AI-Assisted Bookkeeping: Making the bookkeeper irreplaceable

Aden: Next up is AI-assisted bookkeeping: making the bookkeeper irreplaceable. All about using AI to take the grunt work off bookkeepers so they can focus on the judgment calls clients value. So please welcome Glenn Castle, Regional Product Domain Expert at Dext. Glenn, welcome, and thanks so much for being here. We'll let you share your screen and let you get started.

Glenn Castle: Awesome, great to be here, Aden. Let me share our screen, and we will kick off this session. Alright, hopefully we are all good now, and everyone can see the screens. So, welcome everyone to today's session on AI-assisted bookkeeping. I think every other headline out there at the moment is talking about how AI is coming for bookkeeping. We actually have a different view on the Dext side of things. We see it as a way for a lot of bookkeepers to now pull ahead, by being able to actually start automating a lot of the manual tasks that you've traditionally been doing, where maybe your time is not being best placed. Whether that's splitting out different transactions, coding different line items, chasing your clients, all of these other bits and pieces that everyone struggles with on a day-to-day basis.

Glenn Castle: So really, what AI is going to enable us to do is start doing things a lot more efficiently. And I'm pleased to be joined today by Kelly Marshall and Niamh Campbell. Kelly and Niamh, we've had a lot of chats over probably the past few months, and I'm really excited about some of the things that they're actually doing at the moment, and that's really what we wanted to share in this session today. What are they actually doing? But we'll just start off with some introductions. Kelly, are you able to introduce yourself, tell us a little bit about yourself, your practice, your background?

Kelly Marshall: Yeah, so, obviously, I'm Kelly Marshall. I currently run First Class Accounts Richmond. I've been working in accounts departments for over 20 years. I started in small business and corporate, before starting up First Class Accounts Richmond. I learned very early on in my career that efficiencies are just the way to go. And I remember when I started First Class Accounts Richmond, I was introduced to a product called Receipt Bank, now Dext, and I just sat there and went, where has this been all my life? And now I'm using AI to help me make products like Dext and Karbon, that I have used in my practice for 10 years, more efficient.

Glenn Castle: Awesome. Thanks, Kelly. And a little bit about you, Niamh.

Niamh Campbell: Morning, everyone, thanks for being here. My name's Niamh Campbell, I own and run First Class Accounts in Wanneroo. I've been a franchisee and bookkeeper for 10 years. I come from a hospitality management background, and that means processes and that sort of thing are really important to me. So, again, I was introduced to Receipt Bank very early on in my bookkeeping career, and it's something that's always stayed with me. Very similar to Kelly, I'm all about the efficiencies and making life a little bit easier, and the AI introduction has just worked with that as well.

Glenn Castle: Yeah, awesome. And I guess a question for you, Kelly: how did you get started with using AI and some of the things that we're going to be talking about today?

Kelly Marshall: Yeah, so I grew up in a household that was very security conscious around things like IT. My parents worked writing IT standards and policies and procedures for government departments, so it was always drilled into me. So I was a bit of a latecomer to AI, I was very hesitant. But I started using it probably at the beginning of the year, to help me write emails better, because I do numbers better than I do writing, and just pull together ideas. And then I visited Niamh in August, and she helped me get started with Claude, and it's just been an absolute game changer. Being able to connect it to my systems using MCPs, it's now an integral part of my day-to-day workflow.

Glenn Castle: Yeah, I love that. And even though you've only been using it for a few months, I think everyone's a beginner, right? You see how quickly everything is actually changing. Something that you knew last week is suddenly...

Kelly Marshall: Completely irrelevant today.

Glenn Castle: It's like a knowledge-level playing field kind of thing. But I guess for you, Niamh, how long have you been using AI, and where did you start with it?

Niamh Campbell: Well, I actually asked my ChatGPT today how long I'd been using it for, and it's 3 years and 11 months, so December '22 was when I opened my first ChatGPT account. And same as Kelly, I was using it to write an email here or there, or asking it questions instead of going to Google, because I think ChatGPT or Claude have replaced Google a little bit. And then, probably about a month before Kelly came to visit me, a friend had told me about Claude. Claude was all over the place, and everybody was talking about it, and she sent me a YouTube video, and my brain doesn't sit there and watch a YouTube video, so I took the transcript and put it into ChatGPT, and created a step-by-step guide on how to set up Claude. And that's how I got started, and that's the guide I then passed on to Kelly to help her get started.

Glenn Castle: Yeah, exactly. And I know you guys have been bouncing a lot of ideas off each other over time, and I think that's a big part of what this session is as well: hearing from other bookkeepers in practice, what they're doing and how they're doing it. Some great stories there. Alright, so we're actually going to get into some of the detail around what you guys are doing as well, which I think is going to be the most benefit for a lot of the people on the call.

Glenn Castle: Today, we're just gonna start off with a little bit about where firms are actually at with their AI journey, so just uncovering a little bit about what you've touched at the moment, and then getting a bit of an idea of where you can actually start using things as well. This is not necessarily a Dext-oriented session, if I can get that word out, but we're going to start talking about how you've started connecting, and all you guys are using Claude, we know that, into some of the different tools as well. So, where practices can start. Obviously, a big part of this session is really around Dext, what we can do. I know you guys have been using Dext since the Receipt Bank days as well. And the last section we want to talk about is really around the guardrails and the security around things, which I think is a really prominent issue for why firms maybe have not taken the leap of faith with it yet.

Glenn Castle: Kelly, obviously, with the background that you just gave us a bit of an intro on, you've got some strong thoughts around that as well. So these are the areas we're going to go through. Before we get into things, this is a survey that we actually did at Dext. We produced a guide called AI in Practice. This was around about April or so earlier this year, so I actually suspect these figures have changed quite significantly since then. But the one that I think probably hasn't changed: 57% of the people on that survey considered themselves a beginner. And I think, like we just spoke about, that beginner mindset gets reset every single week with how things are actually happening. 46% of people said they were curious but cautious about it. So, what are the security implications of it? Where is my data going? What can I actually trust it with? What activities can we look at?

Glenn Castle: And then we took a look in that guide as well at where AI has actually been used at the moment. So a lot of it was around AI for communication. So think about things like, I don't know, Google completing your email responses back to your clients. Those types of activities are really what we're talking about there. 41% around task automation, and then 40% for meeting transcription as well. But there's lots of different tasks AI is going to be used for. Different people are going to do different things, and hopefully today's session will give you guys some ideas on what we're talking about.

Glenn Castle: Now, when we start talking about AI, there's really three areas that we're mainly breaking this down into. So one, we're going to start with what's called embedded AI. This is AI that you're already utilizing within the actual software applications at present. At Dext, we've had AI in our platform since around about 2016. That's the way we're processing the volume of invoices and receipts that we are, or have been in the past. But it's also morphed into things like our AI Assist product, or Xero's got JAX, MYOB's assist, et cetera. So all of these other tools are embedded in the accounting software that you're using on a day-to-day basis.

Glenn Castle: We obviously have the chat tools out there. Niamh, you've obviously been using ChatGPT for quite a while, which I think in those early days really proved useful for lots of different things, whether it was drafting emails to clients, what things need to look like, what I could do differently, maybe even the research side of things as well. And now we've moved into the area of what we're referring to as connected AI. So this is starting to connect all of those different chat tools, as we mentioned, Claude, et cetera, into some of the software that we're using on a day-to-day basis. So whether that's Ignition, Dext, Xero, et cetera. This is what we're going to refer to as connected AI, which is what we're going to spend a number of examples on in this session.

Glenn Castle: So, just out of interest, we've got a bit of a survey that should hopefully pop up on your screen in a second. We just wanted to get a bit of an understanding of what AI tools you are using in your practice. So whether that's Claude, Copilot, ChatGPT, Gemini, or other. And just while that's being answered, I know there was a bit of a question in the chat around that, that everyone seems to be using Claude now. Maybe, Niamh, if you wanted to respond to why you started using Claude over ChatGPT?

Niamh Campbell: Yeah, well, I see Jacqui's comment about FOMO. I think that was really it. I was at an event, and everybody was talking about Claude, so I went home, and I was like, it can't be that much different to ChatGPT, and it was. I think ChatGPT has now brought in the co-work side, and it does a lot of the similar stuff, but at the time, Claude was the one that was doing all the co-work and the on-your-computer stuff, so that's why I went down that route.

Glenn Castle: Yeah, absolutely. I think those answers are probably coming in now. I think we can probably end the poll. So, around 46% of people are using Claude, 25% Copilot, 27% ChatGPT. And as we see with all of these different models, each of them seems to be moving ahead, doing different things as well. And I see something in the comments around using Grokbot also, which I think is one that's starting to get a little bit more prominence, albeit a little bit niche at the moment, but yeah, lots of different tools that are out there.

Glenn Castle: And I think you guys have all been experiencing your clients starting to use these types of tools as well, and starting to ask you guys questions. My wife runs an accounting practice as well. You can always pick when a client has stuck your advice through ChatGPT or whatever it is, and come back to you, and then you've got to go through and determine what is and isn't right, and what the context is that it didn't get. So I think it's an important thing to start utilizing in our practices ourselves. We've probably all seen examples of where our clients have completely got it wrong as well, and where we've been called in to fix it. Kelly, I don't know if you've got any stories on that one at all?

Kelly Marshall: Yeah, one of the most important things that I learned when setting up Claude was about letting it know what my business is, and what we do, and how we do it. And then, when I'm working on client stuff, knowing who the clients are. A lot of people haven't done that part when they've just jumped in and asked it a question. And without knowing the niche information, it's given them very incorrect responses. So I've had clients email saying, oh, can I do this? And I'm like, if you want to be in breach of Fair Work, go for it! But I might have to fire you. And it's one of the things that I've gone through with Remy, which is what I call my Claude. It's come back with sources being random blogs written by random people, and I'm like, no, no, no, we get our information from the ATO, Fair Work. We make sure that we source it from proper places.

Glenn Castle: Exactly, not some influencer on Facebook, right? Awesome. So I've just got another poll that's going to crop up here as well, and we just want to really get an understanding of where you see your firm with AI today. So whether that's mostly manual; AI helping with some things, but you're still checking everything; some of the work running automatically; or we've got AI agents in there that we may have built, acting with some guardrails on there as well. So we'll just give you guys a bit of a chance to answer these questions. And look, this is different across each of our tasks as well, right? There's some things that we're probably going to be able to trust AI with, and there's some other things that we're not necessarily going to be able to trust the AI with. So we'll just give you guys a quick moment to answer those, and maybe we could just end the poll there, and we'll have a look at the results.

Glenn Castle: Alright, so we've got 28% mostly manual, 54% is the leader: AI helps, but you're still checking everything. And I think that's the key, right? It's got to earn its trust in the workflows, as to what things we can trust it with, or what things we want to rely on it for by exception. And we've obviously seen the TPB guidance that's come out recently around that exact thing, because at the end of the day, we're responsible for everything that goes out to the clients.

Glenn Castle: So, where your firm sits, this is a little bit of a follow-on from the question that we just had. Manual, we're still talking about data entry, spreadsheets, paper onboarding. AI-assisted, so we've got things extracting receipts, et cetera, drafting emails. Automated is things like some of our bank rules, auto-publish, blah blah blah. And then the last one is the agentic area, which is where we're moving to now, where we've got some of our workflows set up, where actions are starting to be performed, but within the clear boundaries that we've got in place around all of that.

Glenn Castle: So, what I'd like to do now: up on the screen here, we've actually gone through and had a look at both Niamh's and Kelly's practices, at where some of the automation has actually come into place for them. And quite importantly, you can see on the screen what that work looked like before, how much time it actually took, and what it actually looks like after. Both of these examples are obviously utilising Claude. I guess for you, Niamh, where did you start first, and what did it replace for you?

Niamh Campbell: Yeah, I think the first one I started with: because all of my clients come on board straight away on fixed-price weekly packages, I used to do that as a spreadsheet matrix, so I'd enter all the information into the spreadsheet. And even though I do use Ignition for my engagements, I didn't love the choices option in Ignition, because clients couldn't really see everything that was included. So I'd create a Canva PDF with the three packages for them, and that whole thing used to take me about 45 minutes to an hour to get it all ready and sent out. But now Claude's built me a calculator, so I can just enter all the information into Claude, and it then generates the PDF of the 3 packages, ready for me to send off to the client. So it says 10 minutes there, it's probably even less than 10 minutes, it's probably 3 minutes. It probably takes 7 minutes to get the information that I need.

Glenn Castle: Yeah, absolutely. And with that as well, how many of these are you churning out? Because obviously this is looking at one particular example, so when we start multiplying it, I think that's where a lot of the efficiencies have come in.

Niamh Campbell: Yeah, and also, I think the fact that Claude is now connected to my Ignition, so once the client comes back and says, yeah, I'll take the second package, I can just go into Claude and go, can you create that engagement letter now in Ignition, and it will take everything from it, instead of me manually doing that. So it all links in together.

Glenn Castle: Yeah, absolutely, and I think that's a really good example of how we're using that tool to start connecting different systems. I guess the other one that we've got there that I'm interested in as well, and I think a lot of people will be doing this one already, is the meeting notes. How are you using that with Claude to transfer that into actions?

Niamh Campbell: Yeah, so I do use Vinyl as well, so between Vinyl and Claude, I get the information I need. With Vinyl, I get the file note and bits and pieces, but sometimes I've got more information in Claude, so I'll take the transcript from Vinyl and put it into Claude. So it's between Claude and Vinyl. I've sent meeting notes to clients 5 minutes after a meeting, and they're like, how'd you do that so quickly? But it's mainly for me, just so that I remember what needs to be done, because the minute a meeting finishes, I've normally forgotten what we spoke about.

Glenn Castle: Yep, absolutely. And I guess for you, Kelly, how did you start choosing some of those repetitive jobs that we've spoken about in your example, and then, importantly, what are you doing with those extra 2 hours, the time it's freed up?

Kelly Marshall: Yeah, so I started testing things like email management with my own business, having Claude go in each morning and clear out the noise. Just forward invoices to Dext, et cetera. I then had a client who allowed me to start testing more systems, even just on her work. I've got Remy connected straight in to her email, and it goes in and it clears out the inbox. And unlike using forwarding rules, where things can be missed, like notes that are in with the email, Remy knows that if there's a note in there, to leave it in the inbox, and it gives me a note to say, hey, you need to go and pay attention to this.

Kelly Marshall: But I was very careful to start, like I said, with my own stuff, teach it the rules: don't delete, archive. Mark as read, so we know it's been actioned. Just things like that. I've now started rolling that out across more clients as I've gotten permission to do so. It goes in now in the morning. I get up in the morning, log in, and it goes and checks the clients' emails, forwards to Dext, et cetera. And yeah, that's saving about 2 hours of time, and I get that started in the morning before I go to the gym and do the one thing that AI can't do for me, and lift weights.

Glenn Castle: Yeah, absolutely. And we've got a slide coming up on that one, so we'll be able to show people exactly what that report looks like that happens on a Monday morning for you. So I guess maybe one for you, Niamh, then. What does your technology stack look like, and where have you made some of those connection points? Because I think that's a big thing people are wondering: how do I go through and do this, and what can I do? We've touched on a couple of examples there, but I'm interested to hear what your technology stack looks like.

Niamh Campbell: Yeah, so if I start using something and it works for me, I'm gonna stay with it, so I'm not always chasing the new shiny thing. So a lot of these apps I've been using since we started the business in 2016. For bookkeeping, Xero, MYOB, QuickBooks. Dext. I do use both Xero Tax and LodgeiT, because I get LodgeiT for free, so I keep that with my QBO clients. I've got Karbon and Practice Protect, HubSpot, Typeform, Ignition, and Content Snare, Mailchimp, Microsoft, Zapier connects all the things that don't work together, and then Canva and Vinyl. And anything that does connect to Claude, I have connected it, but locked it down so it's not able to change anything. It's read-only, until I know what it can do or what it's going to do. So yeah, it's a pretty big tech stack, but there's reasons for all of them, they all work for a reason.

Glenn Castle: Yep, awesome. And I guess for you, Kelly?

Kelly Marshall: I've got a very similar tech stack to Niamh. And yeah, just using the MCP connectors. The ones I've found most helpful have been Dext and Karbon. I was put onto the Karbon MCP connector pretty early, and being able to pull out information on trends, with how long bookkeeping, or BAS preparation, or IAS preparation was taking over the years, was just game-changing. Being able to see where the time spent was trending up, and being able to go and see, okay, where can we be more efficient with that client? Pulling out information that I don't even know what to do with yet. But it's there, and I will get around to figuring it out.

Glenn Castle: Yeah, and I think a good example, when we were chatting the other day, right, is, okay, you've pulled this information out, at least it's learned from it, so if you do decide something in the future, you've got that trending history. The other thing I just wanted to touch on, a big part of this particular slide, is the documentation of all your processes, right? Your standard operating procedures that you've got in there. Maybe Niamh, from your perspective, why is that so important, and what has it meant for how you've built the skills to use in Claude?

Niamh Campbell: Yeah, my business grew very quickly, and before it grew, I was doing all the work, and it was all in my head. And then I had a team of four offshore who were doing the bookkeeping, and they're all amazing, I love them, but they were doing all the work, and I still didn't really know what was going on. So I've now got 4 staff in Perth that have either just finished or are finishing their Cert IV.

Niamh Campbell: And as I'm doing something with them, I'm recording it and creating an SOP out of it, so that for the next person that comes on to do the job, we're starting to build that up. So at the moment, I've got about 70 SOPs built, just from the work that we're already doing, just recording it and using Claude to build it out, so they all look the same, they all read the same. And we can then start using them for training of new staff moving forward, and redoing all my checklists, making sure they're all in the right order, and all of that sort of stuff. Yeah, the SOPs for me have been a game changer.

Glenn Castle: Yeah, absolutely. Probably one that doesn't get the attention that it deserves, but it's obviously core to how you run your business, right? So what we just wanted to have a quick overview of here as well: this is the Dext workflow, how data comes into the system, what Dext actually does with it. A big part of that is obviously the automation side of things. The data health and insights area, which is basically connecting through to Xero and QuickBooks, is where a lot of the data that's going to get pulled into Copilot, ChatGPT and Claude at the moment comes from, through the Dext MCP connector, along with some of the practice productivity stats as well.

Glenn Castle: What you'll probably see out there at the moment is that a lot of these connectors are very new in terms of their availability, so maybe not everything that you want to do is available there. The rest of the ledgers are no different, but where we're seeing a lot of practices benefit at the moment is getting that data health and insights within here, and we're going to go through and have a look at some of those that Kelly and Niamh have actually built. How that works is obviously connecting through, and in this particular example, we've got Claude here. This Dext Client Health is actually a skill that's been built, and both Kelly and Niamh will have their own skills that they've built themselves, that enable a standard output to be produced based upon the data that's going to be connected through.

Glenn Castle: Skills are easy to go through and create yourselves. Claude, Copilot, ChatGPT will actually be able to help you go through and do that yourself, so you don't need to be some expert in writing any of this. The AI tool is the knowledge creator for you. I guess one for you, Kelly. This is an example of your report that we've got here. Are you able to take us through this Monday exceptions report, what it looks like, what the process is, and what it means for you and your practice?

Kelly Marshall: Yeah, so obviously there's information about data health, et cetera, in Dext. It's always been available. But I've always had to go check it myself, and there's never been anything that said, oh, this is where your priority needs to go today. So, with Niamh's help, I built this report, and it goes in and gets information about the oldest invoice that's with certain clients, how many lines are unreconciled, all the information that was already available in Dext, but it prioritizes it for me, so that I can then go in and see where my priority needs to lie that week. I can go in and say, okay, why has this client got 600 unreconciled lines? Why has this client got an invoice that's been sitting in Dext for 34 days?

Kelly Marshall: And it's even giving me comparative data of how long between a transaction happening and when it's actually been received from the client. So I can see where delays are coming from the client, so maybe I need to better educate them on using Dext more efficiently and forwarding things as they happen, rather than waiting until we ask for them. And that's all efficiency, things that create more efficiency.

Glenn Castle: Yeah, exactly, and they're really big examples: understanding where that bottleneck is, and having that information available so that you can action it. Because in the past, you may know that there's a problem, but not necessarily be able to put your finger on it. And that's what these sorts of tools are about, because they're so good at gathering all of this information, it can then be surfaced with you. Niamh, I know you've got a similar type of report that you create as well. Are you able to take us through this? And then, importantly, we've got some red clients on here. I know this is obviously sample data, but what do you do with those clients that are red?

Niamh Campbell: Yeah, so it's a very similar one to Kelly's, but the biggest one for me is the movement line. It gives me a comparison of how many prepare items were in Dext the previous week, and how many are in it this week, so whether someone's jumped in and actually cleared Dext out, or whether there's just been more added. So it just gives me that visibility over all of the clients, to go back to the staff member and go, is there a reason? What's going on with this one? But the red is based on the health score in Dext, which is across everything, so it gives me the health score and how long it's been since it's been looked at. They're those clients that you don't want to slip through the cracks, so it helps me stay on top of those ones before they realize there might be a problem and think it's time to leave. So it's flagging them before they flag it.

Glenn Castle: Yeah, absolutely. Being able to be a bit more proactive with things, right, is what it's all about. Another one, I guess, for you, Kelly: looking at the auto-publish tracker that you've got here, what has that meant for the efficiencies that you've been able to create in your practice?

Kelly Marshall: Yeah, so one of the things I did a few weeks ago was go through and do a full audit of all my clients' Dext files, and find where there were blockages. So where Dext was maybe accidentally picking up statements as invoices, and they were going through, or where the settings weren't set up right, so they were getting through, and so the team was defaulting to just turning auto-publish off. And I was able to work through those things with Claude and go, okay, what setting in that client have we not got set up properly, and can we fix this? So it just helped me answer questions a bit quicker. I've taken all my how-to docs, put them into Claude, and created a skill, and it now knows how I want things done in Dext. So it's able to apply that when it goes through, and help me move things along.

Glenn Castle: Yeah, absolutely. You had some great examples in there of the practical things that are happening, right? You think, okay, let's turn auto-publish on for everything, but hey, if you've got a statement in there that's been auto-published, being able to have this sort of tool identify that, rather than it showing up in your AP report 7 days, 14 days later, right? So it's about all these sorts of things.

Kelly Marshall: And they paid it.

Glenn Castle: Exactly. Cool. The last one I just wanted to touch on, Kelly, for you, that you've got Claude building as well, is around some of the control accounts that we're utilizing data health for. Are you able to take us through how you use the control accounts and the importance of that for you?

Kelly Marshall: Yeah, so obviously, again, this is all information that has been in Dext forever. But again, you've got to go look at it, and know where to look, and pay attention to it. And sometimes you would get busy doing other things. Having this report emailed to me, I think it's on a Tuesday, I tried not to get them all at once, highlights to me again where my attention needs to go, and what has potentially been missed. What have the team forgotten to do? And do I need to improve notes in Karbon to make sure that these things are getting done properly? So it's helping me create efficiencies across my entire tech stack, by bringing the things that need my attention to my attention, rather than me having to remember to go looking for them. And with the traffic light system, it's like, oh, god, okay, we don't want red ones!

Glenn Castle: Yeah, absolutely. And the last one I just wanted to talk about: I know you've both built your own tools here as well, where you're starting to look at some of the time that's been spent in Dext, and then also tracking this against Karbon. I guess for you, Kelly, when do you run this comparison, and what is it actually looking at for you?

Kelly Marshall: Again, this is a weekly one. As business owners with team members, if we don't keep an eye on these things, then they can very quickly get out of hand. And when we're putting a lot of effort into creating these efficiencies and providing these programs, we want to make sure that the team are using them. So being able to pull the data, and again, this is data that's always been available, but some of it's over in Dext, some of it's in Karbon, being able to pull it together into one report and actually have it give me the information: okay, the amount of time that's being spent processing in Receipt Bank for this client has gone down by 5% in the last week, but the time spent on bookkeeping went up by 10%. And I sit there and go, why? Why, if we're being more efficient over there, are we not being more efficient here?

Kelly Marshall: So again, it's data that's always been there, but as business owners, we haven't always had time to go and actually pull it all together, and now we've been able to create these programs. And most of these programs really didn't take that long to build. I ask Remy a question and say, can you give me this data? And it will go, here you go, and I go, cool, can you now give me this data every Monday morning at 8am? And it will go and set a scheduled task, and it'll pull the first report, and I'll sit back and go, I like this, I don't want that, I like this, I don't want that. And I just have a conversation with it. I live with my earbuds in my ears now. I don't have to type.

Glenn Castle: So it's about being curious, right? That's what it comes down to. What do I want to achieve? I think that's a big part of it. I guess for you, Niamh, your reports are off to the right-hand side. How do you use this for your fee reviews and that side of things?

Niamh Campbell: So at the moment, it's just too much information for me. I'm getting all the information in, and sometimes I'm not looking at it. I think I just got excited and threw a whole lot of them together. But I also promised all of my clients this year that there wouldn't be any fee increases unless it was very much out of scope. I wasn't just doing an increase. So I suppose this information is going to be there for next year, so that I can look at fees and make sure that all of my package prices are scoped correctly, and use it for that. So I'm just getting the information at the moment so that I can tweak it, so that when I do need to use it in a year's time, it's ready with the information that I need to have the conversations.

Glenn Castle: Yeah, exactly. I think that's a really useful section, just to understand what's possible with Claude and how it's connected to all of these different systems. The third part that we want to run through in this session is the Dext AI Assist. So this is where we're starting to talk about AI that's embedded within the applications, that you can control. Think about how a lot of information actually comes into tools like Dext. Traditionally, they've relied on that standard automation: if it's a Bunnings receipt, code it to materials, et cetera. What AI is giving you the ability to do now is actually start putting context across a lot of those documents. So you can start seeing how we're able to put some prompts into our client, or even at our practice level if we need to, that can actually start impacting those receipts, invoices or documents that come into the system.

Glenn Castle: There's a whole lot of things that it can do. We're going to skip through and show you some particular examples of what's actually being meaningful in practice. But essentially, this is an AI agent that is embedded within Dext, and for anyone that's familiar with Dext, it's taking those supplier rules to the next level, right? In the example I've got there, if it was Telstra, code it to telephone. Now we can start saying all invoices from telephone providers, so it doesn't care whether it's Vodafone, Optus, whatever it is. We don't need to write a rule for each one. A whole lot of different examples as part of that.

Glenn Castle: Kelly, I know you've played around a little bit with AI Assist, and have been using it for a while. One of the examples that I love that you gave previously is how you were actually learning to write those prompts, and I know you use Claude for that. Are you able to take us through how you did that?

Kelly Marshall: So when I did my audit of all my clients on Dext, where I came across a situation where auto-publish had been turned off because this had to happen, or that had to happen, I would just go to Remy and say, okay, can we write an AI Assist prompt so that we can turn auto-publish back on? And it would give me the prompt. Again, I do numbers, not words, so it would give me the prompt, I would put it in, and we'd test it for a little while. And yeah, it's just, again, been an absolute game changer.

Glenn Castle: Yeah, exactly. And again, I think this is being curious, right? You know you've got all this work that you need to do. You've used Claude to try and solve that problem, albeit within some of the embedded AI within Dext. And the important thing I know you've spoken about as well is making sure that it's built up that trust over time, so you actually know what's happening before you turn things on to auto-publish, right? Because you don't want to have your AP end up a complete mess, as an example.

Glenn Castle: So we'll go through a couple of quick examples. I know you guys love this particular example that we've got up on the screen. This is one that happens to pretty much every bookkeeper, probably multiple times a day: the fuel receipt that we've got there, the purchase of fuel. There's also a pie, a drink, a Coke, or in this case, a Red Bull on the invoice.

Glenn Castle: The reality is, who wants to spend their time splitting all of this information out, because it needs to go to different accounts, or the GST removed, et cetera? So what AI Assist will give you the ability to do, and there's a prompt in the middle of the screen there, is, if you had that embedded in Dext, anything that comes in in the future, the line items will automatically get split out. The other key thing that you'll notice on there as well is the quantity. So the quantity, the litres, have been pulled out. A lot of people have probably drawn their own conclusions out of this: fuel tax credits.

Glenn Castle: I know, Niamh, you put it in the description, but again, it's just an example of how we get information out of the documents.

Aden: 10 minutes, Glenn, just FYI.

Glenn Castle: Yeah, cool. A couple of other quick examples we will run through here as well: even adjusting things like descriptions, right? This is a real-life example. This was actually done 14,000 times on our clients' records, where they were manually typing in the description field, AWS August 2026 Cloud Infrastructure. We can basically start utilizing the supplier name, the month, the subscription, all taken off the document, to actually create those rules themselves. Even something like, in this particular example, being able to assign things to tracking categories as well. So this has picked up the property address for a maintenance invoice, and automatically assigned the tracking category.

Glenn Castle: And for anyone that is a Dext customer, and I know, Kelly, Niamh, you're going to get your own reports of this as well, this is something that we can actually provide to each customer that's been utilizing Dext, as to what's actually happening in their practice. So where are the manual edits? Where is the actual time that's been generated?

Glenn Castle: The final thing I wanted to touch on with Kelly and Niamh, and we are pretty much at time, is maybe just some quick final thoughts, Kelly, on the security layer of things. There's been a number of questions that have come up in the chat as well, around your thoughts and what you're comfortable with, utilising Claude with Dext, et cetera.

Kelly Marshall: Well, my first guardrail, when I first started using it on doing things for clients, was watching it do the work. Watching what my instructions were getting it to do, and checking the output. Not dissimilar to what I would do with my team's work: what they provide me, I check it. And then, yeah, making sure I've got an AI policy.

Glenn Castle: Yep, absolutely. And maybe for you, Niamh?

Niamh Campbell: Very similar to Kelly, but the other big one for me is making sure I'm not using any free software, because free software means I'm giving them something to pay for it. So I do pay for all of my AI.

Glenn Castle: Yeah, absolutely. Cool. I guess the important thing as well is that a lot of these efficiencies that you guys are creating haven't necessarily meant lower fees. It's just a way of utilizing a tool to make yourselves more efficient. So lastly, any final takeaways, Niamh, Kelly, you'd like to give the audience before we wrap it up?

Niamh Campbell: I think I had something written down. If you do have it and you haven't started, just start with something small, or ask Claude, how can you help me? This is my business, this is what I do, what can you help me with? And then it just starts growing from there.

Glenn Castle: Kelly?

Kelly Marshall: Like Niamh said, get started, just go in. And as I said at the beginning, and I think a lot of people have mentioned it in other sessions yesterday and today, make sure you tell it who you are. Tell it what sort of business you are. And if your number one rule is, any data I give you access to, you are not allowed to delete anything, tell it that, and make sure it saves that into a memory or a skill or something. And learn how to use skills.

Glenn Castle: Yeah, awesome. All right, I don't think we really have time for questions now. I'm getting looks from Aden that we need to be wrapping things up, so I appreciate everyone's time, especially Kelly and Niamh, for the amount of time you've put into not only building what you can do in your own practice, but into this session today, to help out your fellow bookkeepers and accountants that are out there. Really appreciate your time. So thanks, everyone, for the time.

Aden: Thank you, Glenn, Kelly, Niamh. That was awesome, we really appreciate it.