it might be little things like ChatGPT accounts or Microsoft Copilot accounts for the employees, and people might be using them to some level, but it's not having a measurable impact on the business. Sometimes it's bigger deployments though. So there was one firm that came to us who deployed a little meeting assistant, and six months later, they discovered that this meeting assistant had been attending all of their meetings internally and externally, which is great, right? It's doing its job. But the way it had been set up was all of that information was shared with all staff.

Rebecca Morgan: Oh, that would have made for an interesting Christmas party.

Ethan Glessich: Right. So there's a lot of information in there or processes that firms sometimes have done bigger deployments, like, "Oh my goodness, it's not our speciality, it's not our expertise," like things haven't gone as we expected. The third thing which surfaces quite a bit at the moment is a lot of accounting firms, at least the ones that we're speaking to, are struggling to find good quality staff. And that's a lot of them are short staffed, putting a lot of pressure on the staff that are there to do the work, burnout starting to increase, larger effort. So there's this mindset, "Well, if I can't find these staff, maybe we can deploy this technology, which is kind of like getting more staff and relieve some of that pressure, improve that retention." And the fourth is what I call the compliance catch-22.

Compliance and Security Considerations

What is the "compliance catch-22" and why does it create shadow IT risks? [10:21]

Ethan Glessich: So there's a lot of interest in a lot of the accounting firms, but the firms often take a breath and I think a healthy breath to go, "Wow, there's a lot of compliance we've got to work through first." And so often there's a stop. Now, the reason I call it a Catch-22 is because what often happens in those circumstances is the staff are often experimenting with technologies.

Rebecca Morgan: Yeah, because they want to see, they've been talking to people, they've been at the barbecue and someone's telling them about these amazing tools, so they're starting to download them on their computers, aren't they?

Ethan Glessich: Yeah, and that creates a shadow IT environment. And a lot of people are of the belief that's fine, and depending on how the settings are set up, it can be managed, but when it's in a shadow environment, you don't have control over that. And particularly with new technology like AI browsers.

Rebecca Morgan: Yes.

Ethan Glessich: Default settings out of the bat, if anyone's got a ChatGPT free account, the default settings is to train on your data. So all of that information goes to those providers. And then the browser, where you've got all of your CRM, you've got all of your client details, if you might have your accounting software running in the browser, all of that, every time you're using a — depending on which AI browser you're using, but some of them when you prompt, it will take everything that's on that page and send it straight to...

Rebecca Morgan: And just to come back to that what you're talking about with browsers. So obviously people might have free or paid accounts with say, for example, ChatGPT. But just recently Chat's actually launched Atlas, which is connected to Chat, obviously, and to your Chat account. But, having said that, that's now a new way to browse the internet. It's almost like wanting to replace Chrome or any of the other browsers we'd use. And so what you're saying there is that's just a recent development, but it's opening up security issues.

Ethan Glessich: Absolutely. Yeah.

Should staff be allowed to use AI browsers? [12:11]

Rebecca Morgan: And on that, because as soon as you started talking about browsers, a few questions came through on that. Some people are saying here, "I've been told we shouldn't allow staff to use those browsers." What's your view on that?

Ethan Glessich: It's a complex question. So there's definitely some gains which can be obtained from those technologies. And if you proactively manage the compliance risks, they can be managed. But the cyber security risk is another dimension that these are very new technologies and there are known and quantifiable cyber security risks associated with these which haven't yet been resolved, like prompt injection and other techniques that other parties can take control of the little brain, the little AI sitting in our browsers through some advanced techniques. So it's very — compliance can be managed if it's very proactively and strategically approached.

Rebecca Morgan: So you need to know who's on the browsers, make sure you know what their settings are set at, at least at a minimum.

Ethan Glessich: That's right. But then navigating the cyber security area, it's a very new territory and there's a lot of unknowns. So each firm needs to approach their own risk profile there, but it's new ground.

Rebecca Morgan: And it sounds like the ongoing management and understanding what's happening in that space.

Ethan Glessich: Yes.

People-Centred AI Implementation

Why is "people first, tech second" essential for AI ROI? [13:27]

Rebecca Morgan: All right. Well, coming back to whenever I have a chat to you about AI, and we've had a few conversations now leading up to today, I always like the fact that one of your mantras whenever you're sort of looking at anything is through this lens of people first and tech second. And you were saying the other day that in your view, that mindset really is essential when you're coming in to trying to introduce AI particularly into an accounting firm. Did you just want to talk to why that is so essential?

Ethan Glessich: Yeah, I think it's — I think it's how you get an ROI. So there's a lot of — the technology gets a lot of focus. And quite rightly, it's amazing.

Rebecca Morgan: Yeah, it's the exciting bit, isn't it?

Ethan Glessich: Right. But it's not a switch that you just flick on and the gain is achieved. If it was, we'd all automatically have these gains. Right? The gain, at least at the moment in the way the technology is operating and set up, the gains are achieved when people are using them effectively, when they fit into the workflows of staff.

Rebecca Morgan: Because as exciting as all of this is, they're just another tool, aren't they?

Ethan Glessich: Right. And if we just focus on the tech exclusively and the people we don't need to know about that, there was a firm that we were working with in the productivity space, not in the AI space, just in the productivity space. And we were working on planning and prioritisation and other things, and they said to us, they've got all these Copilot licenses, the directive has come from above, and we should all be champions of AI. They just flicked the switch and we're expecting the results, but that's not how it works. It's a change management process. People need to learn, people need to feel confident in using it, and the way that we support them delivers that or not. And so I think it should be more of a focus on the people and even more so a focus on the problem we're trying to solve then how do we bring the staff into achieve that.

Addressing the Job Displacement Question

Will AI replace accountants? [15:15]

Rebecca Morgan: All right. So I think we need to talk about the elephant in the room now, and it's also a big elephant sitting in the Q&A platform here, because that whole idea of putting people first and tech second has prompted a few questions coming through here. Somebody has said here, "Ethan, most of my staff are worried about AI replacing them, and whenever we have a conversation about AI tools, there's resistance." And also other people are sort of saying here, "Look, I'm worried about AI replacing my team or even myself from an accounting point of view." So, what do you have to say to that? How do you respond to that?

Ethan Glessich: That's a tough question. There's no way to sugarcoat it and I think there's two lenses to look at this. One is long-term, one is short-term. And the long-term is nobody knows. You don't know, I don't know, even Sam Altman doesn't know. Like, nobody really knows how far this technology will go. There's a lot of speculation, but as of today, can AI replace the accounting firm or replace accountants? And the answer is no. It's not possible today. Can it replace specific tasks? Can it replace specific workflows? Can it radically drive up efficiency? Yes. Can it absorb significant parts of roles? Yes. 100%. And more often than not, that's in the more of the junior space where a significant percentage of the roles can be absorbed. But can it replace entirely? No. So, from my perspective, we can think about it in the short-term, at least. It's not so much, "Will it replace us?" It's more, "If there's other firms who are using this technology competitively and refining their offer accordingly, is it, 'Will they outcompete us?'" And that's the practical short-term question. Long-term, it's a bit more difficult, but the reality is if you are at the front of the technology versus being at the back, when the changes are starting to occur in the marketplace, who's going to have the better vision, the better visibility to be able to see the opportunities and capitalise on them?

Rebecca Morgan: Yeah, so this is more about protecting your firm and protecting the jobs moving forward by getting on the front foot here.

Ethan Glessich: And it's what we can control versus what we can't. And if we focus on what we can control and try and utilise it at the moment to deliver the benefits, but then position for whatever is going to unfold in the future, I think that's the way to play. At least that's my personal view.

Rebecca Morgan: Yeah, and I think that view definitely sells it for say the owners or the principals of an accounting firm, but I guess it doesn't allay the anxiety that perhaps individual staff members or more junior staff members might have.

Managing Change and Staff Adoption

How do you bring resistant staff along on the AI journey? [18:00]

Rebecca Morgan: So we've just got another question that sort of comes through on that, and one of them says, "I hear you, Ethan. I agree, AI is the way forward to progress a firm." He goes, "But I've got two different types of staff. Some are excited and others are just point blank refusing to engage with AI." And I guess their question here is, "What should I be doing from perhaps a leadership perspective here? Like how do you deal with that?"

Ethan Glessich: That, yeah, that's a good question. I think it's difficult because the technology is evolving so fast. So in 2023, 2022, there was so much going unknown that I think a clear stance was, "We'll wait and see." But in 2024, 2025, it's evolved so much now that there's no stopping it. It's coming through like a train. And people are using it regardless. So my view is we need to lead it as the executives, as the leaders in the firm. It needs to be led, it needs to be strategic. It can't be staff will go and play, because if you think about it in an accounting firm, what we generally see, and you probably see the same thing with your members, is there's almost like an upside-down pyramid from tech expertise. So the more senior you go, it's generally people that know accounting ridiculously well, know tax ridiculously well, but are less competent or less experienced with technology. And juniors, it's the opposite. They don't know the tax and the accounting regulations as well, but gee, they're a wizard at technology.

Rebecca Morgan: Their mindset's probably just not there.

Ethan Glessich: Very different. Very different. So there's this mismatch. And I think now is the time where even if we don't address it, we have these shadow IT issues starting to creep in that we mentioned before that can create some bigger challenges. So my view is it needs to be led, but it needs to be a journey.

Rebecca Morgan: Yes, you're not going to instantly solve it in a week, are you?

Ethan Glessich: That's right. And you can create an environment, in fact, this is what we do. So one thing that we personally find really useful is we frame it as an experimentation, and we set it up in sprints. Similar to some of the work we're doing with NTAA is we're working on these initiatives and experimenting to see what's going to work well, and supporting people on a journey. They come back together, share what's working. You get this cross-pollination of information in the room. Some people are really connecting with voice-to-text technology, some people are really connecting with meeting assistants. And then they share their insights, and what happens the next week?

Rebecca Morgan: Yeah.

Ethan Glessich: They're both using the opposite sides of the technology, and little by little, as you create those opportunities and structure the journey, that's how I think we can bring people on board, whether they're champions or whether they're sitting back, it's that journey.

Rebecca Morgan: Okay, that's helpful. Yeah, that's something to definitely think about, isn't it?

Strategic AI Implementation Framework

What is the 3M framework for AI implementation? How does it relate to AI strategy? [20:53]

Rebecca Morgan: All right. Well, let's get practical now talking about implementation, because like you said, we need someone to lead it, and it needs to be with all of the staff within a firm, no matter where they're at with adopting AI. So as we said, firms, individuals, staff, they're all experimenting with different tools. So you've got ChatGPT, you've got Claude, Perplexity, a whole host of other various AI assistants. So in my mind, the question is, what comes next after you play? And in fact, I've got a question here saying, "Ethan, I think at the moment all I'm doing is playing with cool AI tools, and I'm having a lot of fun and having some wins, but how do I go about getting a clear strategic approach?" And I think that's the million-dollar question, isn't it?

Ethan Glessich: Yeah, and I think I don't want to diminish the importance of playing and experimenting. I think that's fundamental. It's learning and just it's hours in the seat communicating with these models gives you an experience and an insight that you can't get any other way. So that's really important. But if that's the only thing, and I should say footnote, once the compliance and security set up beforehand, but if that's the only thing, then it's very unlikely to deliver an ROI. And so an ROI focus for us is we're focusing on, first of all, what's the problem, what's the outcome we're trying to achieve, and then what's the most efficient way to do that. But then we use 3Ms as a bit of a framework to really navigate our way into the implementation. So the first one is to map the workflow. And this is something that we did long before AI. It's understanding that workflow through the business. Whether that's that knowledge information meeting workflow we spoke about before, whether that's an individual, whether that's in reconciliation, whatever it may be, that particular workflow, mapping that so we understand it. The second is matching the tool to that workflow. And that may not be the newest version with all the bells and whistles. It's the appropriate tool for that particular workflow. And the third thing is arguably the most important, and that is the measurement. We want to set up the measures, and not just retrospectively, but up front we can look at that workflow, we can look at the technology, the investment required to set up that technology, and we can, okay, what's the gain here? Because sometimes that equation doesn't make sense. It's just the investment's too high for the gain we're looking at. And if we come at it with that 3M mentality, we can know that up front. And as we're implementing, we can then track our progress along that journey to confirm that we're actually achieving what we want to achieve and realise those gains.

Does every accounting firm need a different AI strategy? [23:22]

Rebecca Morgan: Okay. So by hearing that then, I'm assuming every firm you walk into or every firm thinking about implementing AI is going to need to have a different strategy.

Ethan Glessich: There's a little more nuance to that. So yes, that is true to a point, but no, there's also, so in our experience, there's a common pathway that we move through that allows us to use the technology, and we've used that same approach for the majority of accounting firms that we work with.

What are the six types of AI tools — and which order should you implement them? [23:47]

Rebecca Morgan: So this is the strategy you use?

Ethan Glessich: Exactly. And we're going to go through it in more detail in the seminar, but I've got some slides. So first what I'd like to do is just chat about the market and how it's evolved from a technological perspective, as it relates to accountants. Because there's a lot of different things that are happening outside of this, but I think from a practical perspective of what's the market what's evolved for me, and then we can look at strategy. So, how do we approach that if that works?

Rebecca Morgan: Amazing. Yes.

Ethan Glessich: Okay. So if it all started with ChatGPT, obviously, and these large language models. And that was the first place. Okay. Now, very quickly from that moment, companies like Microsoft said, "Oh, great, we can take that and embed that into our Microsoft 365 technology." And that went from these chat applications into copilots. Even more recently, we see the evolution of that, and it obviously went through Google very quickly, but even now in the accounting space, so with Xero, we've got Jaxs, MYOB, there's various tools, or QuickBooks are releasing agents. So there's this integration of these copilots across productivity and accounting suites.

Rebecca Morgan: And the way I see a copilot, it's a little AI friend that's been embedded in software that kind of follows you around.

Ethan Glessich: 100%. Yeah. It's usually sitting in a sidebar, a particular location. And it's got a lot of context of what you're working with in that particular tool, or the documentation that you're working with.

Rebecca Morgan: Yeah.

Ethan Glessich: So we went from chat to copilots. Stepping up from there, we moved into this — a lot of companies started to move into this stand-alone technology. They went, "Oh, really interesting brains. A lot of entrepreneurs took these brains and then started to use them to solve some problems." That then led into these knowledge centres. So, knowledge centres is just basically a data — our data, or some kind of data, that we put a little brain over the top. Yeah. A little LLM over the top, and then we can interact with that data in some meaningful way for us. The next evolution was then automations and agents. And this is much more recently in the last few years is really started to gain —

Rebecca Morgan: Everyone's talking about agents at the moment, aren't they?

Ethan Glessich: Right. Yeah. And then finally custom builds. So that's where you take the technology and you build something that's bespoke for you. Okay. So they're the six main evolution that we've seen in the industry.

Rebecca Morgan: Okay.

Ethan Glessich: Now, what does that mean from a strategic perspective? The approach that we take is not one to six.

Rebecca Morgan: I mean, that would seem logical, but no.

Ethan Glessich: It would, but no. That's the way that you play with it. If you had to start in 2022, you would have been playing with these things in that order.

Rebecca Morgan: You would have progressed as the technology progressed.

Ethan Glessich: Right. But that's not the way we approach it strategically now. The way we approach it now is we map these things on axes of importance or effort and value.

Rebecca Morgan: Gotcha.

Ethan Glessich: Okay. So, we can see, for example, ChatGPT's doing quite a lot of value, but it also takes quite a bit of effort to get the value from it. And that's trending higher effort because it's getting more complex, it's getting more advanced. It's moving in a direction which is more difficult. Where some of the others you can see like knowledge centres or some of the stand-alones are much lower effort, and the value is really high.

Rebecca Morgan: And probably cost.

Ethan Glessich: Right. Yeah. So this map is the strategic map for us. Okay. And we start at the left-hand side. We look at, okay, what are the quick wins, what are the implementations, and it's usually in the stand-alone space, the plug-and-play technology, where we can get some significant productivity gains in — we're talking in weeks, sometimes months, you know, taking people on that journey.

Rebecca Morgan: And they're the sort of products that you can quickly install, and they're designed for the end user to be able to intuitively work it out.

Ethan Glessich: They solve specific problems, and the great companies that have done it have done it in a very intuitive, easy, and natural way. Okay? And that's where we start, because people feel their way into it, they get immediate gains. And from there, it's small investment, significant returns, small investment, significant return, small investment, significant return. So, by the time you get to having a conversation about agents, people already understand the technology, they're already on top of prompting, they're already understanding and getting some gains, and they can take those gains and reinvest in optimising things significantly.

Rebecca Morgan: And then your staff are effectively trained to a level they need to be to start using those other mechanisms.

Ethan Glessich: Bingo. Yeah. Once you start to get more confident in utilising prompting, you can then use those knowledge centres. You can then — all of those transcripts that you've been creating from your meeting assistant, you can use that asset to build your proposals or other things in either a manual, an automatic, or a semi-automatic way.

Rebecca Morgan: And hopefully you've sort of developed your team, too, that they're sharing this information, so there's sort of continual learning as well.

Ethan Glessich: Yes. Yeah. So I think at the SME space, particularly, it's very difficult to revolutionise the entire business around AI. And I know some firms are doing that, and some believe that is the best approach. I can see that it has value, but it also has a high risk associated with it. And it can put you in a position to then pull away in the marketplace, but I think small step-wise improvements is where we're at at the moment with the current technology and the current market as the most low-risk, high-reward approach. So, it's moving in that left to right capacity.

How has the approach to AI implementation changed since 2022? [29:16]

Rebecca Morgan: Okay. And that's sort of why you've got that strategy, or that strategic approach there. So, another question's just come through now, because obviously it's clear you've been sort of growing with this technology while you've been talking to accounting firms about productivity. And the question that just came through is, "Have you had to change your approach?" So, obviously AI really for the past three, four years has started to evolve. So, what's the difference in your approach when you were talking to an accounting firm say three or four years ago as opposed to now?

Ethan Glessich: Yeah, the — I think two things have really started to evolve. So, first, the technology was ahead of compliance for a long time. It was like, "Wow!" How does that relate to the Privacy Act? And now we've got a lot of advice and a lot of structure to really start to navigate that. So, we can get on the front foot of that really quickly. I think early on it was more about experimentation and just success in the early days was, look, like, we could do something cool with ChatGPT. Yeah. And that the reality very early was what it was. Where today we can get really strategic, and we can even run parallel streams. So, you can start to get the team moving while we're mapping workflows at the same time if you're wanting to really accelerate. And I think the difference is, a lot of firms have pulled ahead with AI, but at the same time, they've also had to invest a lot and make a lot of mistakes along the way.

Rebecca Morgan: Because no one really knows what we're doing, are we? We're trying to make sure we're protected, but we're trying to also advance.

Ethan Glessich: That's right. So, they're not as far ahead as I believe they think they are. Because now there's a roadmap. Like, we — it's become a lot clearer that this is how we can navigate really quickly to get to those gains. And if we do it strategically, I think we can leapfrog really quickly.

Rebecca Morgan: Yeah, which is good for people who are going, "Gee, I've had my eyes closed for a while here, but now I know I need to start looking at this."

Ethan Glessich: And that's an important lens because I think about this AI piece as like a wave. I'm not a surfer, I'm a paraglider, but we have waves in the sky as well. And it's like this swell that's building, building, building. You can feel it getting bigger and bigger. There's this like AI wave that's coming, and there will be a moment that will be too late to start to paddle out and to catch that wave. That moment hasn't arrived yet, but it's accelerating. Like, the last year or two, there's been this real shift with these reasoning models and agents where this wave is now starting to accelerate. So, it's — for me, it's now is the time to really dive into the water, so to speak.

Rebecca Morgan: Got you.

Quality Assurance and Accountability

Who bears responsibility when AI gets it wrong? [31:53]

Rebecca Morgan: All right. Well, do you have a moment, because there's a lot of questions that are coming through. I might just pause and see if I can throw a few at you, because you've definitely prompted some thinking out there. I love this one. In fact, this is a theme that I, you know, there's quite a few people that are coming through, and it focuses on quality and accuracy assurance, I would say. In a nutshell, someone's just asked, "Ethan, who bears responsibility when AI gets it wrong, particularly in the tax and compliance space?"

Ethan Glessich: It's always the human. Yeah. And in fact, there's particular requirements for human oversight. It's always the human. Yeah. The, and that's really important because one thing these models haven't yet solved is the hallucination. Now, hallucination is a technical term that they use. It just means they deliver things that sound really confident and accurate, but it's not.

Rebecca Morgan: Yeah.

Ethan Glessich: It's not a fact, it's not actual.

Rebecca Morgan: Because it's pulling information and almost trying to people please you, isn't it?

Ethan Glessich: Right. We're going to go into this more in the seminar, but in essence, the models are predictive algorithms, and they've just got an enormous data set. So they're trained and they sound fantastic. But generally speaking, the current hallucination rate varies a lot between different models, but to give you an idea, like 60 to 70% of the time, when you're just asking a basic question and getting a basic response from one of these models, it's getting it right. 20, 30, 40%, depending on the models and the environment, is wrong. You think about that. Now, there's ways to mitigate that risk, we —

Rebecca Morgan: Yeah, I was going to say there's different prompts and knowledge centres and the like.

Ethan Glessich: That's right. There's different technologies we can use, and in some contexts, you can get it down to below 1%, or even a fraction of a percent. But just as a general prompt and response, the this is still a major challenge that the industry is facing, and it's embedded in the systems. So the oversight requirement is really important.

Rebecca Morgan: Yeah, and I always sort of say to people myself, and I don't know if you agree with this, that you've become the reviewer. So, the AI might be able to give you some raw data, but it's still your responsibility to apply your professional knowledge to make sure that you're comfortable with whatever you're actually producing, whether it's for a client or something internal.

Ethan Glessich: 100%. At the end of the day, it's your responsibility. Now, in the interim step, there's a lot of ways that we can bounce between so that models can find problems with the outputs of models and reduce our review load.

Rebecca Morgan: Stack them up against each other.

Ethan Glessich: Exactly. Exactly. Like Game of Thrones. Yeah.

Client Communication and Professional Ethics

Do firms need to tell clients they're using AI? [34:01]

Rebecca Morgan: There can only be one winner. Perhaps, who knows. All right. What else have we got here? This is another common question that's coming through, client communication. So, obviously if we start to use AI in any of its forms within our firm, a lot of people are saying, "Am I under an obligation to advise clients that we've used AI?"

Ethan Glessich: Hmm, this is a really good question. So, the, from a perspective of privacy policy, compliance with the Privacy Act or other body obligations, there are certain requirements that we must undertake. If we're using technologies in the room like meeting recorders, or meeting assistants, etc., there's legal obligations, borderline moral obligations from that sense. If we're polishing little bits of text here or there to make improvements, it still also gets a little bit grey, because from a privacy perspective, we can't necessarily just remove the names and put that into a model, and that's considered, "Oh, now it's anonymous." Because if you think about it, these models have an enormous amount of data, and they can put things together in a way that no human can. Yes. So, the way that we think, "Oh, it's anonymous," doesn't actually mean it's anonymous for the model. And the model can join the dots in ways that we didn't expect. Yes. So, from an optimisation perspective, I think it's general common courtesy, but there also are legal requirements that we must fulfil in particular context depending on how it's being used.

Rebecca Morgan: Yeah, so we need to tread lightly. And I know I've seen people like someone said here, "If I'm just using AI to define wording or perhaps analyse numbers," they would say, "Surely that's just a normal tool, and I don't need to disclose that." But what you're saying is where you are —

Ethan Glessich: You might think that's the case. Yeah, yeah. But particular, OAIC has defined that it's greyer than it appears on the surface. It's not as simple as, "Oh, I'm going to remove this data and now it's anonymous."

Rebecca Morgan: So it's definitely something I think from an ethical point of view for the profession as a whole, we probably don't have the answers on this.

Ethan Glessich: I think the solution is we get on top of our compliance first. We update our privacy policies, we cross all of our Ts, we dot all of our Is, and then we're in a position to proceed with confidence versus not doing any of that, and we'll just experiment and what are we going to do. I think if we make those decisions up front, we move through that due diligence process, and it doesn't need to be a massive process, but we get on that first, then we're in a position to use the tools with confidence and not have these situations.

Rebecca Morgan: Yeah, okay, that makes sense.

Choosing the Right AI Tools

Should firms use general AI tools or accounting-specific systems? [36:38]

Rebecca Morgan: Another question that came through is about making choice about tools. So the question was, "Is it better to use general tools like ChatGPT, or should we just be looking to invest in accounting specific AI systems with tighter controls," and just whether Ethan has any views on this.

Ethan Glessich: Yeah, I do. They might be a little bit controversial. Okay, so —

Rebecca Morgan: You're in a safe space.

Ethan Glessich: Yes. Right. So, I think about it like this. I'll share my thinking. These major producers of LLMs or this technology, OpenAI, Anthropic, Gemini, Grok, etc., they are at the peak of this technology. They are attracting world-class talent with ridiculous salaries. You might have seen Meta's recent spree of poaching staff, offering tens of millions sign-on. Like, it is ridiculous. It is out of control. But the talent and the quality of the tools that they are developing, consequently, is second to none.

Rebecca Morgan: Are you suggesting we follow the money?

Ethan Glessich: What generally happens in my experience is that at that place, you get amazing technology doing amazing things. Then you get a lot of firms that start to bolt on the technology, and they don't have that same talent pool. They simply cannot capture that talent pool, and the quality of those implementations is unable to be deployed at that same level.

Rebecca Morgan: Interesting.

Ethan Glessich: The second thing that we observe is when you bolt on a technology, we saw this with Apple, it often struggles. And you can see new players coming into the market who have built solutions from the ground up, focusing on AI first. We don't see that in Australia in the accounting space yet, but we do see it in the US. Okay. So there are other countries in the accounting space where there are players who are delivering this sort of technology. Not yet in Australia. Will it be a sufficient size market, or are there bits and pieces is yet to be defined. But for me the key question is, it always comes back to, "What's the problem we're trying to solve, map the workflow, what's the best tool to solve that?" It may be the reconciliation capabilities inside of an accounting software is the most relevant tool for that particular workflow, or it may be an alternative tool can deliver a superior outcome. You'll only know when you go through that 3M process.

Rebecca Morgan: Absolutely. And a few people have said, because there seems to be a thing, "Oh, but surely accounting specific AI systems would have tighter controls." But I'm guessing your answer to that would be, not necessarily. Like, you've got to do an analysis on each one.

Ethan Glessich: Yes, and yes. I think it's a fair assumption. Yeah. But what are they say about assuming?

Rebecca Morgan: Oh, well, they're just assuming that if you use accounting specific AI systems, surely that would have tighter controls.

Ethan Glessich: Yes, no, I mean, when you assume, you make an ass out of you and me. Right? Because it's an assumption, and the reality is we need to go in, we need to have a look, and this is the same for any AI tool we use. In fact, it's the first thing that we always recommend. It's kind of a takeaway from today if anyone's like, "Oh, where do I start with this?" If you're experimenting with any tools, if ever you've installed or downloaded a tool, go into the settings, have a look, and make sure that training of the model is turned off.

Rebecca Morgan: And they often phrase it in really friendly ways. Like, "Help us, help it for everybody and yourself."

Ethan Glessich: I feel mean when I turn it off because they say, "Will you help everyone?" And I'm like, "No." Exactly, right? But the crux is that's the default setting for almost every tool in the marketplace. And we have to go in and have a look and manually change those things regardless of the supplier. And that's the same if it's a — we can hope that the accounting suppliers do it more effectively, and of course, they will, but the solution is it's our responsibility to go and check those terms and conditions.

Rebecca Morgan: We can't blindly trust any particular provider.

Ethan Glessich: That's right.

Practical Applications in Accounting Workflows

Which workflows benefit most from AI right now? [40:46]

Rebecca Morgan: Fair enough. Another question, probably coming back to perhaps implementation and choosing the right direction when you start to implement AI within a firm. This was a question: "What kinds of accounting workflows are most improved by AI right now, and which ones should we still keep, in your view, entirely human?"

Ethan Glessich: Really good question, and it's evolving and changing constantly, particularly with agents. You can now — your imagination is the limitation. So but generally speaking, across the industry and the workflows that we're optimising, it's generally at the lower level. So, workflows around the reconciliation, bookkeeping, admin, a lot of organising of meetings. And you can have an AI EA who can organise and send emails and structure your calendar and organise all of these logistical lower level workflows is where a lot of the value is. Also in writing the reports, or reviewing, or drafting the reports, proposals, etc., there's a lot of workflow and a lot of optimisation that can be done in those spaces. We're seeing a lot of gains in those niche workflows. As you level up, and this is why it can be difficult for the junior staff, because they don't know what they don't know. And you need that sort of senior level of knowledge to know, "Oh, that's a hallucination," and "That's factual." But as you move up to the interpretation and even the generation of the financial statements, there's some interesting things that are evolving, but it's really not at a level yet that I would have the confidence to deliver the outcomes. So, it's really in that higher level where it's really, really important to have the human driving.

Rebecca Morgan: And again, I — from that, too, the human's always going to be the reviewer, isn't it? Whatever output you're looking at, we have to be confident that that is correct.

Ethan Glessich: Yeah, the — I mean, humans make mistakes as well, and they slip through review processes. But I think we can't lessen or slacken our review processes just because there's a model in the middle now. I think we need to increase it, or at least maintain it where it is, get confidence and trust, so then we know, okay, what level of review is appropriate.

Rebecca Morgan: Got you.

Next Steps and Learning Opportunities

What's the key difference between this fireside chat and the AI seminar we have co-produced? [43:23]

Rebecca Morgan: All right. Well, I think I want to come back to something we said right at the beginning. And that was you've got some great insight into giving us the strategy and approach for AI implementation, but as I suggested right at the beginning, we've been working with Kognitive and the NTAA to look at putting a deeper seminar together where people can start to actually take those first steps and have a bit more confidence and detail around those first steps there. For now it's a co-branded seminar. It's going to be an online seminar that will be available soon. But I guess for someone who's thinking about what you're saying here and deciding whether or not it's worth attending that seminar, what's the key difference between the conversations we're having today as opposed to what we're putting in that seminar?

Ethan Glessich: Yeah, it's hands-on. There's no fires. No fireside chat. No fires, or maybe a fire. But it's all about hands-on implementation. There's really three things we're going to go through. So, the first one is understanding the technology. And this is like the aha moment, where they go, "Oh, it works like that! It's not this sort of unknown amorphous thing anymore." It's like, "Okay, it's a little brain and here's how the little brain works." And that makes it very relatable and easy to understand the core of the technology.

Rebecca Morgan: And that would be helpful even for someone who hasn't touched any of these AI models, or even for someone who has been playing with them.

Ethan Glessich: Yeah, because often playing with it doesn't mean you understand what the actual brain is. And when you get that, it's like, "Ah, okay, now I can sort of interact with it in a more effective and efficient way." So we're going to start there. Then we're going to go through those six key points. Well, actually, the five, the custom builds is not so much in the SME space. That's right. So we're going to see and we're going to spend more time in the quick wins, those practical gains to get people confident with the technologies and they can install and follow along in the journey and literally like it's a workshop. And then finally, we're going to touch on compliance and the privacy piece. So they've got confidence, "Oh, okay, I can start to navigate this process and get on the front foot of it, so I can begin to play with these technologies." And this isn't a, you know, something that we've pulled together. This has been evolving over years of work. And each of these workshops that we've been doing with accounting in the room, all of those learnings have come through and evolved into what we're going to be going through.

Rebecca Morgan: And I know speaking to you on this, the process that you go through, you've seen it work in accounting firms and you've seen it bring teams together at all levels of an organisation. So, valuable there. Well, certainly if people are interested in registering for that seminar, we obviously you can do that through the NTAA website. And at the end of this session, we'll also have a QR code that you can find a link to register for that. I know personally, I've got a lot out of a lot of those learnings that you're talking about, and it really does help bring everyone along for the ride. All right. Well, I think I've got one last question for you, Ethan, before we conclude because we've got a lot of information out of you. So, if members or viewers can take away just one thing, one thing from our conversation today, what do you think, or what would you like that to be?

What's the one thing firms should take away from this conversation? [45:36]

Ethan Glessich: Hmm. It's a big question. Yeah. Exactly. I don't know. I think that the technology is here, it's moving, and we've all got to work through our own risk profile of what we're going to do at what time speed. And there's — for me there's one risk that's not in any of these compliance documents, but I think it's actually one of the largest risks that's here at the moment. And it wasn't here in 2023, it wasn't here in 2022, but it's here now. And that is the risk of not taking action. Because it's a freight train, it's a wave, it's a swell, it's coming. There's no way to stop it, there's no way to avoid it, and if we're doing a risk analysis of the different pathways and how we want to pursue, I think we need to include in that analysis the risk of doing nothing. Because with all the momentum, that in my opinion is the greatest risk of all. And that's why we've so heavily invested in driving this space because we want to be at the forefront so whatever happens in the future, we have options to navigate and ride that wave.

Rebecca Morgan: And as you said, it's not too late. Like, we're still at the beginning of this, aren't we?

Ethan Glessich: Yeah.

Conclusion

How can I register for the AI seminar? [47:06]

Rebecca Morgan: Now as I mentioned before, we now have the QR code up on the screen where you can register for the seminar. And that seminar is all about implementing AI in your practice. So you could register via that QR code. Alternatively, you can also register via the blue button, which is just below the viewing platform. Also, we'd love to know what you thought about today and the insights that Ethan in particular has shared with you. So if you could take a moment to fill out that evaluation form, we'd certainly love to hear what you thought.

Rebecca Morgan: Ethan, thank you so much for coming on board today with us in this Fireside Chat. I loved hearing about your insights in this area and I'm really looking forward to seeing the detail coming through on the seminar that you've been preparing.

Ethan Glessich: You're welcome, Rebecca. Looking forward to it.

Rebecca Morgan: Wonderful. Thanks everyone. I hope you have a lovely day.