472 AI’s Impact on Accounting Firms - 5 Year Roadmap Q&A

A fun one today. I recently had the opportunity to speak to a room about 300 accounting for owners or so in Tampa, Florida at a conference for Pasba. Pasba's like a membership group for accounting firms. It was a Q and A on all things AI. Super sharp room of people, which I'll I think you'll see from the questions here. Digging into just like the very nitty gritty day to day how to actually put AI to work for my firm in a meaningful way. And I hope you can come out of this with a couple of ideas because this group specifically, they had just gone through several days of pretty intensive AI stuff. I like sessions and training and all that. My hope is that I can kind of like mentally get you to that place where they were thinking about bigger, more complex problems to open your mind a bit about what AI could be doing for your accounting firm. So without further ado, check this one out. I'll also link pasba. Down below is a really cool group if you're looking to share the firm running journey with some other smart folks. Thank you for that warm introduction. It's great. It's so good to be here. Thank you for having me back. I'm glad it worked out. I love pasma, man. Like PASBA is such a kind of eclectic collection of firms and community is so near and dear to my heart. It just makes such a difference to know people that do what you do and be able to bounce ideas off of them. So. So I love this. So excited. There's so much AI enthusiasm. Obviously I'm a big time nerd on that stuff. But like Carl said, if there's. I imagine the most helpful thing I can provide is probably like you're real close to being back at the battle station and reality sets in and you start looking at the inbox and you're like, what do I do now? Tomorrow, I think was yesterday, I think was very brain expanding, got a ton of ideas. But I mean we've all been to conferences before where you didn't stick the landing when you got back. So that's probably the most constructive version of this. I would imagine. Anything I can answer or build on or disagree with from what you've covered thus far to try to help you stick the landing when you get back to back to work.

Yeah, we got Craig Howes over here.

Okay. So I could break it down in two different questions. One is either a one year or five years from now, we look back, what tools are we going to want to implement when we get back? So at least to have a soft landing from all this AI stuff that's going on today. So one year from now and then maybe even five years from now, the firms that are going to look back and go, man, I'm so glad I did this in 2025. For me, the easy answer is still chat GPT. I think the, the important thing is the breadth of the rollout. Last year, I mean, last year it was, you know, it was a portion of the room where the, the owners, the most progressive thinkers were using chat GPT 2025. I think we've hit a tipping point with deep research and improved models where it's almost a moral imperative to have it in the hands of your team. Like if I'm going to employ somebody the next three years and I didn't introduce them to AI in that, in that time, I mean, your skill set, like it's, it's getting less relevant by the day if you're not learning this stuff. I don't know, ChatGPT is kind of like the chicken Parmesan answer there. But I would say the breadth of who, who you get it to everybody in your firm, that's the big thing.

So a kind of a follow up on that. If we're going to go back and you know, pour our heart and soul into, you know, building out some tools and empowering our staff and you were to look at the state of AI in the different engines right now, I mean, how would you rank them for suitability for small business accountants right now? Or should we not get hung up on that, that the tools are just going to evolve and we might have to shift gears.

So I always come back to ChatGPT because it's the only platform that's been at or near the top from day one. And I mean, Microsoft has said from the beginning we've got all the same models. But if you've used Microsoft's AI products like it is rough, so it's impossible to know who's the leader going to be 12 months from now. Honestly, the bigger thing is just literacy, like understanding how to use the tools and embedding that expertise into your entire team because that stuff transfers like there could be Google's crushing it right now. They've got some really good stuff. Their bigger problem is almost the user interface and pulling it together in as compelling of a way as ChatGPT. But their raw models are actually better than anybody else's. ChatGPT. I always come back to it because it's, it's, they're doing some interesting UI stuff like custom GPTs are unbelievably good and they've always been at or near the top. The in terms of which models to use, you don't want to have to worry about it too much. And right now it's kind of broken down to what's the best general purpose model. And that's, that's 4O and then what's the best reasoning model. And I'm sure reasoning models have come up at this point. But reasoning models, basically they think for longer. It's for more complex tasks. You just always got to know what's the best general use model, which is usually for creative stuff or things that don't require thinking and what's the best reasoning model. ChatGPT makes it super easy to pick which one of those are best. But like a month ago we got new models and those changed. So you kind of just have to be thinking at any given time know what those two models are.

So should we spend a lot of time thinking about which model to choose for interacting with because you know, you've got the drop down and you've got what, four, four or five models plus deep research that you can always initiate too.

We should. Makes a big difference. Also being mindful of when to enable it to go out to the web because it will, it will sometimes guess, do I need to go get information from the web for this? Other times it'll just try to make up an answer. So I'm doing dishwasher repair two months ago and I take, I give it a picture and I'm like, here's the problem that I'm doing and I'm trying to find this part. What is it? And it just gives me a response, but there's no citations. And I'm like, that surely can't be right. I don't think it actually knows that I did the exact same thing and hit the little button to enable search. So it goes out to the web, impose results. And it like went to a part manufacturer website. I found the exact thing. But you got two people in your firm. One person knows that one person doesn't. That one person's like AI hallucinates the other person's like, dang, that just saved me a whole ton of time. So like, unfortunately it's still fiddly. Like there's little things like that that'll make a big difference.

We got John over here. One quick question while they get him a mic. You know, we pulled the room or I think I'd asked the question of who's afraid of AI, who's deployed it? There's really quite a low level of fear here, but we're primarily owners and leaders in this room. But the next layer down of our production, how do we address that? When we bring this back and present these tools with the risk that people are going to interpret this as I'm just going to train this thing to replace me type of thing.

Your team is very scared. Like they are so carbon, the practice management system, they do an annual AI report. And that was one of the big things that come out of it was, you know, fear. Fear versus excitement from managers, fear versus excitement from your team. Like they were polar opposites. Honestly, I think all you can do is get them using it, get them comfortable with it. There's still issues around what are my team's incentives? Is there anything in it for them to be more productive? And if they see it as a tool to be more productive and maybe that means upside to them, that's really exciting. So if you're putting it in front of them and they're still feeling that fear, like, consider what are their incentives to actually leverage it.

You said the right words because you're at pass, but are that we are supposed to pay team members incentive pay based on their billings. Right? So. So it should incentivize them. But now my question is these team members, like, I'm not the guy AI of like software, things like that. It's not, you know, I can ask it football trivia, but you know, that's about, that's stretching it, right? Who teaches my team? I mean, I can't even teach myself. Do I hire a W2 person and say, okay, I'm going to pay you a quarter million dollars to come work for me, and you're just my AI guy and you, you put all the AI into my firm and you teach my firm how to use it or how. What do I do?

Who do you all perceive in the past group as being ahead of the curve on this?

Matt Patrick, we.

What, what would you say, Matt? I would say experiment. Get, get everybody bought, Learn it together. Give some incentives for people making, taking some initiative to learn it in your.

Practice, make it a part of your.

It's the same as any other training mindset you have of, hey, we're going to go, you know, learn somebody. Hey, you're gonna go try this one, you're gonna try this one, you're gonna report back to us, have some contests, make it fun. I think you, you wanna have a environment of experimentation. I Think we, because all of our other tools required training in a different way. We, we kind of block ourselves. Totally honest with you, right now, all I would do, like go out, get ChatGPT team for everyone on your team. The team plan comes with more security. Have them watch my 20 minute video of all the features you need to know about, cut them loose. Then it becomes a question of what's the feedback loop. So I think you need to meet every single week for say 30 minutes and be like, hey, anybody have any cool use cases? Because that accelerates people's learning. But to be clear, every, everybody's doing this, this kind of name dropping. I was at a dinner talking with the Warby Parker CFO a few nights ago and he's like, we gotta get our finance team more up on AI stuff. We're not sure what to do. We're not sure. Do you come in and would you come and help us with implementation? I was like, does anybody have access to AI tools right now? He's like, no. And I'm like, then, then no. Like step one, make the boundaries, put it in their hands, create feedback loops and then figure out what you need. But, but the hard thing is that's the opposite of say implementing a practice management system, right. Which has to be very top down.

Where do you think AI is overhyped today in accounting or in general?

And where do you think people are.

Not paying enough attention?

Good question. I think accounting AI is very overhyped. I think there's. Today there's virtually no accounting AI that I'm excited about in the accounting vertical. All the stuff we're doing that's really cool. It's chatgpt, it's meeting transcript stuff. It's like accounting AI specific specifically is very overhyped. I think. What was the second one? What should be paying attention to meeting transcript stuff I think is massive. You need to be capturing your meeting transcripts because there's so many things we can do with them. This is, I don't want to get too like 400 level, but chat and like ChatGPT, that's the gateway drug because anything that you do with ChatGPT, you can create automated systems that will do those things for you. So to give you a practical example, let's say I've got a meeting recorder and, and we may have even talked about this last year. Let's say I got a meeting recorder and I've, I'm like, I need to post more stuff on social media. And so you take the transcript from your meeting recorder you go chatgpt, say give me 10 ideas for blog posts or tweets or something like that. But like that requires you going out and doing all of that stuff. But let's say you kind of get that nailed down. You got a prompt that you like and it gives you reasonably good ideas. Well, you can use a tool like, like Zapier, like make like Gumloop, like Lindy. And I don't know that I even recommend you build this as much as maybe pulling an agency to help you build some of this stuff. But if you've got that prompt nailed down and you have a personal goal to be posting on LinkedIn every single day, then we need to create the infrastructure that will automatically send that to you in Slack or teams or an email or something like that every time you get off a meeting. So like we find the cool use cases by testing it and using chat. But how do you make that part of the infrastructure of your firm? Like then we get into thinking beyond chat and building that stuff to kind of happen automatically. And I don't like, again, I don't know that that's something like you got, you got big time geeks like Chris, where that, like that's where we first met was, we were talking about like zapier for accountants and stuff like that. If you're that knock yourself out. For most people it's like go out and pull in an agency that can build some of this stuff for you.

So can you talk about the future of the accounting industry over the next five to ten years? Uh, and more specifically, how does that change the business model? So all of us in here follow a pretty similar model. How is AI going to change that moving forward? Not just next year, two years from now, five years from now, ten years.

From now, it's totally impossible to say. And anybody that says it isn't is selling you something. I think the bigger question is what does it do to all knowledge work? I mean if, if I like to tease accountants that if we have our way, AI will replace like all white collar work and it'll just be the accountants left working 11 hour days and we'll be like, we told you, we told you that you can replace us. There's like a, there's like a bigger question of like how does that impact society as a whole? I have a, there's something about the human condition that always finds more to do. I mean industrial revolution, electricity, computers. And there have been all these times in the past where it's like, oh, we're only going to have to work 20 hours a week now. But like the wheels of capitalism just keep on, keep on spinning. Even if you take this as far as, and it is good to think about what's like the extremely powerful AI scenario, what does my firm do then? It's worth thinking that through and what, what you come back to still is, let's say OpenAI ships a model that's tomorrow that's good enough to create the XML file that goes to the IRS to e file a tax return. Right. Like what if, what if that just became possible overnight? Or a language model that could be your whole accounting system. We're actually very close to that. Just being able to have a conversation with an LLM and, and track your books. Like that's probably very close. Well, what's still missing? Like we think about all the other things that we do for people and how we're kind of the arbiter that stands by is this correct or is this not correct? Never mind the fact that like after the fact reporting has really very little impact on the profit levers of the business and what a company should do next. It for sure puts more pressure on being human. And so if there's anything that shrinks really quickly, I think it's the folks that want to sit in the back of the office and just hustle stuff. There's probably still going to be room for those people just less than there was before. But I have a hard time imagining us not being able to find something to do, especially as today. There's just so much more need than we could possibly fulfill. What kind of resources or programs maybe are out there to teach, to train staff on prompts, how to properly use prompts and how to create those. So I'm biased where like I learned from YouTube. I mean there's just so many good resources you can find out on, on, on YouTube of folks kind of sharing frameworks and stuff like that. There's a real risk of overthinking at all and that kind of being procrastination versus just spending more time with it. I think the most important things are spending time with it and sharing your experiences with other folks and having that feedback loop. Otherwise, for me, there's not a ton of thought leadership going on in the accounting space that isn't through the lens of a vendor that's like, here's how to use our tool. Obviously I do thought leadership on it on YouTube and social media, but the way I learn, it's from folks outside the accounting space that are putting out.

Content so when you say, you know, accounting AI is underdeveloped right now or early on, I mean, there would be an automation layer to automate many of our tasks, but there's also a higher level to empower staff to, you know, be more capable in their client relationships because many of our managers aren't as capable to have that high level interaction with, with a client, to provide that kind of advisory level that we've, we've tried a number of different ways. You know, advisory in a can, advisory, you know, how to, how to give people ways to, you know, punch higher than their weight. I mean, how would you, I mean, is that outside of this or is it still too early on?

So this is a great, you know.

Top down or bottom up?

Yeah, all the above. And in fact, it's a really big misnomer that AI is coming for the menial work. It's all work. It's, it really is clear up the skill spectrum that it can be helpful. When you come to a specific situation like this where it's like, how can I use AI to help me do this? My mind first goes to what's the source for the correct way to do it? Do we have a, like a model, an explainer and an sop, Any sort of documentation in the firm describing what is the right way to do it? And then do we have examples? The examples most commonly that are valuable. Here are your meeting transcripts. You can attach up to 1 million documents to a custom GPT. So like, how many of those calls have you had in the last decade with your clients? Well, I wish we had all those. So like the starting point is meeting. Meeting transcripts are like the sawdust of a professional services firm. The first step is capturing those. And maybe it's like you, you do this month end recap call with a. Let's, let's just nail that down to a specific client. If we had recordings of every like month end sort of advisory call with that specific client for the last 12 months, when you hand that off to the new person, my gosh, do they have an advantage to be able to like create that GPT and query it about things they've discussed in the past and be like, they asked me this question, I'm not sure what to do. How do I navigate this? And maybe you go to the boss, you know, the person who did it before that's now elevated or doing something else. Now you can see exactly kind of what the boss would have said.

And that's what I mean. The PASBA model, historically was kind of a box based system where the idea was everything you need to process that client and interact with that client you could just hand to someone else. So if you needed, you had turnover, you needed to move it over to it, reassign that box system was valuable. Anyone could pick this up and begin to process. But a lot of the kind of that intelligence would drop off the conversations is we'll start with an initial strategy session, try to record that, you know, setting kind of the tone and then all of our different, you know, management meetings, our tax planning meetings, all that and you know, to bring that in and then, you know, sift it and you know, kind of get a summary and a core that someone could pick up and just go, okay, now I see where we're at with this client. I mean we would like that type of, you know, we'd like to move that to the next.

Yep.

To this new world.

This, maybe you're already working on this, but this is a great use case for pasba who you already have sort of the infrastructure and the explainers of how to do xyz. How do we then augment that with, you know, what's a custom GPT that could support this role or this function?

Well, this would be interesting because this had come up a little bit today. So rather than having each of us create like a custom GPT and pour like all the PASBA manuals in all the PASBA ways and if maybe we did that at the high level, the association level, and then created a way to, so we didn't all have to redo it or we put our firm, you know, have a global PASBA method and then customize it to the firm. I mean, is that a more viable than we all just, you know, scrape that all up individually and you know, piecemeal.

Perhaps anything that can be shared lightens the load. So you can have public GPTs. Downside of public GPTs is anybody with the link can get to them and more or less get to the underlying stuff. So anything that you're building with your own client data would have to be within a, you know, obviously it could be a, a public GPT, but inside of your team plan you can have shared GPTs that only exist inside of your team. So anything with identifying information you can do that with. But everything that rises above that, as long as you're okay with it, living in a public link, you can do that. Another workaround here is just, I mean you've just got a one page explainer for like, here's how to create this GPT, attach these four files, copy this prompt, put it in. You're good.

Yeah, it's really not that massive amount at the, you know, at the PASBA level. So that, that could be very doable.

This episode is sponsored in part by Verito. New sponsor alert. Have you ever gotten that gut drop feeling when a. When a server crashes right before an important deadline, it doesn't make your blood boil. If you've ever gone through tax season, let's zoom out. Any season of life with unreliable technology, you know, just how much your tool crapping the bed at the wrong time. We just can't have it. Okay? That's why we're talking Verdot here today. Okay? You get in your own dedicated servers. No shared resources, no noisy neighbors. VERDO will even migrate you with zero downtime. Back it up with a 99.999% uptime guarantee. No outages in nine years. That's impressive. I mean, most of us can't say that. Think of the low lights of the last nine years. More than 1400 US firms already trust VERDO to keep their deadlines on track. And Verdo, they work with accounting firms, so they get it. What are you rolling? Ultra Tax Drake Pro Series. They're gonna get it done for you. It's gonna be secure and it's not gonna crap crap in March. Okay? Pardon my French. Get to a better hosting platform with Verito. Check out the link down in the show notes. This episode is sponsored in part by Mercury, which I am also happy about because I am a actually a big Mercury fan. A user myself. So Mercury is a FinTech that over 200,000 companies trust for banking and running their finances. Because the traditional way of getting this done, those platforms, big hot pile of new sponsor. There they go, walking right back out the door. Like I said, I'm a happy user myself. It is a modern platform. We liked it so much, we actually moved all of my businesses to it. Now Mercury for accountants, the referral partner program that allows firms to give their clients VIP access to the Mercury product and unlock special perks. Tell me more. Dedicated onboarding and support for firms and clients. Thousand dollars referral bonus for every new. What? Okay. Per eligible client. An advisor portal for easy team and client management. Mercury Pro for the firm. Nice. They got a link down in the show notes to sign up and learn more. Disclaimer Disclaimer Disclaimer voice Mercury is a fintech company, not a bank. Terms apply for bonus eligibility. See Show Notes for details. Can we go back a Minute ago you talked about hiring an AI agency. We recently brought someone in the firm for training and I realized quickly I did not do my due diligence because we may have known a little more. So, so how do we hire that agency? What are we looking for? What can we do in due diligence to know that they're going to bring real, really good, valuable information to us? Has anyone here hired one that they had good success with? Matt had a little bit of success.

We got someone over here. Would you care to expand or does this surprise you a little bit?

Thank you.

No one is engaged.

It wasn't an agency. We decided to go into the tech world and took a 27 year old who understood all this and didn't understand accounting and tax or anything we do at all and ask them to join us and teach every one of us how to use ChatGPT with team version. He's way over my head, which is great because I used to do everything myself and he is having one on ones with all of our team members. He's adding in custom GPTs and he's learning the accounting side and bringing it across. But he brought the tech side in which I think is really, really hard to do to find somebody in the accounting industry who does that. So we kind of went outside and brought somebody in.

He's about to get a lot more busy, I think.

Well, I'm a little embarrassed to say this but it's my son who had a startup business.

Come on, it wasn't a commercial. How much for this?

He, it's really hard to find somebody and he just had the right skill set and all. He had a startup business in the in tech industry and was looking for part time work. We're like, let's try it. I was amazed at how great it went.

So the, these types of referrals are, if you can find them, are gold. I've got like a little Rolodex of sorts, but it has a very small number of agencies on it. And the good, the good thing is they are the agencies that are out there talking about what they do for accounting firms specifically because I don't want to pay them to learn. Oh my gosh, what is this awful tool you have to use for tax or how to connect this or that. So the more you can lean on what they've done for other firms, the better. There's a few, I can't remember their names off the top there. There's a few that are pretty visible on LinkedIn that I know folks have had good success with Chad Davis, runs his automation community. Community Automation Town. A number of those agencies are like presenting in there and kind of co branding with that community. I could give you a specific list after this if it's helpful. I can't remember what they're called, but.

I'd be curious, you know, with the connector layer. So we, you know, we'd love to have a suite of products that were all interconnected. Right now it's a little bit difficult without getting too technical, you know, using Zapier and Nadan and make and all these to connect, you know, things that we really, the applications we really like. How much effort should we put into building that out now versus, you know, the next generation, you know, like an MCP type of connector where it's a little more easy to put together or, you know, to your point about, you know, everyone wants to sell us something in this space around AI, do we wait for people to develop these things within, you know, some of the core applications that we rely on? Or do you think that we, that we're moving to a model, we'll just pick our best in class in these different areas and the connectivity will come. Is that something to be optimistic about? Or, or are we, is it still going to be a little bit clunky for the near future?

It's going to be fragmented. I mean, hopefully it's only the near future, but we're not close to it not being fragmented. I feel pretty strongly most accounting firms have no business building software. There's a lot that have gone down this path and burned tens of thousands of dollars. What we're talking about here, you could pull somebody in, they can whip it up in a week for a few hundred bucks. That sort of thing I get excited about. If it solves a problem I have today, don't wait because maybe there's a solution for it down the road. You have the problem today, like, let's, let's make the solution. But then, especially when you're in a community like this, to be able to see what other people are doing and be like, hey, we just, we hired this person, we learned about this guy's son and he came and did this for us. And then it's like, well, does everybody need the same thing? It's already built. Like, that's, that's what gets me excited.

Got a question right here.

I have two questions. First question is, what area have you seen the most success with the AI use? Is it accounting, the tax prep, tax advisory, or client communication or something else? And Second question is, how do you communicate the use of AI with your clients to maintain their trust and transparency? On the first question, client communication is the Easy1 Because AI is good, so good at narrative things. And it's, it's something a lot of us aren't good at, is we're actually better at the number stuff, which AI isn't as good at. I know you talked yesterday about, you know, financial statement delivery and stuff like that and, and the, the, the juiciest opportunities right now it's, it's the more narrative oriented things, but it's increasingly becoming, asking it to do a, a specific job. So right now it's very good at reviewing a balance sheet and a P and L and, and just the same way that you do where you do that initial skim and you're like, I gotta look into that, I gotta look into that. There's a new loan like that. It can do all that stuff in a way that I want my prep staff to know how to use to do kind of an initial review. So we're starting to get into those very technical things where it is pretty good. But historically it's been the narrative stuff. What was the second question? How do you communicate use of AI to clients? So, yeah, it's hard because AI is just proliferating so many things that oftentimes we're not even aware of. My take, there's a, there's a degree of like, what do you feel good about that I actually think is important here? For example, let's say I work with creative artists right now who are like, you can, you can get out of here with that AI buddy. You work with clients like that. Like, yeah, there's like, that's, that's a different conversation to, to have that be part of your service delivery. For me, the, the obvious. No, no. Is never put a synthetic voice or visual out there where it's not clearly identified as not being. You know, I think that's that any, any like, anything that's trying to act like the real version of you, like, that stuff can be very useful. You just got to be clear that it's not the real deal. For me, personally, I don't think I would feel the need to disclose. I think it's just, it's part of work and becoming more so every day. But I could also, like, that's also something that probably folks have to decide for themselves.

A bit that had come up in one of the practice discussion groups was to create a video avatar of a manager to deliver kind of an analysis of a financial statement or to you know, say, watch this, this is what we, this is what I'm seeing, you know, in the form of meaningful monthly client contact. But when they were doing it, the, you know, the avatar or the synthetic individual didn't look realistic. It's like, well, how realistic should it be? We should not expect it to, you know, replace that. I mean, how meaningful would that client contact be if it's just a, you know, synthesized version of a relationship?

I think it's something we need to be aware of as, as it kind of interacts with many of our natures being accountants. I was talking with a guy that runs a software company that does sort of like automated AI advisory stuff and they shipped a built in version of the synthetic, like delivery of you. Like you give it a few images and it creates this thing of you. And this was maybe nine months ago and the users were so excited about it. My reaction was like, oh, that's gross. But like I have a very different lens because I do video stuff and like I don't have the same perspective as, as regular folks, but folks were so excited about it and I think there's something very attractive there for many of us for whom getting on video and being vulnerable, all that stuff is really, really hard. But we're not far away from this work being increasingly automated and if we're not willing to, to be human and, and try to make connection and all that, which by the way is what every single one of you are doing here, like that's what makes this powerful as uncomfortable as it is. If we're unwilling to do that, we're like barreling into an age where that's kind of all there is. Like a lot of folks will tell you like 10 years time, that's, that's what's left. And so then what is my firm? Well, my firm's actually a hub for a specific type of entrepreneur and like I can give them expertise still and all that, but that's not annotating freaking PDF files and getting the naming convention right and putting it into my file storage system, which is a huge percent of the percentage of the time of what we do today. So I get the appeal of it. There's a little bit of razzle dazzle where like some clients may even think it's cool, think that's going to be short lived because everybody's going to get sick of AI if they aren't already. And if you're on LinkedIn, you already are. But like there's, there's going to be a time very soon where folks are like, that just kind of feels like, kind of yucky. Like you couldn't have actually just taken a smartphone video, just gotten on a loom for a minute and done the same thing.

We got to raise your hand if you got a microphone. Go.

So, fun fact, the classes I took this morning, I transcribed them on my laptop and then pasted it into chatgpt and asked it to summarize. And then we emailed it to our friends. So it had a little summary from each class. And it was really helpful because, like, if you're taking notes and then maybe you don't hear the next thing the presenter says, it kind of captured some of that. So that was cool. And so I was going to mention that, like, I tend to after something like this, I have all these notes, I go back home, and then you get in the grind, and then you're back to working. And before you know it, two months, three months have gone by and you haven't even looked at your notes. So I'm hoping that my friends and I from this group can have like a monthly, you know, session and just kind of challenge each other to try something new with AI and then share with each other maybe just like one thing a month.

Yeah, yeah. We talked about, like, what's the feedback loop inside of your firm to keep your peers learning from each other? The whole value of this is the feedback loop of each other. So that doesn't even need to just be inside your firm. Like, how can you sort of learn from other firms as well?

So if we were to time warp, if we were to time warp in the future, three years from now, you know, how big of a direct competitor, with all these advancements in AI and all these technologies, how big of a direct competitor do you see into it being to the small business, to the accountant, to focus on small businesses like a lot of us do. And where. What should we be doing now to position ourselves to kind of be ahead of that and prepare for that, that intuit isn't being our direct, competent, taking away our clients?

Yeah, got to get up market. AI is the best thing that's ever happened to productized solutions. They're going to get better, like they just will, and they're going to nail it. Right now, I would argue there's a huge gap between productized solutions and good professional help. And that sucks. Like, there's a lot of business owners where they can't afford 800 bucks a month, like, they genuinely can't and we'll be like, we can't afford not to. But it's like, no, that's actually a huge spend for them. But what's the alternative? Like you go down to 300 bucks for a bench or you know, into a live bookkeeping or something like that. That sucks too. So I'm not too concerned with that. The only folks that becomes problematic for are the people who are selling work right now at the same price as those productized solutions which we have no business doing. So yeah, another way to look at this is, you know, since TurboTax has gone into the business of basically being an accounting firm and having a full service option for, for small business that's not, that's not going through and putting stuff in like, that's like the usual. You work with a pro from beginning to end. Just like working with a firm federal only starts at 1800 bucks. I don't know if you see those like fee surveys that, that, that people put out.

We do them, we do them as an association.

You look at like NAEA fee survey and I think a lot of the folks in this survey are more like retail tax. I mean a huge percentage of these people are charging 350 bucks for an S Corp return. Where you're like, how is that? That's surely that's not the average. So the folks that are setting up camp in the same place where, where they're working, that is a problem. But most of y' all aren't there and like we just need to make sure we avoid it. It's also another argument to not do like the synthetic thing, like don't be the productized solution. Because that's kind of the only advantage.

We have is that going to bifurcate the market between the value add and then the commodity layer. I mean it's always existed the commodity layer of our services, what you just described and we elevate out of that. I mean how could we use AI to enhance what we do rather than drive it down to that commodity layer?

I think it already has. I think right now we've got that really big divide and it's an argument against getting to turnkey and how we price projects and deliver projects and all that. It's also a really big argument for specialized expertise. If you work with a hundred folks that are kind of of a similar ilk or run similar businesses, like you become an absolute wizard and the right and the wrong ways to run that sort of business in a way that TurboTax will never be, even if you can Hop on a call with an EA and ask them questions. Or, you know, same thing for live bookkeeping. If you can hop on with a, you know, Mike, 27 year bookkeeper, like, you know, the guy that you see in the, in the marketing stuff, like, even if you hop on a call with Mike, they don't have the domain expertise you have. So I would say it is an argument in favor of trying to find some specificity in the folks you serve.

Yeah, I think we've got some microphones out. There you go.

Jason, I'm such a fan.

Thanks for being here, man.

We all are.

Yeah. You know, so I listened to your podcast, went all the way back from last time. It's. I just.

Have you started the second lap yet, though?

No. Well, I, I go a mile and then I have a, like, you know, my brain just explodes. And then, you know, I got to turn you off. It's too much water hose at me. So here's. I'm going to pose something to you. Why do the processing times take so long on like, let's say I want to process a disabled Access credit and I want to feed a vendor list to ChatGPT and I wanted to review the disabled Access credit web accessibility site to see which one is like 85% or above that might qualify for that. And then create a spreadsheet and then create the disabled access credit form so that I could put in the tax return. Why does it take so long to do that? And will that get better?

Would you give AI a break? Man, you're asking a lot. Imagine working. Imagine working with this guy. What's taking you so long? I kid. One thing that's really good to know about AI is, is the like cognitive load of what you can throw at it at a. In like a single prompt at a time is still limited. Like, you can, you can give a human this big long directive and it can do a bunch of steps. AI is not that way. So the best practice for something like that is chunking it down into a, like a sequence of steps that it can do using reasoning models, which do take time, but then you're, you're totally on the right track with like, how do I, how do I templatize this with maybe a spreadsheet or something like that so that the tool actually builds the tool for you. And that's a really cool way to use AI, is it making the thing that you can then go run a thousand things through if you have something you do over and over again that is a long sequential process. And it's an argument for going more down the agency route where they can build an automated workflow that handles each of those things individually rather than like having like say five prompts that you always give ChatGPT in a row over and over again. That's a good use case for like somebody building that for you so that you give it the inputs and it runs it through that same sequence. Especially if you're going to do it a large volume of times. Because yeah, if you give it a whole big long like thing to do all at once, it can sometimes get it. Oftentimes it doesn't. But a checklist is a great example. If I got a hundred item checklist and people will say this in like YouTube comments, they're like, I gave it this thing and it's like this big long thing and it just, it didn't do any of it. It said it didn't, didn't do any of it. And it just can't do that like 100 item checklist right now. Maybe it could do five of those at a time, you know, and someone could build for you a sequential automation that would do 1 through 5 and then 6 through 10, and then 11 through 15 and all that. But it can't do it all at once. How soon do you think I will be reliable enough to handle discovery calls? I've seen Chris Hoffman has a big H vac company out of St. Louis, has had good success with it. How soon do you think that'll be able to be rolled out for like accounting firms? It's already good enough, I'd say, but why don't you use it? And what would your clients say I think is the bigger thing. So the more interesting version of that to me is the assistant for my person who does discovery calls that is getting a live transcript and saying, you still need to ask this, you still need to ask this, you need to steer them this way. That's what gets me really excited. Kind of coming back to the point of not losing humanity in those connections. I think the, the AI assisted human is the solution there. There'll be productized solutions where, yeah, you sit there and you talk with an AI agent and all that. I just don't know that those will be the people we're serving.

So going back to, I guess Craig's comment, when you give it these larger projects, number one, the accuracy tends to fall off and it's not going to follow everything to a T. Second thing is it starts taking a lot longer to actually synthesize the information. So do you think that when you look at AI in the big picture and we all have this fear that next year AI is going to be great and we're all going to be out of jobs and we're going to be figuring out what we're going to be doing in the fields or on the farms or whatever else we're doing. When you actually. Right, there you go. So when, when we start looking at that, the, the, the question from this is, doesn't this really only go towards really small businesses that they only have a handful of transactions per year and it's really not going to be those clients that were actually, you know, identifying as those ideal clients where they are bigger, there's more, there's more complexity, there's more planning, there's more actual human element that can never be replaced.

So like you're thinking the smaller ones probably just all get commoditized and go to product solutions. Yeah. So I think that's, that's a goes hand in hand with product solutions coming up market, they'll get better. It'll be easier for the small companies to just do that in a really turnkey way, which frankly I think is good. Like we don't need more blockers to entrepreneurship. Not like it is today. Your expertise, the, the more money that company is doing, your expertise has a bigger impact just because that profit lever is bigger. So for me to be able to command better fees and the client to see more roi, the bigger that company is, obviously that that gets a little juicier as opposed to a company that's doing a quarter million a year top line. It's like they've got very limited budget constraints. You could give them the most amazing ideas that boost their bottom line 10%. But there's only so much in that for a firm like yours that you can actually command. So honestly, I see that as a good thing for us. I mean, how many freaking little Schedule Cs LLC side company things do we do where they're like can you do the books for this thing? And I'm like, there's no way I can do this profitably. These itty bitty projects that we don't want to do. I'm perfectly happy to let the little stuff walk so that we can do the better stuff. Today's episode is sponsored in part by Client Hub, the practice management system that delivers WOW seasons year round, not busy seasons. Seasons that will dazzle your clients for two reasons. One, ClientHub is drop dead simple, man. If you're too Tired of a practice management system that just feels like you gotta have a PhD in learning something new. Client Hub. You hop in. The vibes are great. It just all makes sense. But two remember, there's a second reason ClientHub offers a game changing client experience. That is you can actually get clients to like collaborate with you and be helpful on the platform. Because the portal wasn't this like afterthought. Bolt on now. Very cool. Recent development in Client Hub Magic Books review Twinkle sounds so Client Hub syncs in the transactions from your QuickBooks file. You can run an AI analysis on it like from inside of Client Hub. It'll highlight any potential issues you can mark. Like on the balance sheet in the P and L, you mark each account one by one, whether or not it's been reviewed. But if you want to troubleshoot, it'll pull you straight back into QuickBooks. Like you don't have to learn like a whole new troubleshooting tool if you'd rather sort that stuff out in QuickBooks. Other cool stuff it'll identify automatically. Expense inconsistencies, purchases without a payee, purchases in parent accounts. Why is, why can't we not disable that in QuickBooks? Large transactions, new accounts, new vendors, new customers. It's Magic Books Review inside of ClientHub. I think you might like it. Learn more about ClientHub. Check out the link down in the show notes. Teamup is a recruiting firm that helps you hire direct in the Philippines. Talented accountants. That's different from the outsourcing or offshoring companies you most commonly hear about where you're actually hiring the company. With Teamup you're just hiring a person because you can just hire people in the Philippines same way that I may hire somebody in Reno, Nevada actually. Great example. That's a pretty, pretty suspicious place. I'd say the Florida of the west side of the United States. How do I trust them? How do I train them? How do I do X, Y, Z? Well, how do you do it for like your onshore people? But you really are just working with them. They're an extension of your team. That's what Team up specializes in. So team up. You do a flat one time price to recruit that person. Most hires are about four grand. This, this one time fee. Team up. No ongoing cost. Most offshoring groups, about half of what you're paying is actually going to the company, not to the person. Team up even has a guarantee if that hire wasn't a fit for the role that you put them into in the first six months Team up will find you a replacement at no cost. Does indeed do that. They don't do that. Hiring in the Philippines is simpler and easier than you think. My last four hires in my firm, this was like three, three and a half years ago. They were all in the Philippines. If you want to learn more, check out the link down in the show notes to schedule a call with Timo.

Do we have any microphones out there? Raise hand. There we go.

Hey Jason. Me again. So I was really surprised at the paid team version versus the basic version. And I started using it about a year ago on the basic version, non paid. And I was really surprised at what happened once you started to personalize your ChatGPT by having the paid version. I'm not sure everybody who's not yet gotten into AI understands the difference between them. Could you just share a few tips as to why you think it's so important to have the paid version and personalize the ChatGPT as opposed to using the non paid version?

The biggest thing with the paid version is security. So the team plan, the prompts are never trained under the model, so you can use it with client information, identifying information, that's the big thing. But the team plan is also the only one. The team plan is the cheapest plan that comes with SOC 2 type 2 security. So like that's, that's the minimum. That's the first plan that's good enough for me. From a security standpoint, that's as good or better than security than most of the tools in our space. You do get access to some better models. You get some cool features like being able to share custom GPT between team members. But then there's two features that kind of roll up to be kind of personalization. There's memory that is it remembering things from past conversations and sort of injecting that into its responses. And then there's custom instructions where you can explicitly tell it, always talk to me this way or do this or that. It's not unlike working with other humans. Like there's so many analogies to how AI lets us down because we don't give it clear expectations and directions in the same way that people will when you kind of cut them loose. If you have very specific ways that you want it to work with you for sure, that's really helpful to have kind of built in to ChatGPT. An interesting thing that's happening right now is context lengths or how much it can include in one prompt that's very quickly going to like infinity. The notion that it can actually pull any amount of information into a single prompt. And we talked about this a bit last year. We're headed to a world where ChatGPT, as of four weeks ago now says it can. It can remember anything. It can actually go back in any past conversation and retrieve things and pull them up in this conversation. Sometimes that's cool and useful, other times it pulls something you don't want. So there's like, control is helpful there too, and you can turn on and off memories. But in a world where the AI remembers every single thing that you ever tell it and incorporates that and that informs every single response, that AI starts looking like a long term asset. Like, that's actually really, really powerful to where the first week you're working with it, you're like, it's like a new employee. But two years into it, when you've talked with that thing about everything from navigating sticky client stuff, team stuff, fixing your dishwasher, like it knows so much that becomes almost like exponentially more useful over time in an exciting way.

We got any microphones out there on the way? I'd be curious. A question would be at the firm leader level. How could. I mean, what would your opinion be on enhancing a firm leader's ability to utilize AI to get, get us more empowered, to be better.

Yeah.

And to handle all. You know, we're very busy. We're just very busy.

Right.

How would we lever that? You've got anything, you know?

Yeah. Actually to Brenda's point earlier, Brendan, when we stop talking to Brenda's point earlier, oftentimes we'll see a version of something that we like. So maybe it's a talk at a conference or a podcast or something like that. And we're like, that's a great idea. I should go do that. And then you never do. Let's just take a podcast. So I run through some sort of workflow change in a podcast and you're on an exercise bike when you're listening to it and you're like, that's a good idea. But by the time you get to the office, it never returns. Take the transcript from that, chuck it in ChatGPT. Create an initial sort of like SOP, a version one of what this could look like using that context. And that context could be a book, could be a podcast, could be a blog post. Think like, what are the, what's the good version of that thing you're trying to do? Explain to AI. Here's the good version. Because if you don't, the default is average. It will give you something very generic. So give it that context of. Here's what I think a good version of it is and some context about my team and how to implement it. I would say the bigger thing that we get wrong is we think that we're gonna do it all ourselves and you just won't like. The better solution to that is come back with a version that you can cut your team loose with. Because if you're the one that shipped it entirely yourself, well, you're also now on maintenance duty and every time it has to change, it has to be you. But this is where we can use a transcript AI to make that initial SOP to make the handoff to the team easier and get them started.

I think we've got Noah.

My Chat history on ChatGPT is an absolute mess like that. Money is an understatement because you use.

It for weird stuff.

I use it for everything. Yeah. Use it to cook dinner and buy my wife's Mother's day present and some work stuff every once in a while too. Right. So with what you just said about memory in mind here, does it make sense to try and keep that more organized and use a bunch of temporary chats or so I don't have just all this chat history junk? Or does it make sense to leave everything in there so it creates that asset and is remembering everything that I've ever talked with it about?

I think that's something that's solved long term. I think it gets better at only retrieving the stuff that it actually needs. So I don't know that I would change my behavior as much as like, as much as it seems pointless for it to know that part number that I replaced for my dishwasher. I had another problem with my dishwasher like a month later and I can be like, hey, this old stinker is having another problem and it may actually be useful that it knows that context. So right now I don't know that I would try to manage it to that degree. You can go in and manually delete memories and obviously archive conversations. I would hope that's something that it just handles for me better over time.

So in that vein, if we were to set up a firm level, custom GPT manager level, a bookkeeper level, do those cross pollinate one another? Would it be able to Remember across custom GPTs right now, or is that going to segment the knowledge?

Custom GPTs I believe are not included in memories might be wrong, but custom GPTs I think are more self contained. But that's the Risk is you don't want them cross pollinating each other.

Well, what if you did say for instance, a client, you know, at the client level, we create a client record or a client, you know, this is that client, remember everything about it and then it, but it's segmented, you know, the bookkeeper works on it, the manager works on it. Can it be asked to, you know, to cross pollinate itself under that client record versus under that role?

Right, yeah. There's not a way to do like two dimensional GPTs. So I would start with what are the things that all the people would need? So that's probably the client specific information. Or it could go the other way too, and then share that GPT with my team members. Then they would still need to plug in like the, whatever context is specific to their situation. There isn't a way to like kind of split it on two axes.

Yeah, because I was also thinking that, you know, we will load in the disk and the Colby information and that sort of thing for our roles and then we'd like to maybe apply them to client activity to help refine, you know, where we could coach or train somebody up, you know, based on, on how, how they interact, you know, how they see the world or how they are, their personality. That gets a little dicey to, you know, to, to create it and enable it across. I mean, I guess I'm trying to think in terms of how we would organize our firm around AI so that we don't segment things inadvertently.

If it's the sort of stuff that's done infrequently, it's not that big a deal to just pull it all together and do that work. If it's something that's done as part of a month end close every single month, or maybe a monthly one on one with that team member. That's the sort of thing that to me rises to the level of let's have somebody actually build this for us. And it goes out and it fetches the variables to that workflow from a spreadsheet or something like that. But it can run then automatically and it's just injecting, you know, here's the disk for this person versus that person. Here's the destination of where we want that. So for infrequent stuff you just pull it together for the recurring stuff. That's a situation where maybe you think about somebody building something.

So if you look over the horizon, is that going to change or do you think it will remain segmented in this fashion?

It'll for sure get Better, like how to. How do you manage information organizationally? And right now we're seeing better ChatGPT integrations into your Google workspace and your SharePoint and stuff like that. And so another spin on that is right now instead of attaching a file, you can point it to say, specific folders in a, in a company folder structure. And you know, once I can do run a chatgpt, that's just whatever's in this folder. Like, there's almost certainly going to be different ways that you could do this.

I'm sorry, in the order received.

So I asked this question the other day, but I want to ask you. So we have the teams chatgpt, but we use Microsoft Teams. Our team uses Microsoft Teams. I would love to have an AI team member because you were talking about how they remember things over the years. And so we would be like, if we wanted to sell our business, like, let's say we named this team member, we give it a name like Jim. And so we would sell like Canon and Associates and Jim. And so whoever bought the firm, they could ask Jim, like, why did they make those weird journal entries for that. And like, Jim would look back at our conversations and he just know that more the history of our clients. But I don't think there's a way to have an AI team member in teams.

You can, somebody could build that for you. The most important thing is that you're capturing the context for whatever would inform Jim right now. And most of us still aren't. We aren't capturing all our meetings, so make sure we're doing that. But custom assistants that live in teams and stuff, there are various ways to do that that will point to like a folder structure or something like that. It's totally the sort of thing that you ought to prototype with a custom GPT though, to be like, here's kind of how we think we want this to work. Is it actually possible yet? And then if you love it and the team's getting a bunch of use out of it, for sure, find somebody who will build that same version in teams or WhatsApp or whatever makes sense.

Okay, well, we've got John and then maybe one more, depending on what he asks.

I can make it quick.

I can drag this on now. So right now we're talking about ChatGPT and we're talking about the difficulty of saying, okay, well I've got a personality style here. Now I got these custom chats that can't access. So what if all of a sudden like we're on chat GPT and we make Matt has 45 customs. And then all of a sudden, like, this Gronk or whoever this Elon Musk's dude is, right? They're like, holy hell, that thing is 10 times better than ChatGPT. Now. It's like file cabinet CS, right? You can't transfer your stuff. Once you got it, you're kind of stuck. How do you transfer all that knowledge to the new, better AI or, you know, chat GPTC got you sticky. And now it's like, we're going to charge you a million dollars a month.

We have that problem with all of our tech, right? The thing to focus on is the, the skills you're building in humans, not the infrastructure. I think the infrastructure isn't too bad to rebuild. What we can do is we can overthink all this and have analysis paralysis, but then the humans don't build the skill. I think that's the bigger thing.

One more. Where are we at?

I have a question. So I've heard different kind of scenarios. I'm very new to, to this. What's the correct layout? Because I hear sometimes here in the group, some people are like, oh, I need, I need a. A custom chat GPT per client or by employee or by department or by one overall one for, like, what's kind.

Of a mess right now.

So what is the correct way to do it? Because then you do it and then you're like, oh, I have to redo it if it's not right.

I would hazard a guess that there's very few people in the room that have this problem today. That would be a great problem to have if we'd actually taken it that far. I think there's a lot more people who will ask the question you're asking and then like, not do it because they're going to be afraid of doing it wrong. So just start. Try to be like systems and process first rather than AI first. Like, what's our process and how can we plug AI into it? And it may look a little different firm to firm. You're doing great. If you have that problem. Yes. Think about it, but don't let it block you. Like right now, organizationally, there isn't a perfect way to do that.

We got. I see Natalie back there. Anyone want to charge back there and give her one more, and then that will be the absolute last. Natalie's been doing quite a bit. She might be one of the few in the room that has gotten a fair way down that path.

Jason, also a big fan. I got started because of you. But I wanted to ask. You said not to reinvent the wheel. One of the things that came up here was having a client GPT that would combine all the data together. So has anybody ever already done that? Is my first question. And I put that on ChatGPT and it gave me the steps with the APIs, putting Zap together the spreadsheets, connecting them all together. My question is, has anybody done that so I don't have to put the time in another three days to get that done.

Nobody's taking it as far as, like, pulling in ledger data. It's something that I talk with a lot of the practice management systems about. There's people working on it for sure. It's not, like, imminent by any means. I still. And I know that just a really crappy answer I gave you here. I'm just. I think about this in such a more scrappy way where it's like, that sounds like a cool idea, but would I actually come back to it every day and would it be super useful? I try to push people to, like, make this super scrappy version of it just to, like, validate. Is. Does this even work? Like, is the AI good enough? Is just the meeting transcripts enough? If meeting transcripts and copies of financial statements, is that good enough? Because then it's like, well, that's just a set of PDFs, maybe that's enough. I don't know. That's a little bit of a wet blanket answer too. But, like, I try to just start more simply with that. And to answer the question, there isn't an amazing version of that on the shelf right now.

Okay.

Because we do something called M2C2. So it would be more for that when you're making a phone call to the client to build that rapport, to build that relationship, to be able to bring up all of their data. They got married, they have kids, it's their anniversary.

Yeah.

All that comes up when you pull up their name.

Oh. I mean, think about our CRMs and how we want to have birthdays and kids and all these things. If you had all of your transcripts and you could run a couple of prompts on six years of transcripts, and somebody tagged all of your clients that way in the CRM. How amazing would that be? Right?

Time is up.

One more? No.

Okay. No.

No.

Carl, any. Any final thoughts or are we good?

Thank you very much for having me. I literally have to run straight to the airport, which sucks, because I want to hang at the party tonight, but thank you very much for having me. It's so amazing for me to meet you all. And even if you haven't done the second lap yet, know that like my reality is sitting in a room by myself talking at a camera. And I know it's weird to come up and talk to me and say something that means the world to me. Please do. I super appreciate it. Maybe not right now cause I gotta go to the airport, but thank you all. Like I love hanging with you. Pretty cool, right? That was a fun one. In Tampa. PASBA does a couple conferences a year. Again, I'll link PASBA down in the video description if you want to learn more about what they do. If you got questions about how to use AI in your accounting firm, drop this stuff down in the comments. Every couple of weeks we do like a Q and A sort of roundup episode. I'm increasingly doing these talks around the country. In fact, if you go to jasononfirms.com/events, you'll see all the upcoming events that I've got. This week I was in Orlando, but I've got events coming up in Los Angeles, in Denver, in Chicago and Atlanta. We'd love to see you there again. Jasononfirms.com events we'd love to hang in person, but either way, I'll see you in the next one.

It.