In this video and podcast, Randy Johnston and Brian Tankersley, CPA, discuss the recent AICPA AI Symposium. Watch the video, listen to the audio podcast, or read the transcript.
Or use the below podcast player to listen:
Transcript (Note: There may be typos due to automated transcription errors.)
SPEAKERS: Brian F. Tankersley, CPA.CITP, CGMA, Randy Johnston.
Brian F. Tankersley, CPA.CITP, CGMA 00:00
Welcome to the accounting Technology Lab sponsored by CPA practice advisor. With your hosts, Randy Johnston and Brian Tankersley,
Randy Johnston 00:11
good day, and welcome to the accounting Technology Lab. I’m Randy Johnston with my co host, Brian Tankersley, and we’re very pleased to speak to you today about the recent AI summit, the symposium that was held by AICPA and cpa.com now this was not the first of these type of events, and I was very excited to be invited. And there were loosely, 17 presenters and 57 attendees, and there was so much doggone learning that occurred in the room. So Brian, I know that you know Eric gaskerson and the new president of the AICPA, Mark Koziol and as well as Mike sarami. You’ve known all those guys just about as long as I have, and they opened up the day with a bit of a welcome and a perspective on AI accounting and finance, and then we got a bit of an overview from Mark Peterson and Curtis Philp, who’s Senior VP of alpine group on the political climate. And if you’ve been attending the AICPA town halls, you’ve gotten quite a bit of briefing on that, but you know, when you think about the overview, we did discuss a number of very important things that are going on and more that we’ll talk about in our podcast with you today. Of course, the focus is, how does AI affect the profession of accounting, and how does affect both public practice and industry accountants? So it’s all about AI in accounting and finance. And there’s, you know, a lot of discussion around the AI landscape, you know, the barriers to adoption the new agentic AI. We’ve got a section on that coming up with here in just a minute, and how data centric works and regulatory, again, sections all coming up. So, you know, in the time that we had, they were very, very we used time very wisely, and the ideas presented to us was, you know, quite amazing. Now I know Brian, you and I have presented on AI for some number of years, and you know, we have talked a lot about privacy, and you know, some of the impact of AI, and have been using the conservative guidance of following the EU’s AI regulations, just like we recommended following GDPR and the UK regulations, as well as the Canadian regulations. And during the week of our recording this session for you, we know that Virginia was putting up new AI regulations at the state level. Now, again, I’m not a national you know, policy wonk on this stuff. We do know that the AI executive order that President Biden issued was rescinded by President Trump, kind of day one or two of the of the administration. So right now, we’re pretty much running without AI regulations here in the United States. So it’s kind of Katie bar the door. And of course, the argument is it enables innovation. So you can take your own position on this, but our basic approach right now is, I think it’s safest to follow the EU, UK and Canadian AI regulations, but there’s nothing binding you to do that at this point. So
Brian F. Tankersley, CPA.CITP, CGMA 03:39
go ahead, if the data goes into an AI engine and it can be extracted by somebody else, you are going to end up with, with a data breach that you’re going to have to deal with. So what I would suggest that you do is look at the PII that you have, and don’t put it in AI engines where you’re not where you’re not positive that’s not being used to train, train in there. Because, again, there are, like we said earlier, there are all kinds of different things, all kinds of different ways that, that the training data can be extracted from the AI models. So, you know, again, it’s, it’s not, you know, Randy, Randy’s right, that there’s, there’s not necessarily governing regulations directly aimed at AI, but there’s still a lot of PII regulation that you still have to contend with that will be a problem for you.
Randy Johnston 04:33
Good insight there, Brian, and you know, the some of the topics that were laid out in the overview was, you know, the impact of AI on research. I know I use that all the time. I know you’re very good at using AI for research. Brian, whether it’s, you know, bookkeeping or audit or tax or advisory all there, we know we’re going to have to upskill our people. We know there’s going to be a lot of career evolution. So this is
Brian F. Tankersley, CPA.CITP, CGMA 04:58
a. Say that you and I have had conversations with both Thompson, head, cch, about AI, especially in the context of research, and the thing, the thing that’s holding them back, in many ways, is this, this hallucination problem we have, because the option, there is no option of being wrong in their world, because as soon as they’re wrong, they lose all credibility with you, and then you go somewhere else. So it’s critical to understand that the reason you haven’t seen as much AI in this profession as you have in marketing and sales and other places like that is that, you know, how many times has a salesperson been wrong in your life? Well, let’s just say it’s more than once. Okay, and you know, again, the you you but you can’t do that with with things where people are giving professional guidance. Yeah,
Randy Johnston 05:48
so that’s your attention. Then next Brian to AR regulation. Now Bill Wyman from tech dynamics presented this particular section, and he really was laying out the regulatory landscape. Did a beautiful job of laying this all out. Now Brian and I have followed NIST for some time. We actually, even before beginning our session with you today, discussed NIST. But you know the bottom line here is that with the current AI political climate, globally, it’s going to be hard to get anything passed in the US environment. And then the bottom line here is AI is going to proliferate rapidly, and it will touch everything. I actually have been listening to broadcasts in my commutes back and forth to Wichita about AI and the impact in every profession. And I think a statement that was made yesterday in one of those was pretty insightful, and that is your biggest risk is if you guess that AI won’t impact you, you ought to plan that it will have some impact. Notice, he wasn’t saying AI is going to replace you, you, and I agree, I think on that, Brian, I don’t see a condition under which AI replaces accountants. But the bottom line is, it isn’t going to replace you, but you better assume that it’s going to make a difference in the way you work. And we’ve clearly seen that in other labs that we’ve done for you with Blue Jay and tax GPT and tomuters co counsel, and you know the CC, H AI products, there’s, there’s definite impact out there.
Brian F. Tankersley, CPA.CITP, CGMA 07:31
And I think one of the things you have to remember is that today’s AI is as stupid as AI will be in our lifetime, okay, in the rest of our lives. So it is, it is going to get increasingly better and better and better. You know, I’m looking, I don’t know if there is a new Moore’s Law equivalent. Moore’s law, of course, was used in semiconductors to describe the the increase of computing power on a logarithmic scale. And, you know, the computing power doubled basically what every, every 18 months, is that, right? And it’s still continuing on. So it’s, you know, I think the, I think we just have to remember that as we’re looking at this, it may not a lot of things that are going to affect us. May not be possible today, but they may. They may be table stakes five years hence. So it’s, it’s an interesting time.
Randy Johnston 08:29
Yeah, and you know, one of the statements made in this section was that AI will be ubiquit ubiquitous, general purpose technology over the next five years. And I actually just don’t know how to help you predict how fast these changes will take place, but I know that they’re changing. So you know, the next up, we had a little bit of a view of just the AI landscape. Now, in future technology labs, we’ll talk about agenda AI with you, but and there’s more to come in this session, but if you look at the AI landscape, it has also changed quite a little bit. Matthew keel from Gartner presented this and he talked about, uh, the expectations, many of you are familiar with, the Gartner hype cycle, and he did refer to that in his presentation, but he was really thinking that, you know, AI was supporting thinking tasks. Was one of the comments that was made. And there were time savings being driven by AI, but a universal response during the day about AI time savings is the time seems to just get sucked up into other activities, as opposed to turning into, you know, dollars, if you will. I don’t want to say it’s just more billable dollars, but you know, yes, we’re working more efficiently, but the efficiency seems to be disappearing. Was a per. Pretty universal response so
Brian F. Tankersley, CPA.CITP, CGMA 10:02
but I am also seeing that, at least in my own use of AI, it lets me create more voluminous and better quality work product than I would have much faster in the past. But what that means is the game and the expectation of what I’m going to come out with is going to be much more beefy than what it was in the pre AI world,
Randy Johnston 10:26
yeah, and when we have been using AI for research, and you know, there’s a lot of deep research tools that are out there, but we’re improving our techniques with that too. One of the statistics quote in this section was that in 2024, 58% of finance teams have started using AI, but one of the things they were very clear about is starting may not be using it a lot, so most of the statistics that you’re seeing on AI, they believe are overstated, because People are saying, Yeah, we are. But just how much is it changing your day to day work? So we know there’s going to be much evolution in this. And further this LED then to the section done by Carrie costelec and Samantha dempti from the ESSA Sox services and from KPMG about AI considerations for audit and assurance, and one of the regulatory changes that just happened was the movement of client accounting services over into consulting and out of assurance, a pretty significant change, which, again, we’ll probably talk to you about on another tech lab, but the question about how you audit AI is also one of those, you know, dicey questions. It was also no I want to note with you in the last few weeks that the anthropic folks showed how Claude reversed engineered on its logic, and it showed how it did math and so forth. That was a very interesting paper, which I did post on the social media platforms. But the logic of these large language models is still not well understood, and even the developers of them don’t exactly know what will happen, but the big breakthrough, as we see it, might be coming in agentic AI and Heather Unruh from Salesforce, Senior Vice President of Engineering, did a beautiful job on her presentation. Brian knows me well enough I don’t very often speak of a presentation as just being magnificent, but this one was and, you know, they they really took the view of what Salesforce was doing with agentic AI and how much difference it was making in people that were using Salesforce because of The agents. And they noticed that agentic aI had five key attributes, defining a role, looking at the data, taking actions, working through a channel, and having guard rails to make the thing correct. And it is a agentic AI is a way to offload repetitive task, improving client service, and there’s going to be a fair bit of improvements, I think, in agentic AI to really handle things the way you might as if you were a person. The other thing was the agentic maturity model that that she discussed, because as Heather was going through this, it was like, Oh man, I wish I’d have thought of that one, but they basically talked about it in multiple levels, level zero being fixed rules and repetitive tasks. Level one being information retrieval. Level two, being simple, orchestration with multiple domains. Number level three was complex, orchestration on multiple domains. And level four, multi agent orchestration. So you know, the thinking here is that the agent can do quite a bit, but a human in the loop is very important. And I think that’s true with all AI, but the whole goal of agentic AI is to free up human time for more valuable work. And she quoted some statistics in her presentations. They said that they would distribute those presentations, but I don’t have a copy of that at this point for our recording for you today, but the statistics were stunning, and I know that they will, Salesforce would share them with you in a sales cycle, I’m sure well,
Brian F. Tankersley, CPA.CITP, CGMA 14:40
and so I would just say that that one of the interesting things that I think you’re going to see related to a lot of these topics, is, if you have, if you haven’t, used the perplexity search engine slash AI engine combination, one of the most interesting things I’ve seen in some time is their feature called deep research. Page that goes through and does multi multi step AI processing of information and multi step information gathering, summarization reporting. If you haven’t looked at that one, that’s particularly interesting, but you do have to have a paid plan to get that but what we’re starting to see are more and more of these tools and features that do the multi step things in the process, as opposed to you having to go in and say, Okay, now take this and do this. Now take this and do this. You know it’s, it’s more like, more like a more experienced staff person, where you say, go do this. And they know that there are seven to 10 steps in there, and they’re good enough to come up with the right ones in there. So this, again, this deep research thing is a is something that is going to have a huge impact on us. And I hope that you hope you check it out, because I think it’s a huge productivity game in using AI. All
Randy Johnston 16:02
right. Now, one thing that I did take note of, because I knew I wanted to have the number in front of me for one 800 accountant, retail accounting, using Salesforce, they claimed that they had resolved with agentic AI, 1000 inquiries in the first 24 hours, with 50% fewer cases submitted. That’s pretty big numbers. In a big four consulting firm, they said that they saved 42 minutes per Sales Lead per week, resulting in an extra 100 and 20 million in margin in a big four firm per week, okay, and then last in insurance audit and recovery, they claimed an 800,000 increase In collector productivity with 150,000 of claims deflected to autonomous agents. So agentic AI actually handling and doing the work was one of those things. Again, there were lots of different illustrations that Heather did in her presentation. But again, I’m I’m not easily impressed, I would say, with presenters, even though I am one, this one was impressive. So that then got us to data centric. Ai, several different people presented here, and in particular, in a future technology lab, we’ll talk about the new general ledger system released by digits. Jeff sievert, who’s the CEO there, and Brian, and I’ve known him now for a few years. He talked about what it took to build AI to get a GL to work. And Robert Pacifico, from MasterCard, or senior vice president, talked about their use in Charles Broome, the VP of growth strategy of business credit, also talked about their strategies. But, you know, with data centric AI, of course, Jeff’s big deal with how hard it was to build an autonomous general ledger and get it right, and how they had to hydrate the data to add contextual information and be able to build, then, you know, meaningful dashboards. But several observations in this section are instructive that accounting is not generative. In other words, the generative AI models generally don’t work here. 70% accuracy in AI is easy, but difficulty increases exponentially. And we’ve learned that from most of the tax products that are trying to do 1040, and k1, and so forth. And any shortcuts that are put in these systems are generally traps that come back to bite you later, that training data, scale and variety matters. But we’ve known that pretty much all along, and that a lot of the model orchestration comes from model sizes, which Brian and I have talked to you in the past with. You know, narrow, small, medium and large language models in the new large behavioral model. So data is a big deal when it comes to data centric AI and another piece that’s quite important, and it’s a little technical, but vector based data models are what make these AI models run much faster. So those types of services are certainly in play. Well, then that led us to innovations in accounting and finance. Now, I always enjoy listening to the CTO of sage software, Aaron Harris, and he was talking about AI advisory services, but we also had AI in tax with Dom Magna from PwC, and Ryan Daly. The in this area. And then also Avani Desai from Shellman, talking about building resilient AI systems and accounting firms. And each of these three different presenters, you know, talked about what AI could be used for innovation in mostly public accounting firms the way they presented, but Aaron, in specific, talked about freeing up resources to do hiring higher value work and to be able to do more better and do more different. Kind of Brian’s claim just a few minutes ago with you that everything was becoming more personalized and more intelligent, and we had to establish AI governments and publish our AI commitments to be trusted. All good insights from Aaron with the PwC, folks with Dom and Ryan, they really talked about AI and tax and how to revolutionize tax processing, some of the challenges, the regulatory compliance, the AI powered transformation, and the opportunities that were there, and interestingly enough, how they were also using agents for AI, agentic AI for tax workflow. And that was pretty interesting, because Brian, you and I started talking two years ago about how the power automate workflows could be processed. And the day before we recorded this session for you, Microsoft announced publishing of these new AI agentic power automate workflows. Again, that’ll be a topic we’ll need to talk to you about in another technology
Brian F. Tankersley, CPA.CITP, CGMA 21:47
lab. But let me say this about AI tax. One of the things that is the A that is the elephant in the room is that I think that, I think that what’s going to happen also is the government side the audits. The government audit side of this is going to also adopt AI, and it’s going to make them much more effective at identifying those outliers and those other, those other things that shouldn’t be that are in in tax returns and things like that. So I think it’s, I think it’s going to let them do more with less, just like it lets us do more with less on the on the business side, and so it’s a, you know, again, I it’s, it’s going to provide a framework to maybe make the interactions that we have with the government sector may be a little more effective, and so that gives me some hope that we make it out of this, this challenging world that we’re in today.
Randy Johnston 22:51
Absolutely so I guess, with that said, the last and key presentation, again, just a short one, was about AI systems in accounting firms, and here the Shellman folks basically said you should you already have systems in place, so you just need to overlay those systems to build resilient AI systems. And that it is really a business imperative, and it’s and using the base systems you have is what makes AI resilient. So they promoted the idea of being adaptive again, having human in the loop, having auditability, having data governance and having privacy and transparency overall, a fine day of leading thinkers in AI who provide tools to the accounting profession and to industry accountants. Again, I was so pleased to spend the time learning from all of these, you know, very brilliant in the people in the room. So Brian, parting thoughts for our listeners and viewers today.
Brian F. Tankersley, CPA.CITP, CGMA 23:57
Well, I think, I think, you know, there’s, there’s a lot of hope here with respect to this, but one of the things that we have to all take care of, that I think, is is imperative for for all of you in public practice, is to figure out who’s going to do the most tinkering on this. Okay? Because what I’ve seen with AI is that it requires some tinkering. You have to try things and see if they work, and that requires time. And when you’re living life six minutes or 15 minutes at a time, you know, there’s no code to charge that to, okay? And so you know, realistically, if you want to get the most out of this, you’re gonna, you know, you’re gonna have to let people throw some time at this, and you’re gonna have to, you know, you may want to get some structured learning. I’ve actually signed up for a Microsoft event that’s coming up pretty soon, that’s that’s to help you get more out of out of AI tools, and help you learn best practices. But I think generally, accountants don’t Tinker well because we’re so freaked out by realization and maximizing profits and. All these other things like that. And I think the the thing you’ve got to understand here is this is like any new skill, and you know, every time you started anything new, you were bad at it. You really stunk at it. I mean, you think about riding riding a skateboard as a kid, or riding a bike, you know, you weren’t good at it when you first started, okay, but you had to practice, and you had to get your knee scanned a few times, and then you got good at, okay, the same thing happens with AI, where we’re going to have to go out and throw some time at this and, and I’m struggling with as much as anybody, because it’s, you know, I’m looking at it saying, Should I really be throwing three hours today at this? And the answer to that is yes, I absolutely should, because it’s, this is the classic, you know, if you go back and look at was Eisenhower’s quadrants of productivity, this is the classic, important, not urgent, tool that that can make a huge difference in everything else that you do. So, you know, getting some time and tinkering and gathering best practices and and just generally, getting comfortable with the engines and seeing what you can and can’t do with it, I think, is one of the most critical things that you’re going to do in 2025
Randy Johnston 26:14
Well, I appreciate you citing a Kansas native son. I actually attended a birthday party on Saturday with people from Abilene who also knew Eisenhower and had been in the home in Abilene, Kansas. I mean, it was kind of like old home. Week. I was talking about Eisenhower’s car and, you know, all those things. One other thing you and I got to experience in the last few weeks, Brian was the accelerator program. Tool Laurel talking about how the AI helps with not only hourly but fixed rate engagements in terms of modeling. I thought that was fascinating. But you know, we’re going to have another technology lab on the 2025, accelerator program. We encourage you to listen to that. And if you haven’t listened to the technology lab we produced on Laurel, that might be a good time to go back and look too. We’re excited about the AI possibilities. We’re excited about having you here today. We appreciate you listening in, and we’ll talk to you again soon in another technology accounting lab. Good day.
Brian F. Tankersley, CPA.CITP, CGMA 27:24
Thank you for sharing your time with us. We’ll be back next Saturday with a new episode of the technology lab from CPA practice advisor. Have a great week.
Photo Gallery:
= END =
Thanks for reading CPA Practice Advisor!
Subscribe Already registered? Log In
Need more information? Read the FAQs