As firms across the accounting profession experiment with artificial intelligence, John LaMancuso, CEO of K1X talks about what accountants are getting right — and wrong — with AI.
Transcription (This transcript was prepared by artificial intelligence; in case of discrepancies, refer to the recording):
Dan Hood (00:04):
Welcome to On the Air with Accounting Today. I'm editor-in-chief, Dan Hood. Everyone in accounting is, or should be, working feverishly to leverage artificial intelligence in their practices. The question is, are they making the progress that they think they are? Now, the sobering truth is that they might not be. That's a suggestion made today by John LaMancuso. He's the CEO of K1X, which makes AI-based tax automation solutions. And he's here to talk about all that and what a successful AI implementation looks like and what it doesn't. John, thanks for joining us.
John LaMancuso (00:34):
Daniel, thank you. Appreciate you inviting me. It's a pleasure to be here. Thank you.
Dan Hood (00:38):
Yeah. Well, and this is a particularly timely topic. Everyone's trying to wrap their heads around AI and making it work for them. You've talked about the sort of five false signals of AI progress, right? Things where people think they're getting into AI and they're doing it right, but they may not be. What are the signals you're seeing at accounting firms?
John LaMancuso (00:58):
Yeah, it's a great question. And I think firms are really, really working extremely hard to get AI embedded into the business. And so what we're actually seeing is that AI features, number one, are not just a complete in and of itself competitive advantage because they can be easily replicable. AI is an enabler of business process. And so I think firms are diving in to use AI where necessary in a tactical way, but they're missing the big picture around how do I embed this as a big picture workflow? That's kind of like number one. Number two is we're seeing that there's a lot of pilots going on with some point solutions. And a pilot is great, but it's not a sustainable adoption across an organization. What really great firms are doing well is they look at the organization's processes in their totality and they're identifying how business processes can be improved as opposed to just some niche successful pilot.
(02:11):
That's kind of like two. Obviously there's a lot of demand and excitement right now, and that should not be confused with business value. Business value is derived when you've got organizational optimization, pause because your processes and your workflows all get improved. And then finally, building AI internally, that's a vision that some firms have. And we think that is a critical area to really, really double click into. Yes, there are some firms that can do it, but these frontier models, these large language models that we're seeing in the open marketplace today lend itself to some governance, maintenance, reliability, and challenges that firms underestimate. So you got to be really careful when you think you can go it alone, and you got to be very cognizant of IRS code 7216 and the intentionality of that IRS code and the privacy requirements there. So you got to be careful when you think you can do it yourself.
(03:23):
And then it's the measurement piece, Daniel, how to look at measuring AI by whether or not it improves operations or it just is there to exist within an organization because it changes a tactical step that is today manual, and we believe that AI can fix it. So those are the big areas that we're seeing in terms of areas in which there should be caution.
Dan Hood (03:54):
Right. Well, and it sounds like when you look for commonalities among all those signals, it sounds a lot like what people really need to do is take, before they start to do a pilot, for instance, or start to think about making their own, is to sort of step back and get a sort of 10,000 foot view of a proper AI strategy so that they're implementing across their whole firm or thinking about it across their whole firm, but also thinking about the universe of AI. You talked about all the different rules around the IRS rules around privacy. You talk about what you can and can't do on your own with AI. A lot of it seems like you can jump in, but you also want to take a look at the pool before you do.
John LaMancuso (04:31):
Yep. Yeah.
Dan Hood (04:34):
So what's behind all those signals? Other than that, are there other things you see mistakes firms making?
John LaMancuso (04:41):
Yeah, you've absolutely nailed it. They're evaluating AI through the lens of technology and rather than through true business transformation. So first is AI is very easy to measure in terms of activity. You can look at a component in your business, whether it's a manual process, and it becomes very easy to measure that activity and say, "Hey, look what we've done. We've made progress here. That activity is now being executed through this AI bot. But are we really looking at the operational outcomes? Are we looking at the business workflow in their totality?" And so it's very easy to automate individual tasks, but you have to look at the broader workflow. And so we believe that you need to treat AI not just as the objective, but how does it improve the overall business platform? So if you start automating segmented tactical manual steps with AI, you're going to end up with a fragmented pool of AI bots that are not working in concert and don't improve the broader picture.
(06:03):
And then lastly, there's a big assumption that implementation ends once a technology has been deployed. That's not true. There has these models, these large language models, whether they're proprietary like ours, or they are frontier models like Chat and Quad, these models have to constantly be looked at. And you have to understand the broader context of the business that the bot is operating in. The technology has to be reviewed. It can't be just deployed once. It has to be reviewed and constantly reinvented, if you will, around how is that workflow changed? How does that workflow need to be improved? And how does the large language model AI solution support it? So it's just not, "Hey, let's deploy, let's implement, let it run. Let's get the results, but let's make sure we've got broader context of how the business needs to improve." And that's what we're seeing is some of the mistakes that are happening out there.
Dan Hood (07:13):
Yeah. Yeah. I want to dive into those in a second, but really just given the pace of change in AI, right? It's definitely, you can't just set it and forget it. This is not you get QuickBooks desktop and just put it in and everybody uses it. If QuickBooks has radically changed every 10 minutes the way AI does, it would be a very different animal. So yeah, absolutely keeping a track of it. So let's dive into the mistakes you see firms making, things they're getting wrong.
John LaMancuso (07:40):
Yeah, it's mainly around this big area. It's around speed. It's around pace. And AI, as we all know, that innovation has created intense pressures to move quickly. Everyone wants to keep up with the Joneses. And as the pace of AI innovation speeds up, so does the pressure to move quickly within a firm occur. But speed alone doesn't create lasting long-term advantages. Here's what does. The firms that begin with the overall business needs as opposed to what can this product that I'm thinking about adopting, this AI product that I'm thinking about adopting, what does it produce? That's the wrong approach. You need to look at the overall business need and step back. So sometimes it's important to stop and slow down and look at the entire business needs before there's a quick jump into a particular technology that's AI-driven to create a certain outcome. So look at the entire business needs, slow down, look at the entire business needs, and then figure out your adoption strategy.
(09:03):
It's becoming really easy to access this technology. I mean, for a very modest amount, you can immediately be up and running in cloud or chat, but it's really tough to execute and become a true differentiator.
Dan Hood (09:19):
Right. Well, you hear a lot of firms when they've got individual staff out there experimenting on their own, often not terribly well-supervised kind of thing. It is very easy to get into it. And I think a lot of firms, they say feel that pressure to do it at speed, but it would certainly make sense. Certainly for anything large scale to, as you say, take a pause and think about speaking of taking a pause. We're going to take a quick break right now. All right, and we're back and we're talking with John LaMancuso of K1X. We started by talking about some of the mistakes firms are making or some of the ways they're approaching AI. Let's call it a suboptimal approach to it. Let's flip it around and talk about what should firms be doing with AI? What things should they be doing and maybe how should they be doing?
John LaMancuso (10:10):
Yeah, that's a great question. A lot of firms are getting it right, and those that are getting it right are stepping back and they're looking at their competitive advantage. What differentiates our value proposition from firm A, firm B, firm C? And so they look at that competitive advantage and they say, okay, what's the holistic business process? Where are the workflows that can be automated? And how do we use AI to drive that? And so this whole long-term value creation really, really depends upon adopting that AI in a holistic approach in making sure that workflow integration becomes a measurable outcome. And so that's the key component, Daniel, is first and foremost, you've got to identify the workflow bottlenecks and then step back and understand how does AI actually solve that? And then you have to go about your evaluation process. There's many AI solutions out there.
(11:15):
Step back, look at the bottlenecks, identify the AI that works best, then go deploy and make sure that you're focused on the entire workflow and not just a sub-component. And so sometimes that means redesigning workflows. Sometimes you have to look at the current workflow and say, "Hey, this isn't working. Step A, B, and C. Nope, we got to go A, B, D, E, and then C again." So look at the workflows. AI does not help with redesigning. AI will enable what's been designed and working properly, but a lot of times you have to step back and make sure you got the correct workflow before trying to automate these individual tasks.
(12:01):
So those are critical steps. And then look for solutions that integrate into existing workflows rather than create additional complexity. Let me repeat that. Look for solutions that integrate into existing workflows rather than create additional complexity. Why? Why? Because what many firms find out as they start adopting AI is that there's this immediate results that come out of that tactical initiative. However, when they step back and they look at the entire workflow stream, they end up creating additional workflows that didn't exist in the past. And so it creates this complexity that doesn't really need to exist. And the firm should really stop, slow down, make sure they're not going to create more work because AI will do a lot of great work. It'll do it very fast for you, but sometimes the end result, if it's not measured correctly in the right workflow context, will end up becoming more work.
(13:13):
So you got to measure the success through business outcomes. You got to make sure that the workflows are very, very organized and not creating complexity.
Dan Hood (13:25):
Right. I mean, one thing that's interesting about is there's a degree to which this is fantastic advice, but it's been fantastic advice for regular automation. In the past, people would always say, yeah, if you don't look at your workflows first, you can automate it, but that just means you'll have bad processes being done faster. It's still just as bad, it's just maybe a little faster. And as you say, then I think with AI, a lot of people end up adding complexity to it, and that's to be avoided. Excellent. As you said, a lot of firms have been doing this right, getting ahead of things and so on. What do you think are some of the hallmarks there?
John LaMancuso (14:01):
Yeah, it's really first and foremost establishing clear governance and accountability around the AI. It's easy, very, very easy to immediately test, very easy to immediately pilot. And then the inertia happens. Results are positive. And then you step back and you say, wait a minute, are we really doing this in the right way, the right manner? Do we have the best practices set up? Are we sure that we can govern this correctly? Is there transparency and accountability around the AI? Do we know that we can trace the data back to the source document? And that's extremely critical. You need to establish clear governance, accountability, traceability, and transparency around what the AI is doing because there's a lot of false positives that occur with some of the large language models that we can buy off the shelf. So you have to make sure the governance is there.
(15:08):
Adoption has to extend beyond the successful pilots. I've talked about how firms that are doing it well measure that very, very well at the onset. They look at the whole workflow, they understand what needs to take place, and they're institutionalizing and adopting on a holistic basis. Probably the most important one is that they're looking at an AI as a return on investment to improve operations and client outcomes. And the last statement there is most important. How does this AI improve the service that we're delivering to a client? Client outcomes come in various forms, paying less tax, paying more accurately, timely, et cetera. And so if the AI enables you to deliver better client success, then man, what you're focused on is definitely on point and how you're implementing it is going to benefit the client. So it's not always, "Hey, how do we increase our business, decrease the overall expenses of our business?
(16:21):
How do we create more operational excellence?" We have to keep in mind what our end goal is, and that's to absolutely delight a client. And so the firms that are getting it right, Daniel, really, really understand that they're keeping that in context when they're deploying AI and they're working really hard towards that.
Dan Hood (16:42):
As you say, always worth focusing on that because I think a lot of firms, their goal with AI is to implement AI. They're like, "I got to keep up with that. We got to have some AI in the firm. How do we do that?" And that's, as you say, only the very beginning, the real results got to be some part of the client experience, some part of the firm's experience, some ROI, measurable ROI beyond just, yes, we have an AI thing implemented in our firm.
John LaMancuso (17:06):
Exactly.
Dan Hood (17:08):
Yeah. Maybe we talk just briefly, I'm curious if you have a sense of where people can start. Obviously, we've talked about one of the things that you use to start is to step back and look at all the processes and look at the firm holistically and say, "Okay, what have we got in place now? What can we fix of it?" Or rethink of it as opposed to even before we get an application or some kind of AI in there, just actually looking at it. Are there other things they could do to get started in AI?
John LaMancuso (17:36):
Yeah, for certain. AI is an enabler. It's not the end product. And so one of the things that's critical is to understand current technology. So what are the solutions that we're using today to execute a tax return? What are the solutions that we're using today to collect source documents and information that go into the tax return process? Understand that current technology, and then identify the integration points. So we have a tax prep software, we have a workspace administration tool. Look at those technologies today and understand the integration points. It's very easy to, like I said earlier, take a tactical step in any of those systems or technologies. What's more difficult is to understand how does the AI tooling integrate into those applications and make those applications better? That's the overall goal here. Remember I said you have to look at it, the entire workflow, the entire process, and that requires the hard work around integration thought leadership and integration coding, the actual step of moving data from the AI application into your current workflow.
(19:03):
So that's a real critical step that needs to take place.
Dan Hood (19:09):
Yeah, absolutely. And it's funny, as you described, a lot of these steps are ones firms have completely skipped over and gone right to, "Hey, let's put some AI in the whatever, some part of our workflow, some little element of it with all that thinking and planning and design work needs to go in advance." Great stuff. All right, John LaMancuso of K1X, thank you so much for joining us.
John LaMancuso (19:34):
Really appreciate you having me. Thank you so much for the opportunity. Thank you, Daniel.
Dan Hood (19:40):
Great stuff, and thank you all for listening. This episode of On the Air was produced by Accounting Today with audio production by Adnan Khan. Rate and review us on your favorite podcast platform and see the rest of our content on accountingtoday.com. Thanks again to our guests. Thank you for listening.
