Hello, my name is Gillian Bowen and this is Small Firm, Big Impact.
If you can help every single person in your business, say half an hour a day or 45 minutes a day or an hour a day, it might seem like a small number, but when you scale that across your entire business or your entire team, that dramatically increases the efficiency of the organization. Treat your AI model or your tool like you would treat an intern or a grad.
When I'm saying you can do a particular use case when you're explaining it to the model, explain it how you would explain it to an intern or grad, you're going to get a really good result.
It's the podcast giving Chartered Accountants the up to date information they need to do their jobs. Each episode, I share resources, tools and expert advice provided by Chartered Accountants Australia and New Zealand and a range of people across our profession. So get following the podcast in your favourite podcast app. Let's start a conversation. Today. It's a big welcome back to Kayur Patel CA for part two of our
discussion on generative AI. Now welcome back. First of all, Kayur thanks for coming back.
It's good to be back. Thank you.
Part two, as we promised, is focused on use cases for accountants. And why you might be asking. Well, all of that is included in part one. So go back and listen to that first. But in summary, GenAI and getting across it is all about efficiency and productivity and adding value to your individual work life, but also your business life and the life of your clients. And what you'll learn in this discussion is that starting small now
is the key to bigger things down the track. So Kayur, just quickly, before we dive into the specifics of the act or, you know, action stations, as I like to call it, um, why do that? Why focus on small things now rather than waiting for the next big shiny AI tool?
Yeah, it's a good question. And, you know, I've been helping clients, um, understand GenAI and then implement GenAI, a safe AI in their, in their environments for the last few months and by far and away the biggest piece when it comes to helping educate organizations around AI is how do I make sure that I get efficiency now? Because everyone's waiting for the big shiny thing, the thing that's going
to say one silver bullet. Um, but what it's really easy, is really easy to miss is the fact that if you can help every single person in your business, say, half an hour a day or 45 minutes a day or an hour a day, it might seem like a small number, but when you scale that across your entire business or your entire team, that dramatically increases the efficiency
of the organization. And that is possible right now with tools available right now, as long as you use a safe, enterprise grade, secure what I like to call safe AI option.
Mhm mhm. Yeah that is a key part. And we'll explore that as well. Um and if we don't explore it we'll put information about that in the show notes. Um so what I was thinking and you and I have had a chat prior to this because, you know, planning is key to a quick and good discussion. We're going to cover three areas. So area one or part one is going to be text based use cases. Then we're going to have a look at data based use cases. And then we're going to wrap up by having a
look at a couple of bespoke use cases. And as I said we're going to try and do that in 20 minutes or less. So first of all Kayur, how can I use text based generative AI right now in my accounting firm?
Yeah. Okay. So the first category is a text and then some data stuff as well. And for each of these categories, um, I might just reinforce one comment from the previous episode, which is remember that these AI models, these GenAI models, you have to treat them differently to like a search engine. Right? Because they're built they're built
based on neural networks. And the example we used last week was, you know, when a baby is learning how to speak, they might look at their dad and say, um, dada, and then everyone will make a huge deal about it. And so that will get reinforced. That data is that person if they say, mama, um, nobody will make a big deal about it. And so then they'll know that that's not reinforced. And so that's how these GenAI models learn.
And so as I'm going through these examples, uh, with your listeners, I would encourage everyone if they take nothing away from today's podcast, but this, um, treat your AI model or your tool like you would treat an intern or a grad, right? So when I'm saying you can do a particular use case, when you're explaining it to the model, explain it how you would explain it to an intern or grad, you're going to get a really good result rather than just typing it into a search engine.
And equally, you know, if you gave a piece of work to an intern or a grad, you wouldn't expect that they would give you a piece of work that was, you don't have to look at it, you just send it straight out to the client. Um, you would want to review and check because that's how that's how the process goes to make sure that we tick the boxes and that we're providing good work for our clients. The same thing with these AI models. Treat it like an intern or grad.
Such good tips. I love that such good tips. Okay, so.
So with that in mind, pretty much any text based thing that you get an intern or a grade to do, you can get an AI model to do so. Here's what I is really good at from a text based thing. First category you can do is very good at creating documentation, creating text. Right. So I use it right now to draft articles for me to draft documents, to draft emails, communications. Um, I even get it to convert text that I write
into my voice, like my actual voice and my video. Very, very good at doing those text based things, provided you're using specific prompts that, um, are engineered in a way, as if I was having a conversation with an intern. So I'm giving it the right background. I'm giving it, um, a conversation rather than just a one liner. Do this for me. I am providing it with other examples of documentation or email that I've done in the past, so it learns how to sound like me or how I
want it to sound. Those are all things that I would, um, provide to an intern or a grad when I get it to write my emails or do text based documentation.
So my brain's ticking away already. So just quickly to confirm, in my understanding say, for example, I paste into ChatGPT, for example, a version that is that I've paid a subscription for or that is just that that's legally protected. Um, I put in some emails that I write and then say, can you create an email that is written as if I wrote it to blah blah about said topic, and it would bring it up?
Absolutely. And it would. If you've given it content like your previous emails, it's going to sound much more like you. And then if you were to provide it with more context, so explaining some background about who the recipient is and how much background knowledge they have, and whether you want the email to be formal or more casual or how detailed you want it to be, it's going to do even better. Just like it would be if if, um, if it was, if it was a human.
And so what you would do is you would write, you're engaging with it, aren't you? You would say, okay, that's a good first attempt. But, um, if I give you this additional information, how would you rewrite that? And then it'll produce another version.
It's very much an iterative process that you hit the nail on the head. It's very much an iterative process.
Great. What else?
So also in text. So um, outside of the creation bucket, it's very good at improving and transforming previously written text. So for example, you might have a piece of advice, um, to go out to a client. I think in the last episode we talked about fringe benefit tax. So maybe it's on how fringe benefit tax relates to their motor vehicle pool, for example. And you might think that the technical information is correct, but you want to craft it
into a nice response. You can copy and paste in what you've written, some bullet points of the technical information, making sure you've got the right you know, the correct answer, and then get the model to be able to convert that into a nice email or notes to file or document or whatever it is that you need to send out the door to the client. I do this all the time, and it means now my emails to clients
or whoever else it might be. I'm literally writing bullet points, and then I'm using AI to convert it into a nice email. And you do that for every email that you send that needs to be, you know, professionally worded, that starts to stack up into some serious time saving. Also very good at transforming text from a perspective of, you know, if you've got some generic text, get AI to convert
that to a first person text. So it reads as if the recipient is it's talking to the recipient rather than talking about something um in general, it's very good at that and then also very good at taking some specific facts and then applying them to a template that you might
have in your organization. If you've got a template that you use for a specific type of report for clients or whatever it might be very good at, then taking some specific details and putting them into your templates in the right way so that it reads well.
All right. So are there any other text based use cases that you want to tell us about before we move on?
Yeah. So the other category I think is summarization. So AI is very, very good at being able to summarize massive amounts of text or documentation into an easy to understand summary. So where do I use this? Anytime I'm going to talk to a client, um, I will get documentation that is relevant to them and summarize it down so that I understand it so I can be proactive when I talk to them. So that might be recent tax alerts from Inland Revenue. That might be latest industry
information or documentation. That might be documents that I found from the client's own website. But I will use it every time before I go and have a conversation with a client. And that's just one example. But its ability to summarize large documentation into something that's easy to understand is brilliant. Everyone should be doing it.
So it's teaching you, it's upskilling you before any sort of meetings or interactions that you're going to have with the client.
Absolutely. And to the extent that I recommend to our own graduates and interns before they look at a new technical piece or information or topic or client industry for the first time, they should be doing that for the relevant documentation, for whatever that technical thing is, or the client or the industry every single time. It will just help them be a) more proactive, but b) just understand that base level of information to help get them up
the curve quicker before they get in. And then do the work.
How much time do you reckon that alone would save, um, a person or yourself? Instead of having to go and find everything, read it all and then write your own stuff.
Oh, I mean, for me, on a weekly basis, I save hours and hours just using that one set of use cases, the summarization.
Um, wow.
You know, like a recent example, um, BEPs pillar two, the big documentation. I know that I didn't need to understand every single thing, but I needed to get a general idea of what it was about and when it came into force for various balance dates. That was something that I would have had to read through and kind of sift through and take quite a bit of time on. But, um, 30 seconds, genuinely 30 seconds.
And is it safe? Are you informed enough? Like is it able to trust? And your argument is that you can trust what it spits out because you've put the source document in the AI.
Absolutely. And and also I'm using this to give me that base level of knowledge and understanding. I'm not using it as the final authority on the topic. Important. And so when I'm using it to give me that base knowledge, I'm far more comfortable with being able to, um, to trust what it says. The other piece, though, is I do make sure that my prompts are specific. The way you prompt for these things will heavily influence the the stuff you get out.
And that's the difference. That's the key note there between you and I and between our listeners and yourself. You live and breathe GenAI every single day. You lecture in it, you teach in it, you work in it. It is your job title. Um, that I would argue you are very qualified to understand what prompts one should put into the AI to get back the best result. So before we move on then to, um, data based, what are the risks here, um, with what you've proposed and how do you respond to those risks?
Yeah. So look like everything there are risks. Um, and those risks are a) that the AI tool you're using and this would be for all of the examples that we give in this session, um, is not safe and secure. So absolutely advocate for nobody should be put in client or their own company information in a publicly available free AI tool. And also not even in a paid AI tool unless it is, um, what we call enterprise grade.
So make sure that you have safe enterprise AI. Um, the second thing is you got to make sure that the information you get out is accurate and reads correctly and well. Now, I actually think accountants are best placed to make sure that that happens because we've already got the review structures in place. Right? So if a grad does a piece of work, normally they're not just going to send it out to the client without a review. Right? It'll get reviewed by the manager or the partner. We've
already got a review structure. So keep the same review structure that you've got in place in place. It's just that the first cut is being assisted by these models. Um, and then the other piece is you can massively reduce the risk if you understand how to prompt well. And a big part of the training we provide to clients is helping them understand how to curate good prompts to get the right answer and the right tone of voice and all of those types of things. Yeah.
That sounds like a key piece of learning for everyone that's listening. How to prompt well, how to engage effectively with the AI to make it produce the best result. We'll see if we can find some suitable reading or information for that. Or heck, we may even do an episode literally just on that. But a little bit further down the track. There's so much to talk about in this space, and I know that our members really, really, really want the information. Let's move on to data. Um,
it's key to the life of an accountant. Um, what can a small or medium sized practice use right now to be more efficient, uh, more productive to add value?
Yeah. So there's probably a few examples that, um, that you can jump into straight away. So if you use Excel and I'm assuming that, that everyone listening to this does, um, you can get these large language models to do two things with Excel. One, you can get them to help
you either create or troubleshoot your formulas. So if you've got a massive workbook for forecasting or budgeting and you're trying to get you're trying to figure out why something's breaking, it's very good at helping you to either troubleshoot or create the right formulas for you. So that's that's probably
one use case. The second thing that some of these models can do, though, is you can upload Excel files to them and just say in its simplest form, just say, I've got, um, sales data for the first seven months of the year. Um, in an Excel file, you could then get the model to create columns for each month for the rest of the year, and then predict out, um, sales volumes for each month for the rest of the year, um,
based on some parameters. So you could say a 10% uplift in this or a change in this variable or whatever it might be, and it will then go and create the file, sorry, create the extra columns, create the data in those columns and allow you to download that file as well. And, and some of these newer tools like Microsoft Copilot, um, you won't even need to upload them and it'll just it just already is able to access your Excel files and just, um, and provide you
with those changes on the fly as well. Um.
How many accountants do you think are across the use or the availability of of that to them?
Um. To be honest, not many. So. Probably very few. Um, and look, that's not necessarily a bad thing, because the ability of these tools to be able to work well with Excel is a very recent thing. So if we were, if we were recording this 2 or 3 months ago, um, actually even more recent than that, that's not something I would be recommending people play with, because I just didn't
think it was good enough at that stage. And it's still early days, don't get me wrong, but it's now got good enough for me to think, well, this is interesting and I should get it trying so that yes, yes. I'm improving with it.
I just feel like this is good. I mean, I know that a podcast is hard to be, you know, breaking news, but I feel like that is a breaking news development that you can use AI that way and excel that way together. Um, and I just think, as you said at the top of the show, um, saving 30 minutes a day scaling that across your business, that to me sounds like something that is, um, day changing, week changing, month changing.
Yeah. And that's before you even get into, you know, like AWS, they have a data visualization app. If you're using that already, you can now use an AI module on top of that to be able to natural language, ask it to do things for you. So create me a graph that compares sales data for the last three months with accounts receivable and cash or whatever else it might be, and that's also extremely new. Um, and, and those are like kind of bespoke data visualization tools. Power
BI is going to have that very shortly as well. So, um, the examples I've talked about is are literally using the same tools we've already got, but um, with or alongside an AI model. Um, that's before you even get to these like specifically built data visualization and manipulation tools that have got AI embedded.
So we're talking about data. So there's two things there for data isn't there. There's a there's programs that have an AI data tool built on top of them, like you'd said with AWS, but in in the sense of Excel, you are putting the Excel into your text based AI, am I right?
Yeah. This so well there's two options. So if you um.
Just that I clarify that that's what we're talking about. Right. But you go ahead first. Yes.
Well for most of the models that's exactly right. You'll upload an Excel file, um, and it will read the file and understand it. And then you'd ask it to, um, you know, in that last example, extrapolate out sales based on some variables and, um, update the Excel file so you can download it. Um, but Microsoft Copilot, as that improves, um, that will get better and better at doing that natively
within Excel as well. Um, so there's there's a few different options depending on which route of safe AI you want to go down, which models you decide to choose.
Yep, that makes sense. I just wanted to make it clear that we weren't talking about a data based AI. We were just talking about how to use AI with your data, and as a result, you are using the text based AI to manipulate your data.
Yes, absolutely. And the great thing about that is it's available right now, right?
Yes. Absolutely. Absolutely. Okay. All right. So we're almost out of time. I want to have a quick look at, um, 1 or 2 or it might just be one bespoke option that's relevant to a particular practice, an example of what that might look like.
Yeah. So this is when you start to get into a little bit more, um, I guess, um, specific models that have been altered in a way to help with a specific problem. And this is good if you've got repeatable tasks that you do in your practice all the time. So one pain point that, um, I'm assuming almost every accounting practice has is onboarding clients, right? Like we just want to get in and start doing the work. You've got to get the right information. You got to do
AML checks, you got to do a whole bunch of stuff. Um, and so you could, um, build using a set of, uh, using an algorithm and, um, and a specific piece of, um, publicly available safe AI to build you an AI based automated workflow to, for example, onboard clients. So a client, you send them a link and it asks them for
some basic information. And based on that information, the AI will understand what further information to ask and where to put that information and who to notify based on what type of problem they need to be solved and when based on that person's availability and their expertise. Um, and
all of those other factors. You can now start to automate all of that, not based on some a small set of predefined rules, but based on AI understanding what the client wants, who they are and who's in the team and their availability and their skill set and all those other types of things. So you can start to now build some specific, um, automated AI based workflows. Um, and that makes sense when they're highly repeatable, when you've got to onboard lots of clients all the time, for example.
That's interesting. That's I'm writing that down, AI based workflows. That's the key takeaway there. And you get, um, that there's AI specialists that are able to help your individual practice do that.
Yeah, absolutely. The one thing I'd say with, with doing that is when you're working with, um, a team or a person to help you do that, there's lots of people that understand the technology. What you really want is someone that understands the technology and the practical applications for your business. And there's no different for you as a
chartered accountant. Um, you know, the real value you add is not your understanding of the numbers or the standards, but it's your ability to relate that practically to your client's business and the decisions they need to make. Um, I would say the same thing applies here. The technology is great. And actually it's you know, there's lots of people that can get across that. But how does it specifically impact the the your business, the people in your business,
the processes within your business? Um, that's the real key to making sure you implement it well.
Um, okay. Well, we're about to wrap up my final then. Um, quick question is, after all this, is it easy to find the reputable program, um, that you suggest, you know, is it easy to find a reputable GenAI program that is, um, of the suitable enterprise grade that you talk about? How do you find it? Where do you start to find that?
Yeah, it is available. There's a couple of I mean, the main options that, uh, the clients that I've been talking to are interested in at the moment, um, ChatGPT Enterprise and Microsoft Copilot studio. I'm not saying they're the only ones. Um, those are probably just the ones that come up in conversation more than others at this specific
point in time. So, yes, um, that that is possible. Um, and in terms of doing that process and onboarding and all those things, I think it's very important that anyone that on boards, AI does it in a, in a safe manner so that they understand the parameters, um, the implications of putting various types of information and they've got some specific rules around what they can use it for, what they can't, the review process, all of those types
of things. And when you get someone to help you implement AI, they will be able to provide you with a templated flow that basically ensures you're safe from woe to go and that they help you with the change management, the process, all of that stuff should be a, you know, a repeatable task for them to help you with.
All of that makes sense. If there's some specific reading that, um, Kayur is, um, suggesting we'll whack that in the show notes. And as a separate, um, piece of amazing reading, I've got the, uh, the team at the CA library to pull together a list of resources that are relevant and specific to AI and accounting and also AI in general. And I've put all of that in the show notes for episode one, and I'll make sure it's also in the show notes for part two. It is definitely action stations. Um,
there's a lot to think about. Don't feel overwhelmed though. It is achievable, and you can see why Kayur suggests that humans with AI will end up replacing humans without. When you could have a look at the efficiencies and productivity. That's what the bid is for me, saving 30 minutes every day or even a week. And you're a step ahead of those who aren't, am I right?
Oh, absolutely. And so I think everyone right now should be using safe AI for text based use cases. Um, absolutely. I think we should have every organization. There should be at least, um, you know, a percentage of people looking at ways to use it with data. And then I think at this point in time, be aware of some of the bespoke things that you can do with it so that as it becomes more available and cost effective, you can get on top of that as well.
Mhm, mhm. Such an interesting discussion. That is all we have time for. And as I said I think we should touch base again at the end of the year uh to see. Or it may even be before that to see what else has been developed. Um, and if you remember who rolls out some of this um, or is using GenAI get in touch. Let's discuss the results. Let's talk about what it is doing for you in your firm, in your business, for your clients. Um, have you checked out
the podcast page on the CA ANZ website? I recommend you do that too. There's plenty of other great content experts, interviews, and resources that are tailored just to you. It is worth checking out, and I'll put a link in the show notes as well to the website so it's easy to find. And of course, you'll see a link to the podcast in the newsletters that you receive from CA ANZ.
If you want to get in touch with me and the team, email us at [email protected] and follow the pod in your favourite podcast app. Let's start a conversation. Thank you Kayur Patel, for being my guest on two epic episodes of Small Firm, Big impact.
Absolute pleasure. Thanks, Gill. Bye bye.