Will AI Replace Accountants?

This is your business unleashed, a podcast about building a business that makes money without you. I want to help you grow your company effectively so you can enjoy your work and your life more. I'm your host, Clayton Aiken. I'm an accountant and business advisor and entrepreneur and a family man. Every episode, I'll provide simple, actionable insights for entrepreneurs who want to get off the treadmill. We'll talk to other business owners about their journeys and industry experts about the tools that they use to gain a competitive edge. Let's unleash your business. All right, welcome to another episode of your business unleashed podcast. Today we're talking about AI and I've got a wonderful guest who is a fellow CPA in the industry from New Zealand and Canada. Welcome Peter McAral, CPA CA. Thank you so much for being with us. Hey, Clayton, it's good to be here. Yeah. So you own an accounting firm fuel accountants where you located in Toronto, who is that side of Toronto? All right, Toronto. So there for anybody who says that I'm very Alberta centric with my content, you can bugger off. I've got a Toronto Toronto fellow CPA on here. This is a national initiative to try and get AI going in our companies. We are deploying AI in our accounting firm at scale. We're a little, you know, sort of iffy and wishy washy on how to roll this out to the team, how to do it safely. What are the security concerns? How do you make sure that people aren't just dumping massive amounts of time into playing with AI rather than being productive with it, right? And those are all the sort of challenges that we're working through now. So, you know, thinking through that has been a challenge. And tell me about your background and what you're doing to help, you know, guys like me, infirms like us sort of sort of get through that. Yeah, well, I mean, it all started when I was, you know, like you thinking about what am I going to do with AI in my own firm. And, you know, that started in, you know, in 2024. And I started by, you know, learning as much as I could, I got training on AI. And I got a course back then that was mostly on prompting and, you know, but really started to think through the consequences of AI, not just the tool, but where is this going? And so that's when I started writing about AI for other accountants because I could see that this was going to be a lot more than just another software that we buy. Yeah, and so now, I guess I could engage you as a firm owner deploying AI and what would that engagement look like? Like how does that work? Yeah, so we've got a number of things we're delivering. So right now our key product that we deliver is called the practice transformation program. And it's a nine part program where we help firm leaders think through the leadership aspects of their firm with respect to what AI is going to do in their firm. It's not a technology program. It's all about leadership and positioning. And then we also have some AI training for the team. We have an AI prompting course for accountants and we'll also we haven't released yet. We have a Claude for accounting or for accountants project that will be able to be used within the team to teach them all those nuances of how to use Claude for an accountant rather than just. So we don't have enough stuff on the internet, but so it's a so shallow. We want to deliver something that really helps accountants really get to use it well. Right, and so what are the biggest, you know, just let's think about just generally a service businesses or professional services business, maybe law firms or engineers, whatever any service business. As we're rolling this out, what are the biggest pain points that you're seeing in the industry and from the leadership or from teams, you know, what are you seeing? Well, I wouldn't say so much there's pain points yet. There's pain points and learning how to deploy it well. And so a lot of firms, a lot of knowledge based companies are using AI, but they don't know what they're trying to achieve with it. And I think this is the problem that accountants are having. I think this is the problem that any knowledge professional is having. We know it's so powerful. We know it can do so much, but we actually don't know what problem we're trying to solve. And so when I'm teaching accountants, one of my famous sayings is that if you don't know what problem you're trying to solve, no tool you buy will solve that problem. And I was just being the two massive conferences and hundreds of vendors, you know, and each of those rooms trying to sell AI solutions to people. You don't know what problem you're actually trying to solve. Their solution is only going to give you marginal improvement. And so that's I think the biggest fundamental problem. That's a leadership problem, not a technology problem. Yeah. And is that an AI specific thing? I think that's just generally right you go to these conferences. I remember I think my first accounting big accounting conference was in Boston. I was working with a vendor out there at the time and I flew out to a count. I was just overwhelmed by all the vendors at this place and going holy smokes and you can kind of see a use case and all of them and you sit down and talk to them and you go, yeah, I could I could see a use case for this. And the reason I didn't sign up for all 80 of them or however many vendors there were in this big hall is to me this just seems like all I'm going to do is that the firm is going to stop. And so that's what I think is going to be the most important thing is that the firm is going to be able to do that. And so that's what I think is going to be the most important thing is that the firm is going to do that. So the second major issue is again it's mental it's not about the technology itself. It's what is AI going to do to my business and this doesn't I don't care what business you're in. If this isn't something you've sat down and done a whole bunch of naval gays in around you're probably not making the right decisions. But where is AI going and what will it do in my industry on the accounting front, you know, I'm looking at where AI is going with transaction processing with tax preparation. And you know, I'm looking at that and I'm seeing that in less than two years, we will work totally different than the way we work today. And our clients will have different expectations and very importantly, the pricing pressures and pricing expectations will have shifted. Yeah, okay. If we don't take account of that now and start preparing our firm for not just reacting to that, but how do we live in that world and how do we add value to our clients when the transactions are being done automatically. Then that's how we think about repositioning our firm. It's not about all I can just do transactions faster. Yes, that would encourage us if you're a eyes faster than my AI. Like that's not clients are going to pay for that. No, exactly. And that's the pricing challenge and that's a strategic critical issue. And this is where many accountants and I would say any lawyers and any knowledge, you know, business, if they haven't started talking about this now, it may be too late because these things are going to come and they're going to come in about 12 months. And they got everybody who has not yet thought about it. I'm trying to find right now. I'm trying to find the podcast that I just listened to on diary. It was I think it was Mo Gadat from from like the former Google exec. No, maybe I think it might have been I think it might have been him anyways. This was in earlier in June. And I guess let me step back because he kind of opened my eyes that podcast so to open my eyes on how quick this is coming right. And we have to respect that I think Canada is a very it's a slightly different situation than the US for a number of reasons which we can talk about. But when we look at AI deployment and people go, hey, aren't you scared? Aren't you scared that this is going to take your job because if you go through all of the job replacement, like if you go through the studies. And we talk about it over the last five years to talk about AI and job displacement and job augmentation. I mean, our industry is at the very top of that list of industries that are set to be disrupted, right? But with the example that everybody uses. Right. And why? Well, because we work with data. And so data is very this stuff is very good at parsing data. I mean, I can the use cases are incredible. And we can talk about that. But I have a different approach. I think that this is either going to be terminator, right where, you know, sky net. What is what is the line from the movie where it becomes self aware and I can't remember the date. I should write down the date because it's a famous movie, right? But it becomes self aware and what happens when it becomes self aware and this. Podcast diary of CEO podcast on early June really went through that and what and what this man believes is coming and not in a couple of years like now, he believes that a gentle AI, I think is pretty much on our doorstep. If it hasn't already or not a gentle AI, what do they call a GI, what is that? Like the general intelligence where basically AI can perform human thinking tasks better than any human can. Right. And we've got talk of all this AI are self programming now and they're sort of self developing and you go, OK, well, this is a. of a snowball. And then we've got on the other hand, we've got Sam Altman sort of changing his tune from a few years ago where it was a bit of fear porn going, this is going to replace everybody's job and we need to get ready for it. And now, like recently, he's kind of come back from that and gone, "No, it doesn't have created the job disruption that I thought and I was wrong." And I don't really, I personally don't put a lot of trust in Sam Altman, I believe is a very good salesman. And so you go, "Have you just switched what you're saying in order to sell more licenses or is this actually happening?" And so from our perspective, from a job displacement perspective, I'm thinking about, "Okay, AI is here. It's really good. We're rolling it across our team. I've got projects, CRAs clearly using it because I think we're going to ask them at the roundtable, but the volume and speed with which their issuing post-assessment reviews has been really demoralizing this year because now I have to respond to them all. And every time I see a new one, my soul gets crushed a little bit. Even though I've got a project in Claude that can respond to a CRA post-assessment review in approximately three minutes now, once we go get the data out of the file and in fact, co-work can go get the data out of the file if you trust co-work, which we haven't gotten there yet in our firm. And so everything is going to get faster and everything is getting faster and these major tech execs, I think it was one of the top execs at Anthropic a couple weeks ago, said, "Hey, everybody in the world needs to stop developing this now until and give our economies a chance to catch up." That's scary language from an executive at Anthropic to be using. My perspective and our firm's approach has been, we need to get this out to the team as quickly as possible and figure out and align ourselves with the most ethical AI companies, which is, when we look at the behaviors of the contracts for example that OpenAI is signing with the US government that Anthropic refused for example, that's an indicator to me of ethics and why we switched from OpenAI to Anthropic and it turns out Claude for our use cases way better anyways. But that's not to say that a new player might not come along, there's a new Chinese player that's trying to overthrow Anthropic right now. And so all this to be said is, where is the industry going? Right? So where is the industry going? And what does work for a CPA firm look like in 10 years time? Where are we going to have to focus? Well, I have no idea what it looked like in 10 years time. I can tell you what I think it looked like in two years time. All right, let's have it. With time, the transactions will be mostly managed by themselves. We will be at a point of what I call real time bookkeeping or as near real time as we possibly can. Which the last glimpse of that that we got was blockchain, which that kind of went has it gone nowhere? It went nowhere, right? And blockchain, by the way, requires open banking, which it works, you know, coding transactions in the US is different than coding. And here's one of the key differences. And let's riff on this for a second. And we'll get back to your two year thing is the US generally has open banking where you can AI into a US bank and pull the actual core transactional data, whereas in Canada, we don't. And where our banking regime is a lot more, let's say regulated. And it's trickier to get a bank feed data out of there. And so you went well with blockchain, it solves all that, but that kind of went nowhere. And so US has all our regulate we have indirect tax, for example, right? You can't. So so the IRS in a lot of circumstances will accept a credit card statement as a source document support, whereas we don't. We need the invoice with the GST number with, you know, so there's a bigger auditing requirement in Canada than there is in the US. So really, let's talk about Canadian firms and Canadian, you know, so sorry, back to your two year plans. So yeah. Yes, you're absolutely right. I I've always said that the bookkeeping in Canada is slightly, I wouldn't say harder, but there are more nuances to it than bookkeeping in America. Can we just say we're behind the rest of the world? I actually think it's it's more refined in Canada. Okay. And because of that GST factor, the unfortunate thing for Canadians is that all of these products that are developed for the American market and in the American market, it's predominated by cash basis tax fund. Sure. They don't have input tax credits like we do. And so they don't really have to worry about sales tax. And they can do what I call sledge hammer bookkeeping. Here's a bank feed, slap a code on it, post it. Where your main objective with that bookkeeping is to complete some compliance exercise. Exactly. Yeah. So it's very low value. It's not very nuanced, but it can be done fast. And in Canada, we can't support that because we have these extra layers that are a requirement for our basic compliance. So every business is going to be GST registered. So you can't just slap us a default tax rate on something without some additional insight into that transaction. If you're in BC, oh God help you. You've got now a whole new layer because you've got two types of sales taxes. Or Saskatchewan or Manitoba or Quebec. Yes. Exactly. That's right. So you know, so it's way more nuanced than than the US's. Plus in order to get those taxes right, you almost always want to see a source document because that's the only way you know for sure if you've got the numbers correct. And if you don't get that correct, as you said earlier, CRA is way more aggressive at reviewing or auditing than the IRS's. Yeah. So just for any of our clients listening, that's why we're so much we're annoying. I'm sorry for being so annoying. That's why. As a countance, we have a higher duty and I'm sure you are the same. You know, I want my clients to sleep well at night. I don't want them to be worried that they are going to be audited by the CRA. And I tell my clients, I can't stop you from being reviewed or audited by CRA. But what I can do is make sure that the work we've done is robust enough that should you be audited or review, you have nothing to worry about. Yeah. Yeah. Because we get our job right the first time. Right. And if we go to sledgehammer accounting, we aren't doing the job right the first time. Yeah. And that's a problem. Okay. So, you know, going back to a few use cases, the simple stuff and I spoke about this at the wage point some of the couple of weeks ago and I've got another talk coming up on AI for two geared towards accounts is like this is overwhelming stuff. Like the pace of change is even overwhelming. Right. And I wrote a I wrote a presentation in early May. And by the time I had to deliver the presentations six weeks later, my presentation had changed because you know, Claude had released four new models and the use cases that we could use it for were substantially different. And now I'm just coming off of, you know, pretty well, a month of vacations. I'm hoping to take, you know, a with work in between. I got Starlink. I'm always on. But the amount of brain power that I want to commit to keeping up, it's impossible. And it's almost I almost found myself through my June crunch in a bit of a, I don't want to say depression, but my head game was it was rough there for a bit, man, because I'm trying to keep up with all these changes, trying to properly guide my team whilst trying to keep up with the compliance workflow and increased CRA activity, which we're seeing a lot of. And and by the way, that's clear across the country. We're seeing a lot more CRA activity. I mean, when you spend money like you like we've been spending money, we got money's got to come from somewhere. And so instead of, you know, we got to pay for a billion dollar condo bailouts in Vancouver somehow. So how do you do that? Of course, you would you target small businesses and trying, you know, get them to justify their 123 dollar donation claims on their personal tax returns. I guess that's CRA's approach. So anyways, so that's been a very frustrating. And I've taken a couple weeks off and gone and been in my trailer. And you, I mean, you're, you got a 30th anniversary coming up. I understand congratulations. And you're off to Alaska, right? Yes. And you go, how am I going to keep up? Well, I'm in Alaska, right? Like this is I'm going to be behind it. I'm going to have material to cap keep up on. So it's really overwhelming. And you go, where do you start as an accounting firm? Right? Like what are where do you even start? And I think a lot of people have caught on to it. This is really good at helping me draft letters and emails. But that's just like, that's the minimum acceptable standard now. Right? It's very good at that. It's low hanging fruit. And, and by the way, edit everything that you can to get rid of the double dashes that's just such a sign of AI, right? Like, and the lists, the list dumping. And I can tell when something's written by AI now. So where do you even start? So I'd see there are three levels of AI. Doesn't matter whether you don't count them to lawyer, even a plumber. There are three levels of AI. The first level is the efficiency level. That's trying to do things faster and better. Using it with that, using it to draft emails is a method of doing that, right? It's like, hey, how can I do more with less? How can I get slightly better quality? And in the accounting world, that's where most people are focused. We are trying to do our coding faster. We're trying to, you know, have some skills or some scripts that get run that that, you know, like you said with your CRA, I've got one of them just the same with with a CRA response letter, you know, how do we do more things faster? The problem is that all the research shows us that that's going to get you 15 to 20% cost savings and that's about it. Okay. The thing with going after cost savings is there's only so far you can shave. It's not endless. You can only go down so far. The further you go down, the more you start hurting your customer service delivery. The second layer of AI use is the strategic thinking layer. A lot of people, especially the knowledge professions, have worked out that Claude or ChatGPT is really fantastic at helping you see things you didn't see. This is where I spend a lot of my time. Whether this be writing content, whether this be evaluating things that you're doing, letting it coach you on an example. Every time you have an interaction with somebody, I now record all of those meetings with a trans-criber tool. After that meeting, I can go back to Claude and I can ask Claude to critique that conversation and I can tell Claude the things that I was thinking about and the things I want the most advice on. It will now either interview me or it'll tell me how it perceived that conversation. Even when I say I'm really interested in how I handled this particular area. It'll tell me that. Then it'll tell me, "Hey, here's one thing that you didn't ask me about that you actually need to know about." Interesting. Yeah, it's really good. So our teams could be using that for their client calls to say, "Here's what you missed. Maybe think about talking to the client about that next time." That's a really good call. I love that use case. There's different layers even within that. You can tell you this, "Chasic thinking at the higher level of where are we going?" This is what I recommend every business should be doing today is asking, "Where is my industry going? What are the risks facing my industry? What are the pricing pressures that will come in six months, 12 months time?" If you haven't thought about those, nothing you do at the lower end will solve those problems. Then you could do it more on the personal coaching side. If you're a manager and you've got staff and you want to have a difficult conversation with a staff member, Claude or ChatGbT can coach you through that conversation. You can do a dry run with them first. Or if you've had a conversation and you want to know, "Hey, maybe it was a sales conversation. I did this the other day." I booked a sales conversation. I recorded it. At the end, I went back to Claude and I gave it the transcript and Claude shreds off me. That's a really good idea. A lot of sales conversation. I want to follow the spin process. Tell me where I got off track and tell me what I could do differently next time. It's fantastic at doing this. Here's the thing. It doesn't mind offending you. But you have to give it permission. Most LLNs are what we call sick authentic. They are trying to please you. I have a standing instruction in my Claude.endee file. I'm master prompt. I tell it that your job is not to please me. Your job is to critique me. Your job is to help me grow my business. Your job is to help me see things I have not seen for myself and to tell me where I'm off track. Your job is not to tell me that everything I do is great. It's natural tendency. This is great. I haven't had to hire you and I'm getting all these great tips. I'll send you an invoice afterwards. The first thing was efficiency gains. I want to riff on that a little bit more. Second thing was personal coaching or is that the third thing? The second thing is strategic thinking. That can be everywhere from what's happening in my industry to the coaching help me think through this problem. That personal coaching is a great example of that. The third thing is growth. This is where I think we're seeing a lot of startups in this space but I think a lot of established businesses have not yet seen the advantages of this. Using AI to help you ideate, create and deploy new services or products that you didn't have before. All the research on where there is the ROI in AI, this is the area that we are seeing the most results in. I said with efficiency, you can only gain so much efficiency by shaving things down and making things faster. If you're looking to start a new service line, you've got a zero baseline, everything is upside. There's no ROI because there's no ROI. There's an ROI but you're not comparing it to anything. You're creating something new. The ability to use AI to create new things is absolutely incredible. You can write a book in a day. You can create a training program literally in a few hours and it's good. I would never want to just have it create it and then you just launch it but if you give it the right material and work with it, that's why I love that phrase co-work for Claude because that's what it feels like I'm doing. I'm co-working with a trusted partner. You can create phenomenal material that's stronger than you would have created yourself faster than you would have created yourself and easily pivotable. If you do something and you pile up it and you go, "No, that's not very good." Or I've got some gaps here and again, take that back down to the lab before. Take the recording of what you delivered. Ask people for feedback in that session. Take that recording and now ask AI to analyze your live session and give you pointers on how you might change the content to improve it. So you could have this dynamic feedback loop and you're now operating at this third level. You're creating new things that could generate new revenue. The thing that's going to happen to a lot of businesses and accountants are not immune from this but I think it applies maybe a lot more to some other businesses as well is that our businesses, there are going to be people whose businesses shut down and there are going to be whole new businesses that get created. Thousands and thousands of new businesses are going to be created because of AI. It doesn't have to be the dichotomy of close or start up. It's reinvention and this is where as accountants we have to ask, how do we reinvent or transform our business so that we can still add value to the world in a world that no longer values the compliance because the computer takes care of most of it. Yeah, I call it data grinding. That's super interesting to say that and it goes back to a previous point that I left unfinished, which I do often terminator. That's the one option but the other option is like we're building railroads here. I talk about building railroads and why do I talk about that? Well, if we go back to the industrial revolution, it really kicked into high gear when we were able to start transporting stuff across large distances, machines, metal, whatever. Now we can trade with different regions more than fur because fur, you load it on a cart and you have a horse dragon across the country. Well, now I've got a railroad. I can transport people easier. I can populate the west. I can transport machinery and then everybody at that time, I think there's some writing that I've read that people are scared. This is going to take our jobs. We're going to have a whole flood of immigrants or whatever coming to our region now and taking our jobs. That was a main fear then. Then the internet came along and in 2000, you go, "Holy smokes is going to end the world." Why 2K all this? It ended up creating more and more and more. I'm a railroad guy. I think that we have a huge, huge opportunity, opportunity of our lifetime to use this technology to further humanity and make our businesses better. The delta there between closing, staying open, that middle ground, the difference is going to be the entrepreneurs who choose to embrace it and learn it and deploy it across their teams. We're going to fail at it, but you got to iterate just like anything new you're trying. The only path forward that I see here, and if you're a Terminator guy, this is a bit scary, but it's to use the tools and to learn the tools and the entrepreneurs that learn the tools and use the tools will win the day. I also see this coming back to, you go, Microsoft Word, I can't add, what do we need that for anymore? I don't know why I'm paying for that subscription anymore. Because Claude's better at it in all material respects. It's just better than Microsoft Word. I can build my own AI and host it on a server. I can buy a Nellah Lem and host it on my server. I don't have a server, but you know what I mean. I have and give my team access to that and start dumping subscriptions. I can even have Claude start writing software that I am paying licensing on otherwise and start saving my, I think we've spent 130 grand on software last year across the whole firm, including client software. And you go, how, how, you know, when you talk about efficiency, I'm thinking on that level too, rather than just helping me write emails. It's like, well, can you help with website coding? Can we write a new client portal? Can we, you know, all the things that we We can have a real custom solution now that we had to buy something off the shelf from before. We can just program that ourselves and our subscription costs go way down because we've got something in house now, right? So I'm a railroad guy. I think there's a railroad. And going back to this podcast, this diary podcast, they were saying that with this next level AI, which has likely already arrived, they're going to see serious job disruption in the next 12 months. That was the thrust of this podcast and you go, okay, but then this guy, MogaDet, at the end of the podcast, he kind of got into the light, which is 10 years time. It's a very different world. We're going to come out of this fog. There's going to be a lot of chaos in between and we'll be better for it. He's very hopeful and that relies on sort of a, what do you call it, an AI that's apathetic to humans. That's right, right? Apathetic is that the right word? And I think there's a better word for it. Anyways, where this is the new world and we're creating more because of it and maybe we can do better at space exploration or maybe we can do better at resource development and development of things that are good for us. Hopefully, that's where we're headed. Can we wrap on security for a minute? What do you think about that? Obviously, I'm not worried having a cloud-based LLM, cloud in the cloud. My whole team has that. I'm scared to roll out co-work to them. And here's why I'm scared. Cloud co-work can drive your computer. So it can use your CPU on your desktop computer to move files, edit files, delete files, execute files. And it's the deleting and executing and moving and all of that that I'm kind of nervous about. Or what if this thing just goes rogue, which we've seen AI go rogue lots? It just does weird stuff sometimes. So how do you, you don't have to answer this, but give me some ideas because I know that maybe there isn't a clean answer. But how do we safely deploy these tools across our team? Yeah. But this is one of the challenges with cloud right now, is it's a great tool for personal use. It's a little harder to deploy at the team level. You know, you could be in the enterprise system and they have some extra tools in the enterprise system to help with that. But they still don't solve what I call the client confidentiality bleed problem. And there's two aspects of this. One is the going rogue side of things. Now there are ways that you can minimize that. Okay. The thing I'm more worried about from an account perspective is you've ever used chat GPT back in the day where they switched on memory. And now it was remembering things from prior conversations. Okay. And that was wonderful because all of a sudden you realize that chat GPT knew you so well. And then you have a conversation with it and all of a sudden something you would you would ask that a week ago about client A started surfacing in a conversation that you were asking about client B. And you suddenly realize that memory is not always a good thing. Yes. You want isolation memory. You want it to remember what you don't want visits a bleed memory. Right. And so just at the basic level, let's just throw this out there. You need to have the right subscriptions if you're going to use this for business purposes. Okay. So in cloud that's minimum the team plan. If you've got anything lower than the team plan, you can't use this for client data at all. And I'm not saying that you can use it for client data in any other subscription level. Go check with your IT provider and your insurance company. It's on you, but the firewall of training larger language models comes around your environment when you hit the team level. And I think every AI has that level that you can sort of firewall your data, but you're talking about that at the pro level, you know, the not quite team level where you're not training the model. Okay. And I think that's the baseline. So when we talk about security, there's three levels of security that we need to be aware of. One is, am I training the model? Okay. And anything you do that trains the model is a no go, especially for an accountant. You never want to put proprietary information, your own intellectual property. You never want to put client information, and especially personally identifiable information, social security numbers, social insurance numbers, stuff like that personal revenue numbers. You never want to put there in the system with our training on that data. And one thing people need to remember here is that you know, people often say, well, you know, why would I upload that to the web? You're not uploading it to the web in the same way as you are like publishing a website or posting a social media post. No one's ever going to be able to see that document that you uploaded or that prompt that you uploaded. Sure. The ones we've been trying to avoid. And without these protections. Well, even without these protections, if you put in a prompt and you include some client confidential information, no one is ever going to see that prompt or see that document by itself. Let me give you an example. Let's say you're doing a tax return and you've got a famous client. Okay. And you know, will you selling the on because she's Canadian. Okay. Let's say you're doing a tax return for selling the on. And so you put a T1 return into a non into a training, a training model prompt. I like a free model. And you ask it to critique the return and tell you if you've left anything out. Okay. Great. Seem innocent. No one's going to see her T1 return, her tax return. Okay. However, one day a month from now, someone might be googling or going to the AI and asking, what Celine Dion's net worth. And all of a sudden in that response might be, well, in 2025, Celine Dion earned $786,000 or where did it get that from? It got that from the data you submitted. So you haven't posted the T1 to the internet in the way you're thinking of, but that data now becomes part of its brain, but it can answer based on part of its brain. Yeah. Okay. So that's why we never use a model where your subscription layer allows the model to train on your data has to be non-negotiable. Okay. That's like the minimum. That's the minimum. Exactly. At the on the right plan. Yeah. You've got to be on the right plan. You've got to ensure that your data isn't being trained. Okay. The second model is data retention. How long can the LLM provider retain that data? By default, it retains it for quite a long time. And anyone in the account who the legal profession might say, well, actually, that's not what we want. We want to make sure that that data is in the retention's good. You want it to retain data for a very limited period of time. Because it's better. It makes you like easier. That's easier. Go back to it. You can come back tomorrow and go, I want to continue that conversation I was having yesterday. And if that conversation's gone, it's like, well, what you see is you're now going to start all over again. So you want retention. You just don't want retention forever or you want to know the terms you're retaining data. Okay. So we've got data, we've got feeding the model. We've got data retention. And then the third layer is this data segmented enough that it's not populating across clients. And that is the hardest one to work around. Because if you've got a model that's generally got memory, that's a problem. Now, this is where co-workers really good. Co-work when you're working in folders only has no memory. So there is no data bleed. If you're working in a folder for client A and that's your root folder and you ask it a whole bunch of questions and give it a whole bunch of documents that you would normally be confidential, it's looking at all that information. You then change your working folder to client B's working folder. It can't see. It's structural. It can't see without your permission the data in that other folder. It's limited to this folder. So co-work, I like from that perspective. Because it builds some architectural barriers around the file structure. Now once you throw in projects, projects is where the memory gets added. So if you've got a common project called client work, well now it's able to think across clients. So you've got to watch that. And then the other problem in co-work and any of these LMs is the MCP layer. So for those that aren't aware, that's called model context protocol. This is how you connect other systems into your AI. And the problem in most AI tools is that those layers are universal. They apply to your account, not to your session or your working structure. So if I've got, let's say I've got a client that uses Shopify and I've got a Shopify MCP and I'm able to connect up to client A's Shopify account. And then I pick up client D and they use Shopify. But I haven't connected Shopify because usually I only have one connection type active at a time. And so I'm talking to an AI about Shopify. Well the AI doesn't necessarily know that this is client D, but that account only belongs to client A. And you might find it starts to pull in data from that connection because that connection is universal. That's an awkward client meeting meeting that happens. Absolutely. Yeah. Oh yeah, your Shopify data doesn't line up with your accounting data by $300,000. You can't work at Y. And then all of a sudden you realize that I just aren't even the products I sell. So where's that data coming from? Yeah, yeah, yeah. Yeah, yeah. Interesting. Yeah. Okay. So this is the, you know, the harness that's available. technical term, the harness is the sandbox that you work in. And none of the tools have yet provided us with a harness suitable for client accounting work. Microsoft is probably the closest and there's the worst model. But they have at least because they've been in the enterprise space for so long, they are the closest to be able to create a non-bleedable client harness. Interesting. And then what about is there a way to help me go, well, what if one of my team connects to a client folder and deletes, it deletes a whole bunch of stuff? That's main fear A with co-work. Main fear B with co-work is that through some social engineering, an executable file ends up on one of my team's desktops and gets and cloud executes it. And all of a sudden we're in a world of pain, like a virus or a fishing, whatever. Who knows? Yeah. So again, with cloud, we've got a number of layers we want to think about the element. Computer uses a separate layer that I didn't mention earlier. Generally speaking, cloud isn't going to execute an executable file. So, and first of all, it's not working on your desktop. So that risk is reasonably minor. It's still present, but it's not as bad as people assume. The data deletion risk, you can actually switch off data deletion as a policy in your, at least in the enterprise layer. I'm not sure about the team layer. This is where you actually look at, well, where are we storing our data? Are we storing our data in a system that we could actually roll back and manage? Now, whether that be a shared Google Drive or a box account or some other synchronized system, so that if something goes rogue, you know that you can still go back and get that data. So we're back to backups. Like this is a 20-year-old conversation or 30-year-old. I don't know whether it will ever be away from backups of some version control. Version control. Yeah, it's the same idea. Because at the end of the day, you never want to have a single point of failure. And a user that just clicks the wrong button or an AI that thinks it's okay for me to delete that file. You never want to have that back. Well, and so in co-work also, another thing you can do is say, listen, every action that you take, you can set this up as a company policy, every action that it's set to take, it has to run by the user, right? Yeah. And get approval for it. You can turn that off and say, yeah, I can do whatever it wants. And on the one side, that's pretty good because you know that somebody actually has to review every action that it's taking, which is great. And now somebody is responsible for whatever happened. You've got a person that's responsible for that. On the other side, I've had it go and do web research for me to help me with my presentation or whatever. And every single new web page that it wants to open, I have to approve, approve, approve, approve. And it just it's irritating, right? So there's a there's a there's a trade off there. Yeah. Yeah. And the the computer uses probably the area or the browser use where it's really uncertain as to, you know, how secure it is. Because, you know, there's a whole threat of what we call prompt injection. I'm going to the details there where it pulls data that might try and tell it to do something. I would say that's actually getting less and less of a concern because the the LMs are aware of this and they know how to filter the stuff out. I'm more worried about the reddit problem. I mean, there's so much bad advice on reddit. Yeah. How does the AI when it's looking and searching the web, you know, how does it know that reddit advice generally bad CRA advice probably okay, you know, how does it rank that? I'm more worried about that than I am about prompt injection in and of itself. But then you've got the whole computer use. I mean, if you're going to give your your your AI the ability to open up an application or go to website and log in and do something, how do you know that it's only doing the things you've authorized it to do? Right. Yeah. And again, we're still brand new. I mean, this this stuff has been around for what six months. Yeah. So it's still brand new. What if it sends an email to one of my clients in the middle of the night? Yeah. You know, and it's and it you know, what I you know, I could see a world where cloud thinks it's being really clever and sends an email firing all my clients overnight if you've set it up incorrectly. You know what I mean? Yeah. Or you know, so okay, good. So that's that's kind of. And this really just comes back to a right some permissions issue. I mean, this is the same stuff that network security guys have been dealing with for 25 years. Probably for 45 years. You know, what who has the permissions to do what in a certain environment? Sure. That's what it comes down to is how do we trust our agents or our AI? What permissions do we give them? Are they allowed to draft messages, but not approve them? Are they allowed to send messages? You know, are they allowed to post stuff online? Are they not allowed to post stuff online? Where where do we put the limits and and how do we enforce it? And one of the challenges a lot of businesses say, Oh, I can tell my LLM to not hallucinate and it won't. I'm like that that that line never works telling me LM to not hallucinate it hallucinates because it doesn't know it's doing it. Yeah. Yeah. And that never works. If you tell it, so you know, a rule put in a prompt is what we call a soft rule. And AI ignores the rules you give it frequently, the soft rules. So the question is how whether the hard rules, whether the architectural constraints, and you can program those. You have to program that's anthropic programming that isn't it? Or is it if you're using something like Claude, absolutely. They have to build in those frameworks. And so this is where you know, but for accountants, you know, we, you know, as much as I love Claude, I'm not sure that Claude is the best environment for us to be running our client service and our team deliverables from for all these reasons because we don't have a secure enough rights managed harness to ensure that we have data privacy within that setting and restrictions on what the agents can and cannot do. So because of that, we're looking more at a third party tool that's been designed explicitly with those issues and might. And they are out there. They are coming. Right. Right. But now I'm back to a software subscription that I didn't want to, I was trying to get away from those. Right. Yes. So, okay, good. The future of the industry, I want to riff on that just a little bit before we, before we button things up here, or two button things up here. I see a world, I see a renaissance of human connection. And I frankly, I believe it started during COVID. We got very lonely in COVID. We messed a lot of our kids up during COVID by not allowing them to go to school and interact with their peers, I think. This is my opinion. And in fact, we, you know, are from our from hosts an annual charity gallery to try and give a bunch of money to youth and adolescent mental health initiatives at the Alberta Children's Hospital that they're doing. Because I think that we're, I think this is a scary new world. And I don't know, I got an eight year old and a 10 year old. And I don't know what their future looks like, you know, aside from learn to program. And maybe learn how to swing a hammer. But even that's, you know, Elon's trying to solve for that as well. And so the future is sort of unclear. And the only the only way we can really deal with this is to try and understand it, which is, it's ununderstandable in a lot of cases. So, future of our industry with the, with a renaissance of human connection that I believe began during COVID and it will only be accelerated through this. And well, what does that mean? Well, I think that the jobs that will be safe are the ones that have a high degree of human connection. Because I think we're going to be craving that as a species. And I think frankly, it's the only way we survive it is if we, we step back a hundred years in the client relationship where it's very much human centric meetings are so important coffees in person becomes so important because now you're stickier than the next accountant that comes out there who's using tools almost exclusively and doesn't actually want to meet with you ever. And we did it, we're guilty. We, we've tried to kill a lot of meetings through a lot of processes that we've developed. It's a huge time saving. And so how much, how much stuff are we discussing during those meetings that can be delivered in other formats other than a meeting? Well, great. Now what is the meeting for? And, you know, I just wrote a two-part article series on this exact topic. Okay. And the, and the, the genesis of this article series comes from that, and I'm sure you've had the same thing. I'm sure most of our listeners have had the same thing. If you're using the meeting recorder in your Zoom meetings or in whatever tool you use and you show up at the meeting and both you and your other person both have their recorders in the meeting. And the standard jokers will one day, our, our meeting record will be able to have the meeting without us. And we all laugh at that. And the reality is that's quite possible. And so it raised for me a couple of questions. First of all, if everything is becoming an agentic, what does an agentic accounting firm look like? If our clients expect to be able to interact with their bank and with their other partners in, you know, through their AIs, what do we have to do as accountants to be in that same space? But then that raises the next question. Should our recording agent be able to have the meeting without us. What do we miss and lose when we are no longer face to face? And that's the human aspect of the conversation. Sure, we love the automation. We want the technology to work seamlessly. We don't want to have to log into 26 applications to do everything. But at the same time, how do we keep it human? Because keeping it human is the way we keep the clients, where we keep that relationship and stay sticky, as you said. And so we have to be very, very cautious about automating everything. And ask, what is it the client is really paying for? And how do I make sure that not only do they get that value, but they feel that they're getting that value? And today, the way we do that is by talking to one another. Whether it be a phone call or a Zoom call or the in-person coffee, you need to be able to make sure that you've got that human connection that's never superseded by the AI. It can be supplemented, but never superseded. Yeah, and so now, if you model that out, you go, what are we going to do? I had to spend the first four years of anybody who doesn't know this, like CPAs know this, but I'm a CACPA. So in order to get those letters behind my name, I had to be a data grinding grunt. I had to take a step back in my career financially and go and make minimum wage. I don't know if they're still paying poverty line wages for starting articleing students out of university. And I had to data grind. And the most successful people in this industry aren't the people who are super good at data grinding. They have it as a baseline so that they could get logic and judgment, judgment's a huge one, logic, judgment, reasoning, et cetera. You develop those skills through data grinding. That's the old way. And the most successful people in this business, let's say the partners at the big accounting firms, the big four, the reason that they are partners at big accounting firms is because they have their very good relationships, generally speaking, right? And they're very good at going and taking their client to a hockey game and making them feel like they're worth a million dollars. They're number one client. That's how you make that connection. And that's how these big firm accounting firm partners. It's the human skills. And so that's always been present, but the vast majority of us never make that level. And the vast majority of us will love the data grinding and will bury ourselves in it and enjoy never having to deal with the client, right? The trouble is is that whole, that's all going away now. You don't have to data grind anymore. That's over. How do we train our articleing students and the future of the profession to have interpersonal skills mixed with logic, reasoning, judgment? What does the training program look like? If you don't have to data grind for four years. And I don't know that anybody's got an answer to that question. I haven't seen CPA Alberta or CPA Canada come out with answers to that question. I think they need to start answering it pretty quickly because if not, the value of my of our letters is going to decrease if we're not focused on training people for the right things, which frankly isn't it. I grew up in a sales background. So in order to put food on my table, I had to sell to people and I don't believe selling is a dirty word, by the way. I had to sell to people. Well, I mean, number one rule in sales is the client's not always right, but the client is always a client. So you need them coming back, especially in our business where it's largely relational and largely recurring. So just, I don't know if you have any thoughts on that, but maybe a closing comment on the future of the profession and how we're going to train our teams and have you put much thought into that because I'm kind of stumped on it, right? Yep, so two thoughts on that. Number one, the way we learn those skills used to be by reps, you just do it thousands of times, over and over and over again. That's going to go away. So how do we replace that? Well, we have to do what airline pilots do and they learn on the simulator before they learn on real people. So we have to find ways to say, how do we simulate the real life of working on clients when we no longer have the opportunity to do the transaction coding in the same way? So create, we've got to look at simulation as a training model. The second is that this new model is going to revo- we're going to require review at a different level and type than it used to. So this is where people are going to start to learn those reps or perform those reps. It's going to be in review. It's going to be much higher volume across multiple clients but on the individual client, it's going to be lower volume. So how do we train people on understanding the client through review of exceptions and anomalies rather than the detailed transaction processing? And I would actually say you actually learn more in review than you do in transaction processing. I don't disagree with you. The moment when my light bulbs started turning on and I actually got is when I hit reviewer level. Yes. Absolutely. And so I'm looking forward to creating a T1 preparer which we're currently working on where Claude can get the baseline T1 preparer or 10-4-year cross-border return prepared. And now all of the people who used to have to parse all that data and enter it, which is our biggest bottleneck by a mile, like when we have to go through a mountain of client documents and figure out what we need and how to enter it into the tax return. Claude can do all that. Well, now all those people that used to do that can now move to that reviewer phase, which has always existed in accounting firms. We're almost eliminating a whole tranche of prep level which should have the knockoff effect of we can do more with the same people. I don't intend to let anybody go ever. One of the things that gets me going in the morning is how many families the collective we are putting food on the table for, right? And so the job isn't, the idea isn't job killing here. The idea is doing more with tools we have. But inherently that means job killing because we're not hiring at that level anymore. So in a roundabout way, you're going, but I also think that the pie is very large and there's so much as a profession that we are not doing for our clients that we could be doing for our clients. And the main constraint has been how much they're willing to pay versus how much energy needs to go into it. Not every client of ours can hire a full accounting department for their business, their small business. That's why they're hiring us. Well, we can do way more now with the same tools which you're not going to charge by the bill of the hour anymore. And so now you've got these fantastic tools that are just going to allow you to create way more value for your clients if you harness them properly with the same team, right? We just have to think about this a little differently. Yeah. And next, the model I teach is, yes, you can cut costs. You could cut staff. But if all you do is cut costs and cut staff, you're eventually going to get priced down where you lose that margin. Race to the bottom. Yep. It's a race to the bottom. If you can reduce your costs, yes, that's going to be necessary. And then invest that extra time in advising your clients and help with them run a bit of business and helping them see the things that they can't see today, you will have a client for life. And I talk about the movement from the rear view mirror. Only if you're able to connect with them on a human level. Exactly. Absolutely. Yeah. So I talk about the the accounting model and we'll close with this moving from the rear view mirror to the dashboard to the one screen. Now the rear view mirror analogy, I think everybody understands. As accounting for 200 years, we've driven the car with the rear view mirror. Last year's tax return. Last year's annual accounts. Last months, financial statements. Last quarter's GST return. That's the rear view mirror. We're moving very fast to a world of zero day or one day close where transactions are able to be kept up to date on a near real time basis. Well, that's your dashboard. How fast am I going? How much gas is in the tank? What control lights are on? That's your dashboard. But what clients really need is they need a co pilot that's helping them see up the one screen with a heads up GPS. The GPS tells you what's my destination? What route am I going? The co pilot tells you here the turns that are coming up. There's road worker kid. Maybe we go around the block for a little bit. By the way, the cop loves to sit just around that corner with his radar gun out. Make sure you're doing the right speed. OK. And by the way, you've done this route. I've done this route 75 times with other clients. Here's what most people miss on this route. The best coffee in town is at that store there, not the one up ahead. OK. And so having that co pilot who's helping you drive and look out the windscreen, not down at the dashboard and not behind you in the rear view mirror is I think where as accountants we have to focus. Because that's how we add value to our clients. And we can only do that with solid real time financial information. I love it. Thank you so much. What an enlightening conversation. I've got a huge list of notes on on I might just have to hire you here to help us with our deployment. Because there's just so much to think about. And going through it step by step is frankly daunting. And teaching our team how to how to we had a we had in all hands the other day about phone calls are cool, right? Just pick up the phone, right? And so, you know, and so thank you so much for. for making the time to comment chat with us all about this. Where do we find you, Peter? - Well, if you're just looking for general advice, I'm at fuelaccountance.com, Peter at fuelaccountance.com. - No, you can't go to that one, we're comparing. - I know that one, I can't go on that one. - But if you'd like a little bit of advice on your business from AI perspective, that's probably the first place to reach me at. If you're an accountant, what's into this? Then I publish at the aiaccountant.ai. I do a weekly roundup of news in the AI industry. Every Monday morning for a accountant, I take the techie stuff and I take the industry stuff and I pair it down to what matters for a cares practice, as well as other court leadership stuff as well. - Right, and we're saying the aiaccountant.ai. Peter at the aiaccountant.ai. If you just want to reach out and email Peter about getting your practice on the right track, thank you so much, Peter, I really appreciate it. - My pleasure, I was grateful to be with you.