The Auditability Problem That's Breaking Accounting AI
Download MP3Attention: This is a machine-generated transcript. As such, there may be spelling, grammar, and accuracy errors throughout. Thank you for your understanding!
Mike Whitmire: [00:00:00] I think there's going to be this, like series of companies that comes out with an agent that can do a lot of this work and is fairly accurate. And then they hit this auditability wall, and it creates a big problem for the company overall in scaling. So I'm like really fascinated to see how a lot of these future companies get built out.
Blake Oliver: [00:00:17] Are you an accountant with a continuing education requirement? You can earn free Nasba approved CPE for listening to this episode. Just visit earmarked in your web browser, take a short quiz and get your certificate. Hey everyone, and welcome back to earmark. I'm Blake Oliver talking today with Mike Wittmeyer of Floqast. Mike, welcome to the show.
Mike Whitmire: [00:00:39] Thanks for having me on, Blake. Appreciate it. Good to be chatting again.
Blake Oliver: [00:00:42] Mike. All this talk about AI agents is driving me nuts. I see press releases, like, every day. We've got like, Sage, we've got Oracle, we've got workday. We've got all these ERP systems releasing all these AI agents talking a big game about it. Right. And I'm just so skeptical because I hear a lot, but I don't see a lot. I don't hear a lot of accountants using it. And then I saw Floqast put out some videos of the AI agents that are now in Floqast, and I just I love the implementation. I like I love what I'm seeing there, especially when it comes to audit trails. Auditability. Because that's a big question about AI. So I wanted to get you on the show to talk about like Floqast approach to AI and what you all are doing there.
Mike Whitmire: [00:01:29] Yeah. Well, I'm glad you like our approach to it. And you're you're right. There are a ton of agents coming out. You see a lot of marketing around it. One of the things we did was we tried to avoid the noise and put our head down internally and think about how I should be applied to accounting, because, yeah, the Auditability is a unique thing that we have to deal with. Sales doesn't have to deal with that. So they can use agents differently, like if it messes up a lead. Oh well, so be it. But the stakes are much higher for us. So we thought we should just, like, put our heads down, think about our accounting days and our audit days. And hey, what's the right way to use it at the end of the day? Ai is really about automating work, and agents are doing non-deterministic work. And so that becomes a little scary when you're thinking about auditability. So how do you combine the power of that type of work with software code with an accountant CPA brain, and how do you kind of pull all that together to start to automate work? And that's that's where we that was kind of our jump off point for building our AI solutions.
Blake Oliver: [00:02:25] So there's a lot of places to start. There's a lot of things that Floqast does these days. How did you decide where to begin when it comes to what you can do with AI?
Mike Whitmire: [00:02:35] Well, we were very fortunate that we began ten years ago, and we've had the runway to build out a whole platform for accounting. So we really help, um, run the entire accounting operations, everything around the GL or the ERP. Um, so, you know, we started with month end close reconciliations, and now it's much broader than that. Um, we're able to help with compliance. We help with consolidations reporting and really that recorded report process by having all those capabilities already built out. We were then in a position to take large language models and think about how can we incorporate them throughout our platform to drive more automation and more efficiency, which put us in a spot to use it in a variety of use cases, like we have a product that automatically emails different, different like participants in the close to gather information and collect it, avoid that request process. And all of a sudden, if the request doesn't come back properly, we can do something as simple as like draft the email response and get the get the correct information for you. So one like rather low stakes use of AI within the platform, uh, on a higher order is something like having variance explanations automatically drafted for you. You know, you can dig into transactions, understand how accounts change, and start to summarize that for the end user and accelerate that process. Another like nice novel use of AI. Um, lower stakes with all that one. Or you can move into things like transaction matching, which does a lot of the reconciliation work for you. We've come up with really clever ways to leverage large language models within the setup and structure of that, and that's one where like, hey, it has to be, right. We need to be able to audit how that work was done and a much higher stakes, and can also save a significant amount of time if you if you're using that solution.
Mike Whitmire: [00:04:12] So another one, but what I'm really excited about is our floqast transform. And that's where I, I'm assuming that's the video you saw where you list out, you know, you go through, you walk through all the steps. You effectively create a narrative around the work that you do. And a big part of what makes that product impactful is being able to integrate with all the different parts of the Floqast platform to do different things. So, for example, you might want to use remind to request a piece of documentation from somebody. You might want to request another piece of documentation from somebody else. Maybe you need to integrate with your ERP to pull in transaction level data. All of that can then be data massaged and structured such that it can feed into our transaction matching product. Then we can go through the matching process. Then we can populate a journal entry. You can post a journal entry from that. Now if you're using Floqast, that's all a trail of audit evidence. That's then associated to our compliance product. So you can able to go through audit readiness on the compliance side of it. And all of that is possible because of Floqast transform and your ability to massage that data, structure it the right way. It needs to be to then interact with the rest of the platform and automate like truly automate end to end processes. So that's why I say I'm fortunate with our timing that we had the platform and then models hit the scenes and we can use them really well, right?
Blake Oliver: [00:05:29] Right. It's like so Floqast is based on the month end checklist. That was the starting point. Right? So so it's it's it began as workflow software checklist software like you know when I when people ask me to describe it I think well you know think about asana. It's like a just a checklist you go through when you do something right and and Floqast did that. And then the genius was tying it to the Excel work papers. Right. The automatic tie outs. Yeah. Um, and linking all that in your document storage system. So. So that was the starting point. And then that great gave you a great foundation because you can now layer on top all the AI automation on top of all the workflow.
Mike Whitmire: [00:06:19] Well, we we started with the workflow and reconciliations inside of Excel. And that's very familiar for accountants and makes, you know, makes sense. You log in, you can start using it. But we rapidly tried to automate as much as we can and actually or like hey, let's start with Excel, but then try to get out of Excel as much as possible. So it's the fact that we had those tools like AI transaction matching, um, or variance analysis or journal entry posting. It's those types of capabilities we had already built that we were then able to leverage and focus our engineering resources on on AI. I. So but what happens is when you start with the checklist part of it, that's the description of your workflow, right? It's like, hey, do this process. And the way we think about AI is we want to take the preparer of that checklist item, have that be a flow cast agent, and elevate the preparer of that work into the reviewer of the work. And now the reviewer is focusing on the the work where the agent had to make a decision or didn't know what to do. And that's where we have this human in the loop that will answer questions or approve things that are new or different for that process. And that's where it's like, hey, we can empower accountants to automate the really repetitive, rote part of this job. Elevate them into the reviewer of the more complicated work that the agent's now doing. So our goal is to empower accountants to be the ones who do the setup of these automation capabilities and do the transformation work, and then be the ones who are reviewing the things that need to be reviewed for audit readiness. So to me, it's a fundamental shift in the role of the accountant. And we get to just like think differently about what the job is going to look like going forward. And I'm I'm pumped about our approach for it because it allows accountants to do like the more fun stuff.
Blake Oliver: [00:07:57] Right? And the reason that you're able to insert the AI agent into the process and, and me as a controller, be comfortable with it, is because you've got the rails set up with the workflow. It's like we're taking this discrete task that used to be done by a human. We figured out that we can do it reliably with an AI agent, so we can insert that agent into the process, and it can have the audit trail, it can have the audit ability. We've got the the the documentation. It creates the work papers. Right. All of that.
Mike Whitmire: [00:08:31] So and furthermore, if there was a decision that was made or if there's not if it's an unknown, then there's the human in the loop. So that is going to happen right. Not everything is the same. So the business changes. New things pop up. Um, and then we want an accountant to review and approve that. And that's what feeds into the audit trail of the whole one. So like good, good example. Um, this is the one I use all the time when talking to prospects and doing, uh, doing marketing and stuff is we did the benefit allocation journal entry. That was the first one that we handled. So this is like we use Ukg internally for payroll, integrate with Ukg, pull down the data, massage the data through Floqast, transform, uh, structure it. So it then goes into our journal entry module product. It then separates all the debits and credits appropriately allocates to the different states, and then posts into your ERP from there. The first time we did it like it worked properly, the second time we did it, we had hired a new, we had hired our first employee in Kentucky. And so that's a change to the process. That's something new. So rather than like hallucinate and just book it or forget about it and not do it, it then surfaces the question to the reviewer of that work of there's this new column, New Row, that says, KY, Kentucky. I think it is like, is this appropriate? Should I contemplate this? And then the accountant can say yes or no and then have that workflow modified. So in the future Kentucky would be contemplated, that would be booked properly and everyone would feel good about it.
Blake Oliver: [00:09:54] And that is possible because you've got the reviewer in the loop there.
Mike Whitmire: [00:09:58] Yeah, exactly.
Blake Oliver: [00:09:59] Okay, so I love this. I want to we went through really quickly your example.
Mike Whitmire: [00:10:04] We covered a lot in ten minutes.
Blake Oliver: [00:10:06] So I want to like dig deep into it and walk us through uh, you know, more slowly for folks who, you know, are not seeing this, right. They're listening on a podcast episode. Let's let's do an example. So, um, so, so floqast transform. That is the overall like how would you describe that? That's the overall like architecture for all of this.
Mike Whitmire: [00:10:30] So transform is it's a new it's a new portion of the product where you log in and that's where you build your agents. Okay. So we also call it like the agent builder. So you go in there and that's where you start chatting with the product and you're explaining your process in in pretty extreme detail. Like, like this is the setup, right? It's like, hey, we take this data point and then what we do is off on the right hand side of the screen, you'll see what looks like an Excel workbook. So it's like a familiar piece of technology that accountants are looking at. And what happens is as you're chatting through your workflow and your process, we begin populating this Excel workbook. And the purpose of this Excel workbook is that's ultimately going to be the audit evidence. So we'll get to the end of the process. And then we're going to be saving this as our audit evidence for how we prepared all this work. But you're chatting through. So in this uh benefit allocation journal entry example step one, it would be like integrate with Ukg. And then we'd have parameters for how we integrate with our payroll provider, Ukg. We'd explain what data points we need to have pulled down for it so that we could populate our journal entry ultimately. But step one is just explaining the integration and how that's going to work.
Blake Oliver: [00:11:41] And you're just typing this in a narrative form. You're just describing what you need.
Mike Whitmire: [00:11:46] Yeah. And one step at a time. Like rather discrete steps. So like integrate with APIs. Step two pull down information around names, dollar amounts state whatever. Step three populate column A with this, populate column B with this bold and make the header gray because that's the format I like. Like you go through all of these steps. Um oh it's.
Blake Oliver: [00:12:06] Like giving review notes.
Mike Whitmire: [00:12:08] It's kind of yeah. You kind of leave review notes, uh, when you go to review it later. But yeah, it's like a lot of a lot of detail. That's why I compare it to a narrative or a walkthrough. But even more nitty gritty than that. Um, but this is also why we want accountants to do the work, because could you imagine, like sitting down with an IT person and trying to explain it? It would just be such a waste of time. So you might as well just have the accountants doing the work. And then what we do behind the scenes is we're leveraging the models to write software code that then does this work on a repetitive basis. So for the most part, that's going to be standardized work that shouldn't change. But then every once in a while you'll have the like, Kentucky example. And that's where a model is going to be deployed to look at that workbook and do a first pass.
Blake Oliver: [00:12:54] And that's important because you're not just skipping from prompt to populating the spreadsheet. You are using the prompt to create a script that then will populate the spreadsheet in a consistent way, in a deterministic way, not a statistical way. Yeah. Every time. Okay.
Mike Whitmire: [00:13:12] And then when when there are deterministic or statistical things occurring, that's where perhaps a model is deployed. So it's a mix of like code plus models being leveraged behind the scenes. But if a model is being leveraged, the human in the loop reviewer review step is necessary.
Blake Oliver: [00:13:29] Got it. Okay, so we build the script or, you know, we connect to Ukg, the payroll system. We're pulling in all this data and unstructured way where then we're then using the AI agent, the scripting to to populate our workbook. Our work paper. Okay, then. Then what happens.
Mike Whitmire: [00:13:48] Then? In this example, we would get to a step where we feel good about the structure of all the data, and then we would want to leverage our journal entry management and posting product for that. So then we would say, okay, integrate with Floqast product from our close automation suite, integrate with the Floqast journal entry module, and we want the following lines populated. So the following rows. With this information you choose the account number. It's going to get posted to the state that it represents the dollar amount. All that good stuff you go through and structure all of that within the journal entry. That journal entry is then reviewed, approved and posted into your ERP. So that flows right through into your ERP. That is then going to be once that's reviewed, approved and completed, all that that Excel workbook I was referring to, as well as the sign offs review and approval. All of that is saved with your checklist item that was the originator of this process. So we'll have it all feed into your checklist item. And then as a final step, if you are going through if you're a public company going through Sarbanes-Oxley audits, you'd want to feed it into our compliance product as well, because then you have the audit evidence available for your internal audit team.
Blake Oliver: [00:14:56] So auditor is looking at that journal entry posted in the ERP system. They can trace it back to what happened in Floqast transform and see exactly. See the thinking of the model. Do they? How much of the like, uh, prompts and outputs are you saving?
Mike Whitmire: [00:15:13] All of it.
Blake Oliver: [00:15:14] All of it? All of.
Mike Whitmire: [00:15:15] It? Yeah. You gotta. You gotta have it all.
Blake Oliver: [00:15:17] Yeah. So then it's very clear to see. Here's what was sent into the model. Here's what we got out. Here's what it did. It's like showing the entire thought process then.
Mike Whitmire: [00:15:27] Yeah. So like I envision, you know, how it audits work today. We'll do the test of one around our systems connected properly to system integrations work. I think audits going to move way more aggressively towards what looks more like an IT audit, and I envision auditors going into doing that process. And then, yeah, looking through the steps that we've gone through and testing a couple of them, getting comfortable with the script related, uh, technology portions and then focusing most of their time on the areas where a human reviewed something.
Blake Oliver: [00:15:58] I wonder how much of that is going to be done by AI agents as well in not too long?
Mike Whitmire: [00:16:04] I share the same, but at the end of the day, I guess, and then we can veer off into a separate topic if you want to with this comment. But the PCAOB as it exists today would then come out and audit the auditors, so they'd have to look at the agent used for auditing. And then how do they get comfortable with how the agent worked? That's where I, I struggle with agents doing accounting or audit because you don't know how they work. And the people who build these things aren't sure how they work. And then they change every three months to like, staying up to speed on that would be it'd be tough. So anyway, that's that's like a big part of why we built. We built things the way we did.
Blake Oliver: [00:16:42] Well, I just I wonder if you could take that record that you're putting into your compliance product and then, you know, just run the entire record of that journal entry. Right. What happened with it in through a model that's trained to think like an auditor and then assign a risk score to that transaction based on everything you know about it, and just highlight the riskiest transactions for an auditor to look at.
Mike Whitmire: [00:17:11] You know, you could yeah, you could feed in data points. Like you could sit there and say, hey, the more human the loop steps there are for any given workflow, let's highlight that in terms of audit risk, or if there's a history of rejecting the work that the agent did and going back and redoing it like that's a higher risk one. Yeah. If you narrow down the scope of it, yeah. That could be a really interesting use case.
Blake Oliver: [00:17:31] Yeah. How many overrides were there. Right. Like if if anyone skipped a control in the process, then, you know, that's a that's a red flag.
Mike Whitmire: [00:17:40] Clearly that one. Yeah.
Blake Oliver: [00:17:41] Yeah. You know, it's interesting though. It's like, um, we can think about these, you know, future AI auditors or something like that. But honestly, there's still so much low hanging fruit in the world of audit. I mean, even, you know, just like selecting journal entries based on which ones were, you know, manually posted versus automatically posted. I mean, a lot of auditors aren't even doing that these days, right? They're still just picking based on amount and some sort of random statistical selection.
Mike Whitmire: [00:18:15] What are what are you seeing? Like you see a lot of marketing and press releases go out about how much Big Four are leveraging AI for helping with audits. Is that in your conversations? Is that actually what's happening in practice or you're not hearing about as much of that?
Blake Oliver: [00:18:29] That's the thing is, I haven't really talked to anyone on the ground using those products. So to me it feels like it's very much still a marketing thing or it's in development. But yeah, I'm not like talking to auditors. I'm not having auditors contact me. Saying like, this has completely changed my life yet. So I wonder, you know, how far out we are from from that really happening.
Mike Whitmire: [00:18:53] And I know you and we, we both talk about it a lot, but the the talent gap and it does seem like these types of tools are necessary given there are fewer staff entering the profession. So yeah. Yeah, hopefully they're actually getting deployed here.
Blake Oliver: [00:19:06] The it's very clear to me that like sampling, testing, it's all going to be done by AI agents very soon. And it just makes sense that big firms will make the investment because it's all about leverage in a big firm. Totally like the the the fewer staff you need to have, uh, you know, under one partner. Right? That's that's more profit to the partner. So like, they'll, they'll do it. That's there's a profit motive for that.
Mike Whitmire: [00:19:36] Like imagine if if our compliance product could feed into the audit software at one of these firms, and it's just living in your audit solution already. The amount of the amount of like profitability and margin you could drive by not. Yeah. And then throwing an agent on the sampling and doing first pass review. And then maybe the human does higher level work like yeah you could drive a lot of that. Uh, my, my dream is us getting plugged into audit software. I think that would be insane for the auditors and for our clients.
Blake Oliver: [00:20:05] Yeah. Right now, it's like one partner to say ten staff, like at a typical regional accounting firm. Okay. Maybe. Maybe higher leverage. Right. Like at a big firm, it could be 1 to 20 or more. But imagine if that, like the future is instead of 1 to 10 people, it's like 1 to 10 AI agents. Mhm. Um I don't know. I mean certainly, certainly the junior staff. Right. Like what are they really doing that is, is analytical. How much are they just following a process? I mean, Mike, I was never an auditor. You tell me.
Mike Whitmire: [00:20:39] Well, I was going to. So I feel like the old man saying this, but I did learn a lot doing the work manually and struggling through it. Like my, my experience in accounting was actually doing all of this work that we're talking about. Automating was actually a very good learning experience for me. And I get concerned around the next generation of staff where they're we're asking them to do the higher order work out the gate. Right. So if if you picture the world we're talking about, all the easy stuff has been done and it's like, focus on this harder thing, which is what you would ask your manager to help you with anyway. So how much I don't know. How do you see that playing out? Like, that's an area I really struggle with.
Blake Oliver: [00:21:22] Well, so help me understand. Give me an example of something that, like you did when you started, that you learned from something manual that we are automating. Amazing.
Mike Whitmire: [00:21:31] So I, um, like, in this world, let's say we go. Really? I'm going to go a little hardcore with it. I remember one of my clients had leased a jet for the first time. Absurd thing. The CEO wanted a jet, so they leased it. And it was the first time this transaction had occurred. So, man, I'm just sitting there struggling my way through it, like trying to understand. Wait, wait wait, you capitalize a jet I don't understand. And that was the debate. And we're going through it and I'm struggling through it. And like reading all this documentation on my own and trying to look at GAAP guidance and like, go through it. And ultimately I get to what I propose as a solution. And then I go to my senior manager and the partner because it was a higher level knew GAAP conversation. So we have that. I get told I did it incorrectly because of one really specific thing that makes it an operating expense instead of a capitalized expense. We then have get to an argument with the client about all this, and ultimately I think it was capitalized ultimately. But that's one where, let's say an agent went through, they would identify it as a like. It's from Netjet. It's a jet and we're plugged into guidance and you capitalize it and we're done with it. And then it would just go to my review level. I'd be like, I don't know this. And then I would go and then the my boss would take a look at it. So I wouldn't have struggled through looking at a contract, understanding how it's structured, reviewing the guidance, like thinking about the nuance of all of it. So it's not like a super easy, you know, do a cash reconciliation example. But that's something that sticks in my mind because it was a rather complex thing to learn about.
Blake Oliver: [00:22:59] But that is the sort of thing that, like, you could use an AI to figure out, right? Take the contract, give it to ChatGPT or Claude or perplexity and ask it to like, figure out the GAAP treatment. I mean, that's that's something that I think these llms can do quite well. Yeah. And so you don't have to then struggle through it.
Mike Whitmire: [00:23:19] Yeah. And I have no idea how that impacts my learning as I come up through the ranks. Right. Like do I I don't know, man. I don't remember as much stuff Unless I struggle through it. If it's just handed to you and done really easily, it doesn't stick in your brain as much.
Blake Oliver: [00:23:36] I mean, it's just like reading.
Mike Whitmire: [00:23:37] The textbook problem. Like, maybe. Who cares? It might be the equivalent of, like, oh, calculators came along and accountants didn't have to do math anymore, so they got to do other things. And maybe that's the continuation of this. Um, and maybe it won't. Maybe it won't matter. Like, who cares that somebody can't do 98 divided by seven in their head? Whatever.
Blake Oliver: [00:23:56] You still learn it in school. Like my son just went through that. He was doing long division this year. Now he's not going to have to do it after this.
Mike Whitmire: [00:24:04] But so so we're relying on school being the really good way where you come out. Like understanding the why behind accounting.
Blake Oliver: [00:24:12] But I guess the problem is right now like so much of accounting is learned on the job. That's the big complaint, right? Is you is you come out of school and you don't actually know how to do anything. So you then you have to learn everything in your first couple of years.
Mike Whitmire: [00:24:27] As we say. That? Is that I remember you and I talking about this a long time ago. Maybe that's more of a factor of starting an audit versus starting an accounting. And when you start an accounting, you actually do the work a little bit more and understand the why behind debits and credits. And then if you move into audit afterwards, you'd probably be better suited for that job. So maybe like maybe it's not as much of an issue if you flip, just flip the whole profession on its head.
Blake Oliver: [00:24:52] I always thought that, like, it's weird that we start as auditors, like most accountants, start as auditors and then go into accounting, which is like backwards. You should learn how to do the work, prepare financial statements, and then once you've done that for a while, then go review the work of others as opposed to like starting off that way. Like how do you review work if you've never done it?
Mike Whitmire: [00:25:17] Makes all the sense in the world. Yeah.
Blake Oliver: [00:25:19] So maybe maybe that's where we'll go.
Mike Whitmire: [00:25:20] I might be projecting that weakness onto technology instead, if that. If that makes sense. Um, so maybe that's the actual fix to the problem.
Blake Oliver: [00:25:30] I mean, the other issue, though is that, like, we're also automating the debits and the credits. I mean, that criticism has been made for years, right? Ever since QuickBooks. Yeah. Um, it took away the, you know, need to do a manual journal entry. Uh, people haven't been learning their debits and credits. I was fortunate to, like, learn bookkeeping on a, on an ancient product where I had to actually make journal entries. And it was it was the kind where you could actually make an unbalanced journal entry.
Mike Whitmire: [00:25:57] Oh, jeez.
Blake Oliver: [00:25:58] Yes. You had to be really careful. Yeah. Um, but like, that taught me debits and credits, and I had to be very diligent about it, you know, and, and I, I had to know what I was doing. It wouldn't fix it for me. Right. So I don't know if.
Mike Whitmire: [00:26:14] The most I learned about accounting was when I was closing the books for, uh, like, group of strip malls in California. There's like some, some dude owned, like, 12 of these, and I was just collecting rent payments and submitting it and doing the books and everything. And yeah, just a little debit and credit struggling through QuickBooks. It goes on QuickBooks desktop back then, just like you do learn a lot physically booking those things and understanding how it impacts the financial statements.
Blake Oliver: [00:26:40] The most valuable thing I tell people is get experience, like booking entries, journal entries, and then go look at what happened in the financial statements. Yes. Right.
Mike Whitmire: [00:26:48] Just yes.
Blake Oliver: [00:26:49] Immediately. Just keep it up on one screen and refresh and see what's happening. And make sure that you actually connect those debits and credits to what's happening in that, in the statement of cash flows and whatnot. Yeah yeah yeah. So yeah, that's that's the problem though, is that like, we're also getting to the point where we can just feed a transaction into an LLM and get the journal entry, which that's Floqast is automating that too, right. Or you have the journal entry feature. Yeah. Yeah. And am I going to can I do that now, or is that something that you're you're building where I could just, like upload a document for a transaction and then it creates the journal entry.
Mike Whitmire: [00:27:32] Yeah. Yeah. So like that's a great example. That product's been in market for close to two years at this point. So then it's a matter of okay, we have this automation feature available and now we can plug transform into it. And rather than manually input a journal entry or start with a workbook, you can just have it populated by transform.
Blake Oliver: [00:27:50] Right. And with enough examples of what you've done in the past. Right. The AI can very reliably produce the next one. I mean, it's really good at Sally. It's like the best.
Mike Whitmire: [00:28:04] Yes. So for any given workflow, very good at repeating it over and over again. And that's the notion. But one of the challenges we've seen is um, accounting departments are really unique, their workflows. And so I can't necessarily take that benefit allocation example and go wide across our customer base because everyone's going to be a little bit differently. Number one, a little bit different. Number one, number two, nobody's comfortable having that data shared with anybody else. Just like that's the reality of the market and security right now. Nobody's comfortable with that. So you wouldn't we won't be able to accumulate a large enough data set to have that be impactful or meaningful. So I see two challenges on that side. And again, go back to like the way we built it, uh, I think is the best way for accounting and avoids those issues.
Blake Oliver: [00:28:51] Yeah.
Mike Whitmire: [00:28:51] Yeah.
Blake Oliver: [00:28:52] You also mentioned variance analysis as a way that you're applying. I tell explain that.
Mike Whitmire: [00:28:59] So variance analysis there are two there are two samples or examples of that. That's going to be your flux analysis. So just how much is my balance sheet or income income statement changed period over period. You're going to set materiality thresholds. And when you trigger one of those thresholds then you're required to leave an explanation around why that balance changed. Um that's really where accounting kind of crosses over into finance and starts to work with the CFO a little bit more in the finance team. And then you have the budget to actual process. So comparing where budgets and actuals shook out and same deal with the materiality threshold. Now once you trigger materiality threshold it's not rocket science to go through the transactions. Like if cash went up by $10 million and 30%. Let's look at some of the biggest debits that hit the cash account. That's going to encompass the majority of the increase there. Let's look at some weird offsets. Maybe there were some credits that we want to make sure we capture in case there's something weird on that side of it. But you can look through all those, we can summarize those. We can send an agent in to understand that and then draft a paragraph listing out the transactions and coming up with the explanation there. And that's one where that's not like the explanations I did in the past. They're not these wonderful essays on how things change. It's like balance went up because of boom, boom, boom, boom, boom, boom, boom, boom. Feels reasonable.
Blake Oliver: [00:30:15] The list.
Mike Whitmire: [00:30:16] Of transactions. Right? Yeah. Right. So so that's like a rather simple use case of it, which saves you some time. Um, and it's nice to have, you know, but it's not, like, as meaningful as one of the reconciliation products.
Blake Oliver: [00:30:29] What about, like, financial statement narratives? That's when I've been able to do myself. Right. Drop the financial statements.
Mike Whitmire: [00:30:37] In.
Blake Oliver: [00:30:37] Financial statements into a chatbot and ask it like, give me the executive summary.
Mike Whitmire: [00:30:42] Yeah. Yes. That feels like a very cool use case to me. I think there's some opportunity there. And there are a couple companies that I see up and coming working on that. Um, that's one where you could absolutely have a do a first pass and a draft at it, and then you would just review it and perhaps make some tweaks based on what, you know, and then be good to go. Um, was that your experience, like, was or did it come out and you didn't feel the need to make edits?
Blake Oliver: [00:31:06] So a year ago it would come out and it would be wrong.
Mike Whitmire: [00:31:10] Chunks of oh, just dead wrong.
Blake Oliver: [00:31:11] Dead wrong. Right. Didn't get the numbers right. But every time there's been a new model that's come out, it's gotten scarily good. And the last time I did it, I've been doing it on the same set of like AICPA nonprofit example financial statements. And every time I do it, you know, like ChatGPT oh three Claude four I cannot at first pass find any issues. Wow. You know, that's cool. I, I recommend everyone listening. Give that a try. Right. Um, do Claude for opus or ChatGPT oh three or whatever the latest model is. When you listen to this and take a set of financial statements, public financial statements. Yeah. Um, and drop them in and ask for like, an executive summary. Um, and that's all you have to do at this point in terms of the detail of your prompt. So just give me an executive summary of these financial statements. But if you want to see something really interesting is say, give me an executive summary of these financial statements for a non-financial user, somebody who's not an accountant, somebody who's not.
Mike Whitmire: [00:32:11] A.
Blake Oliver: [00:32:12] Finance expert, you know, give it to me in plain English, because that's always one of the hardest things, is explaining what's going on in financials to a non-financial user. And like that translation is something that gpts are really good at because they're translators, they're Transformers. Um, that's.
Mike Whitmire: [00:32:32] That's really interesting.
Blake Oliver: [00:32:35] Yeah.
Mike Whitmire: [00:32:35] Another thought to like expand on that is when you're prepping your financial statements, every number within the footnotes, management discussion, everything is supported by a piece of audit evidence, like the auditors are going through and reviewing literally every single number within there. I wonder if you even unleashed, if you unleash an agent to go into the detail workbook that exists behind every number within there, if you would have like an even more robust financial statement that accurately reflects all the transactions that would have occurred.
Blake Oliver: [00:33:07] So like create a project and upload all the workbooks behind every number in the financials, and then basically anticipate the questions you're going to get like that that you can't answer on the spot.
Mike Whitmire: [00:33:20] Yeah.
Blake Oliver: [00:33:21] Now, you could write if somebody asks you like, okay, what's going on with this like lease obligation number. Like you can instantly know. That'd be fascinating.
Mike Whitmire: [00:33:31] Yeah. Like, Lisa, lease obligations is a good example. The detail may not be in the footnote, but the footnote is on the audit side supported by the detailed workbook. And you could grab whichever lease was the biggest addition for the period.
Blake Oliver: [00:33:44] Remember when Silicon Valley Bank collapsed. They they collapsed because they had held to maturity securities. And they had all these held to maturity securities that they then had to reclassify as available for sale. And it sparked a bank run. Right. This was all in the notes, the financial statements for a few quarters leading up to the bank collapse, but nobody noticed it because it's on like page 120 of 140 pages of financial statements in the footnotes. And I wonder if you if you went back and you ran these financials, you know, quarter by quarter through an eye and asked it to look for issues, would it spot it? Um, would it do a better job than a financial analyst does right now?
Mike Whitmire: [00:34:37] Yeah. You like prompted like, hey, dig through these financial statements and understand what are some of the biggest risks to the business at large?
Blake Oliver: [00:34:44] Yeah.
Mike Whitmire: [00:34:44] And I do wonder if it would come back with that. Oh my gosh, you're giving me you're giving me PTSD about that day. Oh gosh. We had a lot of money with SVB.
Blake Oliver: [00:34:54] Oh, man. Um, yeah. I mean, that was a that was a big deal. And it's like, I think this is one of the criticisms, right, of the profession is that like the the financial statements that we produce are really big and complicated and hard to use. Even accountants struggle to use them to like, figure out, is this business going to survive, is it not? There's a lot that's buried in the notes. And so, you know, could an AI basically undo all that complexity for us? I wonder about that. I tried doing it with Seb's financials and the first time I tried, which was probably I know this was a few years ago now, right? First time I tried the context window wasn't big enough. So it couldn't it couldn't even take in a whole set of financials. Uh, that's how long they are, right?
Mike Whitmire: [00:35:44] But you could put it on deep research now.
Blake Oliver: [00:35:46] And now you can do it. Yeah, now you can do it. And I haven't tried it in a while, and I wonder if it would find it when it once the context window got big enough then I could do it. But it couldn't find that risk. It didn't. It didn't notice that risk. Now I wonder if with the latest models it would do it.
Mike Whitmire: [00:36:01] Well, here's here's an interesting one. I, I'm really um, the bill, the big beautiful bill that's going through right now. Um, yesterday I did some deep research on it and I was like, hey, explain the the pros and cons of all like summarize the bill for me and then explain the pros and cons. Um, I'd love to understand how it impacts, um, wealth inequality is a big focus of mine. And what that's going to what that's going to drive. And then like, give me the bull case and the bear case for this bill over the course of the next ten years. And like it did a pretty good job. I'm in the middle of reading it. It was a large essay that got returned, and it's like providing that feedback, but it's pretty impressive how much it contemplated outside sources in conjunction with the actual bill itself, which, like this, is the only way anyone can understand a thousand page bill is by having it get summarized by. Put this in plain English for me, like, what are the ten things that matter in here? It's it's nuts. Yeah.
Blake Oliver: [00:36:52] It's nuts. It's one of my favorite use cases that I, I every webinar I do on I, I showcase that just take take a legal decision, take a bill, drop it in and ask it. Ask for an executive summary. Ask for it in plain English. Right. Like that to me is is just so valuable. Um, and it's how I prep for my show now. Like, you know, we do a weekly news roundup on the accounting podcast, and I could not possibly do all this research myself. I mean, I used to, and it took a lot of time. Right. And now I just drop everything that I want to talk about into an AI agent that basically is like a research assistant. And it does what, like these guys on cable news have teams that do and prepares the notes, and I can read in bullet points and and know what's going on. Um, I can't imagine doing it any other way. It's like a it's like a superpower. I have a team now.
Mike Whitmire: [00:37:45] It's pretty cool. Which, uh, do you have any favorite tools you're using?
Blake Oliver: [00:37:49] Um, I use Zapier. That's my. That's the way I build my automations. So. Okay. I think one of the challenges with, like, building an AI process is if you don't have an interface, right? If you don't have, like a flow cast that you've built, how do you actually, like do this in a systematic way? And so one of the things that's neat about Zapier is you can create like an inbound email address, like a secret email, and it just forwards stuff into it, and that puts it into your zap, into your workflow. And then you can.
Mike Whitmire: [00:38:20] Oh.
Blake Oliver: [00:38:20] Got it. Have it do whatever you want and then output it back via email if you like.
Mike Whitmire: [00:38:26] So cool.
Blake Oliver: [00:38:27] They've also got a Chrome plugin. That's pretty neat. Um, have you ever thought about doing a plugin? A plugin like, uh.
Mike Whitmire: [00:38:34] So we, uh, that might be something we consider now. Yeah. As as the world changes, it's definitely changing how we think about building products.
Blake Oliver: [00:38:44] One of the coolest products that I've seen come out recently is a tool for QuickBooks called Right Tool. Héctor Garcia. The the number one QuickBooks YouTuber created it, and it was basically because he was frustrated with the QuickBooks online interface sucking compared to desktop. So he said, I'm just going to fix all these things that are annoying.
Mike Whitmire: [00:39:04] Gosh, that's still a problem.
Blake Oliver: [00:39:06] Yeah. Wow. So he built a Chrome plugin that, you know, just modifies the interface and gives you the hotkeys that you had in desktop that they never made in online, that sort of thing. And it's hugely popular. And I was thinking like, this is a great approach for any app that wants to like, integrate with systems that are difficult to integrate with because you're just layering on top of the user interface. So you could you could grab data that's not accessible via the API because it's, you know, but it is on a web page that the plugin can access if you go to it. So.
Mike Whitmire: [00:39:44] And is it is his built in a way that's configurable or is that rather standardized?
Blake Oliver: [00:39:51] Um, I think you can like customize the hotkeys to do whatever you want, that sort of thing. Yeah, I'm not super familiar with it because I don't I don't do a lot of QuickBooks.
Mike Whitmire: [00:40:00] But it's interesting. So I know this is at the QuickBooks level, but I my brain spinning I there's probably some application at the enterprise level as we've gone up and signed much larger accounts, and there's a lot of requests for customization. And could you build one plugin that allows everyone to customize, rather than having it be either a build for everybody or give them a developer platform to like.
Blake Oliver: [00:40:22] Yeah, yeah.
Mike Whitmire: [00:40:23] Work with it on their own. This might be a much simpler way to attack it.
Blake Oliver: [00:40:26] I mean, like, yeah, you could just like select. Well, I mean, the plugin could just identify what ERP you're logged into. Yeah. And then just show the options for that. Right. And allow you to customize how it works. Um, just navigate around or just grab data that you need in Floqast without having to use an API. Right. That's the problem is, I mean, this this is the this is the big problem is that there's all these developers, all these companies that are still using systems that never went to cloud or true cloud. They don't have API access. Right. It's like and I remember, uh, at floqast, like you guys had to hack together FTP file uploads in order to get trial balances.
Mike Whitmire: [00:41:09] Dude, it's still like that.
Blake Oliver: [00:41:11] Right? Yeah, still like that. So. So, like, how do you get all the benefits of AI if the data is locked up? You can't. Yeah.
Mike Whitmire: [00:41:21] Agreed. It's very difficult. Yeah. We we we still have prospects we talk to and even existing customers who are choosing to go with on premise solutions today.
Blake Oliver: [00:41:32] Wow.
Mike Whitmire: [00:41:33] And so when we talk about AI, you know, obviously I'm in my like tech bubble here talking to other other people. And everyone's like, oh, accounting is going to be automated in two years. I'm just like, there's a big difference between the tech being available versus the tech being adopted. And that 25 years into the cloud, we still have people purchasing GLS and ERPs that are not cloud based. And so, um, yeah, I think the, uh, pressure on adoption is just going to be significantly slower than the build of all this technology.
Blake Oliver: [00:42:03] Yeah. And and that's why it's like, even though I get excited about all this I potential, I realize there's still just so much low hanging fruit, just basic automation, rules based deterministic type automation. I mean, you guys are probably still selling the heck out of just the original flow cast checklist.
Mike Whitmire: [00:42:23] Huge pain. Still a huge pain.
Blake Oliver: [00:42:25] Yeah, totally. And it's amazing. Um.
Mike Whitmire: [00:42:31] And I think another another interesting thing is, uh, a lot of more technologist backgrounds are going to get into the space because there's such a perception that it can be automated. And so I think there's going to be this like series of companies that comes out with, uh, an agent that can do a lot of this work and is fairly accurate. Um, and then they hit this auditability wall, and it creates a big problem for the company overall in scaling. So I'm like really fascinated to see how a lot of these future companies get built out.
Blake Oliver: [00:42:58] Right. Yeah. Because if you're if it's not auditable, it's not usable. In the end, or you then have to redo all that work and you pay for it on the audit side.
Mike Whitmire: [00:43:09] Yes, exactly. Exactly. Yeah, yeah, you nailed it. I was like, man, all right. If we have agents doing all the work, that's awesome. For accounting, that means the auditor has to recreate all of the work.
Blake Oliver: [00:43:19] The work papers?
Mike Whitmire: [00:43:21] Yeah.
Blake Oliver: [00:43:21] Yeah, we're just doing it after the fact. But, you know, that's going to happen. You know, that's going to happen.
Mike Whitmire: [00:43:27] With many, many auditors are going to have to be replicating work as a result of this. Yeah.
Blake Oliver: [00:43:32] See this Mike, this is why we will never be, uh, at a lack for jobs in accounting. Like, no matter what we automate, there will be plenty of companies that screw it up, and we have to fix it, right? That's that's our role in society is cleaning up messes.
Mike Whitmire: [00:43:48] And there's regulation behind all of it. Uh, so as long as the SEC and the PCAOB continue to exist as entities, which I need to get to that part of the big, beautiful bill here.
Blake Oliver: [00:43:58] Um, well, yeah, you brought it up. The big, beautiful bill, cuts the Pcob.
Mike Whitmire: [00:44:03] Is it? It merges it with the SEC. Is that the proposal or is it changed to cutting it altogether?
Blake Oliver: [00:44:09] It eliminates the pcob and just restores its responsibilities to the SEC without, I think, any additional funding.
Mike Whitmire: [00:44:16] How do you feel about that?
Blake Oliver: [00:44:18] Um, so I'm torn because I have interviewed and talked to a lot of folks who believe very strongly that the Pcob has improved audit quality since Enron. Mhm. I though do not see evidence for that. I have not found any objective evidence that audit quality is better. And the main issue is that the way that the Pcob measures audit quality is fundamentally flawed.
Mike Whitmire: [00:44:47] How is that?
Blake Oliver: [00:44:49] When they select audits for inspection they don't do it randomly. They they take a risk based selection approach so that measurement of audit failures. You know, the part one A deficiencies. The part what I forget the other one. I mean there's they're not selecting like an auditor would. I mean I guess they are. They're selecting based on risk. But it's not a random selection of the audit. So like let's say EY does 100 audits, right. Or Pcob selects 100 audits by EY. If they selected randomly, then we would know, okay, what percentage of audits had issues at EY, but they don't select randomly. They select based on the ones they think are going to have the biggest problems. So that's why you get this odd deficiency rate of like 40%, 50% sometimes.
Mike Whitmire: [00:45:43] So would you, would you bet that the reality is it's any it's less than those numbers. If they're if they're not selecting risky ones then inherently it has to be lower than that percentage.
Blake Oliver: [00:45:54] And, um, I interviewed Christina Ho, one of the board members of the Pcob, PCAOB has been very critical of of the Pcaob's approach. She's the only one who's ever dissented in the whole history of the Pcob. And she says it's if you look at actual restatements, it's closer to 5%.
Mike Whitmire: [00:46:14] So, okay.
Blake Oliver: [00:46:15] It's somewhere between, you know, that 40% number and the probably 5% number.
Mike Whitmire: [00:46:22] Oh, so it's like they find deficiencies in 40% of them. However, they're only significant enough to result in a 5% restatement rate, right. Yeah. Okay.
Blake Oliver: [00:46:32] And I think overall the restatement rate for like public companies is even lower than 5%. Right. It's very small.
Mike Whitmire: [00:46:40] Another another thought is as part of all that Sarbanes-Oxley was also introduced. Is it? Yes. The implementation of Sarbanes-Oxley that made the quality of the audits higher? Or is it the fact that the PCAOB was auditing audits that made it higher? I think you could probably argue Sarbanes-Oxley was the one that improved the quality of audits, not the PCAOB. I will tell you, though, when I was at EY, we were always scared of a PCAOB audit. So like it was a it was a thing that drove behavior. And if that's the purpose of it, it got it got the job done with me and the team I worked with.
Blake Oliver: [00:47:15] So it incentivized you guys to do a better job.
Mike Whitmire: [00:47:17] Yes. Oh, yeah. Fear.
Blake Oliver: [00:47:18] Yeah.
Mike Whitmire: [00:47:19] Fear. Fear based incentive. Yeah.
Blake Oliver: [00:47:21] So I buy that. I, I feel like the issue is that there's just no objective way to measure this, right? Yeah.
Mike Whitmire: [00:47:28] There's no quantifying a feeling.
Blake Oliver: [00:47:29] They didn't write, but they should have. That's what they should have done from the beginning, is they should have said, all right, we're going to randomly select audits. And we're going to measure this over time. And we're going to see if we can improve the quality. And they just never did it. So there's no proof that like the $400 billion a year or $400 million a year, I should say 400 million that the Pcob spends like there's no there's no way to prove it, that it helps. So I feel.
Mike Whitmire: [00:47:59] Like the reality is this is a standard argument in society where it's black and white on either side. The answer is somewhere in the middle. I think probably how it shakes out.
Blake Oliver: [00:48:10] So that'll be like crazy if if Pcob gets dissolved. I mean, that'll be the biggest shakeup in audit since.
Mike Whitmire: [00:48:16] Yeah, yeah.
Blake Oliver: [00:48:16] It was created since Enron.
Mike Whitmire: [00:48:18] No additional funding is effectively cutting the organization and it's just not going to happen.
Blake Oliver: [00:48:23] Right. So there will be yeah.
Mike Whitmire: [00:48:26] Are there going to be audits of the auditors still or will that just not be a thing anymore?
Blake Oliver: [00:48:31] I mean, unless SEC comes up with a program and a way to fund it, we're back to peer review.
Mike Whitmire: [00:48:39] Yeah okay. You're on peer review.
Blake Oliver: [00:48:42] That's what it was before right. It was all the big firms peer reviewed each other.
Mike Whitmire: [00:48:47] At least that's something.
Blake Oliver: [00:48:49] Yeah. But it's your buddies I know. You know, like.
Mike Whitmire: [00:48:51] You know, I don't feel great about that.
Blake Oliver: [00:48:53] You're not so afraid of your buddies reviewing your work.
Mike Whitmire: [00:48:56] I wasn't aware it was peer review before. That. Okay.
Blake Oliver: [00:49:00] Yeah. And that's, of course, why we got into the situation. Because the firms would go easy on each other. I mean, that's my theory. I don't have any proof for that. But, like.
Mike Whitmire: [00:49:08] I think that's fair. I think it's the way the world, the world works.
Blake Oliver: [00:49:11] Yeah. So I don't know. Yeah. It would, I would like to see really independent auditing of auditors. Um, but it looks like it looks like the deregulation push is going to, you know, impact us. You know, the other thing about the big, big beautiful bill, the tax bill, uh, is that if, uh, the, the accounting firms or the partners in, like, blue states are getting hit pretty hard with this, uh, limit on the state and local tax deduction.
Mike Whitmire: [00:49:40] Yeah.
Blake Oliver: [00:49:42] So I, I find that to be a bit ironic, because I imagine that quite a few of them supported the administration. Right. Supported Trump. And now they're getting hit with probably tens of thousands more in taxes every year because they're losing their pass through deduction. So the salt tax.
Mike Whitmire: [00:49:59] The cap is being extended.
Blake Oliver: [00:50:02] So that there's a the cap has been lifted. But then there's a phase out when you get to 500,000 for a married couple. So if you're a partner at a big firm you're making more than half a million a year, right? Yeah. And so it phases out and goes back to ten. And then they eliminated the pass through option, which was the workaround where you could deduct your state and local taxes at the partnership level.
Mike Whitmire: [00:50:28] Okay.
Blake Oliver: [00:50:28] That's gone now. So now there's no deduction for the high earners in the blue states. So basically like you know if you're a partner at a big four firm in New York or LA or, you know, Chicago, yeah, you're paying a lot more in taxes. So the only people that are going to actually pay more in taxes under the, under the bill is like a fifth of the top 1% who happen to be professionals like doctors.
Mike Whitmire: [00:51:01] Lawyers, salaried professionals in blue states are gonna bear the brunt of the tax burden here.
Blake Oliver: [00:51:08] Yes.
Mike Whitmire: [00:51:09] Wow. Yeah. Fun stuff. Fun times.
Blake Oliver: [00:51:12] So that's the world. That's the interesting place we live in. Um, Mike, let's take it back to AI. Maybe for, like, one more question before we go here. Okay. So we've painted a picture of AI agents now becoming the preparer so that accountants on Floqast can shift into the reviewer role and spend more time doing that. Where do you see? Where do you see the profession heading? You know, what is your hope? Are you optimistic? Are you pessimistic? Is it a little bit of both?
Mike Whitmire: [00:51:51] Uh, yeah. So I'll say a little bit of both. I'll start with the pessimism side. I do, uh, I do have that concern around learning without doing the groundwork, but rather small. Let's assume we can overcome that challenge and we'll be all good. Um, I think one of one of the things I've noticed when you talk to accountants about AI is there's a lot of fear around it, taking jobs and kind of had this epiphany. I feel like it's our it's our job to paint the vision of the accountants role and what that looks like going forward. So the way I think about it is it will be much more like the merging of an accountant with a software engineer. So you have the accounting knowledge, supplemented by software engineering tools like a Floqast, where they can then take their accounting knowledge, use our product and automate their own work, and then be in that position to review the work and do higher level, higher level work. So that to me is going to be the future of an accountants role. It will look more like the title would probably be more something like an accounting transformation Information manager or something to that effect. Um, and you're the one deploying the agents, but also then reviewing the agents for your team.
Mike Whitmire: [00:52:57] And that's how we make a more scalable function. Now, if we think about job security for everybody in the future around that. So people of all ages, if you're coming out of college or if you're younger in your career, like get really good at technology, learn these tools as they come out, continue to learn about accounting. And you're going to be a very, very valuable employee like going forward. If you're more experienced, this is where we need you to be really great at reviewing the work that's being prepared. Continue to be really great leaders, run great organizations and hire great, great talent and retain them. Um, and so there remains a role for everybody within accounting going forward. I think it looks very different. And my hope is that it does a really good job of plugging the talent gap that we talk about so much. And so I think the opportunity is there to change the role of accounting, um, plug the talent gap. And hopefully for accountants like make it more interesting and do the stuff they learned about in college. So that's my that's my bull case for the accounting profession. Like that's what Floqast is trying to make happen. And yeah that's that's my hope.
Blake Oliver: [00:53:55] Accounting cyborgs. That's what we're we're going to become AI human accounting cyborgs. I like that vision actually. Like that. I would rather honestly, I don't know, maybe I shouldn't say this, but I would rather manage AIS than manage people.
Mike Whitmire: [00:54:11] Managing is overrated. You're exactly. You're on to something there. Everyone wants to be a manager, and then when they get the role, they're like, oh, these one on ones. And I have to do performance reviews and oh, this is not fun. It's like.
Blake Oliver: [00:54:23] Exactly.
Mike Whitmire: [00:54:24] Sometimes it's nice to be able to put your head down.
Blake Oliver: [00:54:26] The cfo's there. Know what we're talking about. Yeah. Mike, I've been speaking with Mike Whitmire of Floqast, founder and CEO. So great to speak with you and hope to catch you again soon.
Mike Whitmire: [00:54:37] Yeah, man, thanks for having me on, Blake, I appreciate it.
