AI and SMSFs: What trustees should know

Robert Barnes

So how should trustees think about what AI and automation means for their SMSF?

Dr Evan Morrison

Yeah, so at the current point in time, what we’re seeing primarily is the adoption of AI and automation bubbling up at the administration layer. So from a trustee point of view, it’s not quite hitting them just yet. However, they will start to see around the edges of where they can actually get benefit from AI. So one of the things we see from an administration point of view that can apply down into trustee levels is using AI for classification of documentation. So as you receive your share communication, scanning them in through a scanner scanning software and then using AI to automatically classify them into folders on your desktop. And there are tools out there today that do this, that just run on your desktop and that can read the PDFs or read the scans and categorise them appropriately and accordingly. So you don’t have to spend that time. And so it’s all about efficiency savings to just move files around to prepare your paperwork.

Robert Barnes

Are there other ways that AI is changing how SMSFs are managed?

Dr Evan Morrison

Yeah, so what we’re sort of seeing is using AI tools to assist with legislative understanding. And so it’s not to say that they are the ending final advice, but in terms of absorbing the legislation as it comes in and breaking it down, they become really important. So as an example, whenever I see a new piece of legislation myself that is relevant to me, I will use AI to summarise it in a simpler voice for myself. So explain like I’m 10. And then I will say, once I understand it, I will then ask it to explain like I’m 15 and so on and so forth until I get up to the relevant level. But then I use AI to generate questions that I can ask of my accountant or my administrator that are going to be relevant. So given my personal circumstances, and you can explain that to the AI with your knowledge of the legislation, what are the questions that I should be asking to get the best value for my trust or for my fund out of my administrator or my accountant.

Are the efficiency gains transformative yet or are they still quite marginal?

So to date, it has been marginal improvements. So it’s been automation of simple processes or simple activities within the management of the SMSF. Where we see it going though is more complex workflows. And slowly building up that automation layer. And so I think there’s still going to be a need to have the human review process throughout everything you do. However, the preparation stages and just the data curation and collection phases should be far more automatable. And so, which is where the bulk of the time is for a lot of people out there.

Robert Barnes

Do you see AI also being adopted in investment areas?

Dr Evan Morrison

So we’re starting to see the emergence of AI managed fund providers and synthetic ETFs that are based on AI synthetic designs. It is still a dangerous path to go down, especially around the SMSF space because of the requirements of actually not being too risky. We do think that it will help inform decision-making though, in the sense of, think of it as an analyst as an aid. So as you’re making decisions or as you’re thinking through the process, using the tools themselves to assist you with extra analytics or research. So as an example, one of the things that a trustee can do is to leverage the tools to seek adverse findings about a trust or about a managed fund that they’re looking at or a security that they’re looking at. So actually saying, go and search through the news for any adverse findings or things that I need to be aware about, before I make a decision about this. And so it’s not necessarily about making decisions from an investment point of view, but assisting in the research process. In terms of interpreting data, I think that’s probably the next activity on the horizon. And again, it’s not replacing a detailed analysis or a detailed interpretation, But being able to chop and change the datasets, if you have annual reports, if you have reports that you’re running through and understanding what that data means in the same way I described summarising legislation, using generative models to summarise datasets is the next enabler.

And as I say, it doesn’t replace what comes next, but it does allow the trustee to actually work with data in a way that they’ve not been able to before. And so it unlocks new insights effectively.

Robert Barnes

Does AI always make compliance simpler or is there a risk that it could actually make it more complex?

Dr Evan Morrison

I think it does both. It does both. And so on the one side, if we think of AI as an enabler from an analyst point of view, what it means is all of a sudden we have an additional tool to check the work that we’ve done. And so it’s not taking away from the fact that we have to do the work and then we have to review the work ourselves, but we have an additional place that we can have the work validated and checked before we send it upstream. So from that point of view, it does enable us to have more confidence and it removes the backwards and forwards between yourself and the auditor that may go on if you’ve made a mistake through the process or if you’ve overlooked or forgotten something. And so just having a second pair of eyes effectively over your work is really important, number one. And then number two, because it will enable this process, what we will likely see is more stringent requirements around quality. And so as we unlock productivity in quality, the expectation is that we will have higher quality levels. Now, and then the flip side on all this is transparency.

And so this is why I am veering away from advising or talking around actually using it to make decisions on behalf of a client is what we have seen is the need and requirement for AI to be transparent and to be repeatable, which it’s not today. And so until we get to a point where we have 100% confidence in what these models and tools can do, We should never ever think that they are able to answer all the questions all at once.

Robert Barnes

What questions should trustees be asking their providers about how they are using AI?

Dr Evan Morrison

Yeah, look, and this comes, there’s two parts to this question. So the first one is around sovereignty and data control. And depending on what you as a trustee are comfortable with, so long as it’s legal, whether or not your advisor or administrator is allowing data to leave Australia and what happens to the data, number one. And then the other part of it is, is understanding the governance controls that the upstream providers have placed around the use of AI. So as we’re seeing more and more regulation and more and more guidance coming from the regulators, there’s an expectation that transparency is obvious. So, and that there is repeatable process as well as a human review baked into the application of AI. So if you’re talking to your administrator or advisor around the use of AI and they say they’re using it, but then they can’t talk about or talk through the review process they go into and/or how they sample and review what they’ve actually produced with AI, then it’s worthwhile actually asking them to build out those processes so that when the legislation hits around that being baked in, then you’re covered as a trustee because you’ve done everything you need to do upstream to make sure that any decision-making has been done by a human.

Robert Barnes

Does AI significantly increase the risk of serious data leaks?

Dr Evan Morrison

Without a doubt. Without a doubt. So it’s, we’re seeing an emergence of a startup culture of people trying to get rich quick with AI products who cut corners and who are effectively throwing together code there and vibe coding. There’s a phrase, vibe coding, which is using AI to develop software, new products and platforms to do processes or process improvement for clients of financial services. And the trouble there is that in a lot of instances there are security holes in that software and you don’t know whether the data’s being processed. So I mentioned before about asking your advisor, accountant, where their AI is from a sovereign point of view. In the same way, if they’re using downstream tools like this, they should be able to explain to you how they’ve assessed their vendors for the use of those data providers. And so you end up in a situation where your data could be used in training sets in the future. And so nobody wants, when they ask a question about themselves, to have an autocomplete of their tax history coming out of ChatGPT. Number one. But then number two, nobody wants their data to be leaked through sloppy, vibe-coded applications as well.

Robert Barnes

On the flip side, should trustees be concerned if their providers have not adopted AI yet?

Dr Evan Morrison

So there’s still plenty of time ahead. We are, we’re in a phase where people are starting to adopt the technology and they’re still trying to find what works. And so following the leader is actually an okay strategy. Still at this point in time. There are use cases where we’re seeing winning circumstances for advisers and administrators using AI in the sense of getting productivity gains to then lower the costs. But it’s not substantial at this point in time. What I would do is to have a conversation with the adviser or the administrator and understand what the strategy is and if they’re even going to go down that path. And it is a very personal thing as a trustee if you want to go down the path yourself, knowing that as more and more people adopt the technology, there will likely be more and more requirements to do far more data processing because it’s now enabled through practices using AI and whether or not you actually think that you can keep up in that space as well.

Robert Barnes

Any practical tips for SMSF trustees looking to engage with AI in the next few years?

Dr Evan Morrison

Yeah, so the first one is, is when you find the tool that makes sense to you, understand what it does with your data. And so is it shipping it overseas? Is it being used for training? And what’s your comfort level for that? Because it’s your data, number one. But then number two, as you are using those tools, you should be really trying to leverage them as your assistant or your analyst. Not as your decision maker. So you are in control of your fund and your trust. And so as a result of that, you are the one that is responsible for decision making. So it’s not to say that you can’t use these tools. And so you should always apply these tools with a level of, I want you to do this research for me. I want you to be devil’s advocate. So I’m about to make this decision and engage in understanding what it can tell you to include in your decision-making process, but never to actually let them make decisions for you.

Robert Barnes

Can tools such as ChatGPT sometimes be too eager to please and perhaps not be critical enough?

Dr Evan Morrison

100%. So, and this is, it’s very personal to me in the sense of turning off personalisation settings. So the trouble with some of the technology as it exists today is that the providers are providing technology and they’re trying to give it anthropomorphic traits. And through that, it hides deficiencies. So we have a jagged frontier of what’s available to us. So some things AI is really, really good at and some things AI is really, really bad at. And so by treating and making it look like it would feel like a human, we think as humans that it’s not deficient in the areas it’s actually deficient in. So turning off personalisation settings goes a long way to stopping that reinforcement or that sycophancy. And then as I say, number 2 is always asking for the devil’s advocate position. So always running through different personas to try and bring out all the different perspectives that you should be trying to understand when making decisions.

Robert Barnes

And finally, where do you see AI heading in the next few years?

Dr Evan Morrison

So I, look, I do think that we’re going to go down the path of agentic workflows. And so this is not just asking questions. So this is starting to do small activities combined together. So you have a goal that you’re trying to achieve and then what steps you need to take to achieve that goal. And the AI is the one that is starting to make some of the decisions around the process, not the decision. We always come back to not the decision, only the process. But in terms of preparing data and automating the preparation of data, I think that that’s going to be the biggest win over the next 3 years.