The Accounting Podcast

Attention: This is a machine-generated transcript. As such, there may be spelling, grammar, and accuracy errors throughout. Thank you for your understanding!

Blake Oliver: [00:00:04] If you had AI agents doing all that work for your clients, you could make 80% margins on accounting and bookkeeping and and even tax prep services someday. That's nuts. That's like inverse of what it is now, right? Like a professional services firm. Well run generates about 20% profit to the owners. But it could be like 80%.

David Leary: [00:00:27] Coming to you weekly from the OnPay Recording Studio.

Blake Oliver: [00:00:34] Hey everyone, and welcome to the show. This is your weekly roundup of news in the accounting profession. I'm Blake Oliver.

David Leary: [00:00:41] And I'm David Leary Blake. We just saw you. I came up to Phoenix on Tuesday. We had a little offsite.

Blake Oliver: [00:00:46] It was great. We did our strategic planning for the next 12 months for earmark. We figured out where we want to be in 12 months. The quarterly objectives to get there, at least for Q4 anyway, the next quarter. Yeah, the next quarter, which is honestly like as far ahead as I like to think in a small company, in a small startup, like nobody can plan beyond one quarter, honestly. That's crazy. What was cool is that we did it ourselves, and we didn't have to go hire some sort of, like, executive consultant to lead our offsite, because I just used ChatGPT to give me the outline.

David Leary: [00:01:16] Yeah, I kid you not. For listeners, Blake left to go to the bathroom and he came back. He's like, oh, I got a plan of how we're going to run this meeting. He comes back with his phone, so he did it in the bathroom.

Blake Oliver: [00:01:26] And it basically gave us this, you know, very simple strategy to talk about where we each want to be in one year, where we want the company to be, where we want our own roles to be. Once we've set that up, then figure out, okay, what are the 3 to 5 things that we need to do to get there in the next quarter to make progress toward that goal, and then attach some metrics to that? We didn't quite get through to doing all the metrics, but we got at least the, you know, what they call the OKRs out onto paper and figured out figured that out from a product standpoint and head count standpoint. And yeah it's great I highly recommend.

David Leary: [00:02:02] These things.

Blake Oliver: [00:02:03] Yeah. And that actually relates to one of the stories I want to talk about in this episode, which is the consultants under threat from AI, because people like me can now just go to ChatGPT and we can get the standard McKinsey consultant response. We can get the templates, it's got all the PowerPoints and, you know, they're just using templates whenever they come in and do this stuff, right. When when an executive coach comes in and does your offsite, they've done hundreds of them. They're just doing the same thing over and over again. Right. So, uh, it's pretty affordable. I'm going to tell you give it a shot. If you if you're planning your next leadership offsite or you want to do some like planning for your firm, um, sorry to all the coaches out there, but, you know, if you're not leveraging AI to do more coaching, then you're going to get replaced by it anyway. David, I'm going off topic. Let's thank our first, uh, first, let's thank our sponsors for this episode.

David Leary: [00:02:58] Our sponsors for this episode are on pay TV up in digits and I'll jump in on pay right now. Forbes and CNBC rank on pay number one for small business payroll. On pay really knows how to get payroll done right for every client you serve, no matter how complex their software is, easy to use and backed by outstanding service levels, they handle new client onboarding for free and have experts on call to keep you and your clients on track. The system includes multi-state payroll, local tax filings, integrated HR tools, and more with no hidden fees. And when you join on Pace Partner program, you get a custom dashboard to easily manage all clients in one place. Plus, you gain exclusive perks like revenue sharing or discounts, free payroll for your firm, co-branding opportunities, premium swag, and more. Onpay helps you run your practice efficiently while providing exceptional payroll that your clients can count on. To learn more about using Onpay for your firm and your clients, that may be farms. Startups. Restaurants. Bars. Doctors. Nonprofits, gyms, franchises, or dentists. Head over to The Accounting Podcast that is The Accounting Podcast forward slash o A.

Blake Oliver: [00:04:06] We love Onpe here at The Accounting Podcast at earmark. We use it for a handful of entities and it it works great. And I just want to highlight my favorite part is the the zero integration. I know there aren't a ton of zero users in the US, but I'm one of them and on page is nailed it with the integration, everything just matches perfectly.

David Leary: [00:04:25] And you get the controls like inside of because I think same thing with the QuickBooks side. You can map every payroll item to an account. You have control over the sync, which I think a lot of apps don't give you. And then when it syncs wrong, you just get mad because you're not allowed to massage it to make it correct.

Blake Oliver: [00:04:41] And that's and you know, that ties to what we're going to talk about with AI David, because the best AI implementations I see give you that kind of control. And the worst ones give you not control, not transparency. You can't see what's happening under the hood. And you've got a story about the new QuickBooks bank feeds that are AI powered. Being not so popular with Proadvisors they aren't. It sounds like they aren't loving it.

David Leary: [00:05:07] Yeah, it's just to quote unquote, found it created more work than just instead of saving you time. Yeah. The new AI bank feeds in QuickBooks are creating more work. So this is, uh, from Veronica Wasek. So she's a Intuit trainer, so she does training for Intuit. She's a ProAdvisor. And she also has her own YouTube channel. So this is a post she put on her own blog called Five Minute Bookkeeping Calm. So she basically went through and tested the bank feeds. And her conclusion is she went back to the old one. She turned it off. She turned off the new AI bank feeds to use the old bank feeds feature, because it's just not accurate enough, and it doesn't seem to learn the old bank feeds. You could kind of make it learn because you could create rules, right? And it would get.

Blake Oliver: [00:05:52] Every time, every time you did a mapping of a vendor to an account, you could say something like save this.

David Leary: [00:05:59] Yeah or yeah you can. Every time it's Starbucks, always do this. So you're kind of programing it yourself. But now with the AI you don't have those controls. So and her gripes are valid. So you would you would categorize something and then it would do it wrong again. Then you categorize it again. Then would do it again. So you keep fixing it. But it doesn't seem to be learning. And that's the problem.

Blake Oliver: [00:06:19] Ah yeah. This is the problem with AI Jen AI because it's statistical. It's not deterministic, probabilistic versus deterministic. It's not going to always do the same thing every time. And so this is where developers are really screwing up. David I've ranted about this before I think. But I'm going to do it again because if anyone is listening who's implementing AI into an app that doesn't have it yet, don't do what Intuit did. The way that you solve this is you keep the rules based logic, and you add AI to help make the rules. You don't replace the deterministic logic with AI because that's unreliable. You use AI to make the rules.

David Leary: [00:07:03] Yeah, you could. If it's a new client, it pulls up okay. Based on your client's data, here's the suggested rules. And then you can massage them to make them exactly how you want. And that's that's the issue. It's it's just this it keeps making the same mistakes over and over again. It doesn't learn.

Blake Oliver: [00:07:17] And there's no transparency. And there's no way to change the system instructions to tell it what to do. It's all on the back end. Hidden. All the prompting is hidden from you. So this is like, the worst way. I'm sorry Intuit I'm sorry, QuickBooks team, but that's the perfect example of how not to implement AI in your app.

David Leary: [00:07:36] And Intuit screwed it up two ways they rolled out the AI, but then they also changed the the the UI, the experience. So it looks different. It's acting different, but it also looks different. So you're just not. When there is a match, you're not even confident. I feel like I just hit accept the lot because I'm happy it made a match at all and I'm just accepting the matches. But the reality is I don't have time to expand everything out and read what the AI did and make sure it's correct. I'm like I said, it matched good enough and so what? It's matching less and I'm accepting it more blindly, which cannot be good. I'm probably introducing risk to our bookkeeping.

Blake Oliver: [00:08:12] What I want is for there to be a button on the bank feed that says suggest rules, and then it takes you to a screen where it goes through all your transactions, your history, and it says, the AI could do this. And it just suggests all the rules that you could make based on what you've done in the past. And then you just say, apply, apply, apply, apply or accept all. And then you know with confidence that it's going to do that in the future.

David Leary: [00:08:35] Yeah, you really don't need AI to start from scratch every time to recategorize that Starbucks receipt.

Blake Oliver: [00:08:42] Exactly.

David Leary: [00:08:43] Like once AI figures out maybe how to do it once, create the rule and just have the deterministic logic every single time going forward.

Blake Oliver: [00:08:51] And you know why that works? Because that's how humans work. You have to think of an AI agent the same way you would a human. So like you said, David, you wouldn't for you. If you had a human bookkeeper, you wouldn't say, here's a transaction, please categorize it. And then they categorize it. And then you wipe their brain with a, with one of those things like pens from Men in Black. Right. Because that's effectively what's happening here. You you allow them to learn the history. Um, but you also want to, like, know what their knowledge is. So you would ask the bookkeeper, hey, write down what you did in a notepad somewhere, in a document, somewhere, so that you can train somebody else in the future. Somebody else can do your job. And that document, that knowledge should be visible to the business owner. So if you're going to have AI, try to remember how to do stuff. At least give us access to look at the memory, the knowledge that it's creating the manual, if you will. All right. I'm going to get off my soapbox and welcome our live stream viewers. Hello to boring accountant and Yelena. Thank you for commenting, Nathan. Nathan Brown says been following for a long time. Took Blake's advice from a comment on a live stream a few months ago to start my own tax and bookkeeping. Going strong but still worried about AI getting nervous. Wow. Nathan. Congratulations. Um, I feel like personally responsible now for your success. And I am very optimistic. And I would not be afraid of AI if I were you right now. Because as long as you start to use it in your practice, you're going to be just fine. It's going to make you 2 to 3 times more productive immediately. And we're talking like orders of magnitude more effective, with clients able to do more and more and more for them. If I was starting a firm from scratch right now, I could basically jump from just doing like accounting and bookkeeping work to starting to do like CFO advisory work, just because now I have access to this incredible brain called ChatGPT.

David Leary: [00:11:06] On the other side, you're going to get the benefit of AI because small businesses that aren't partnering with you yet are just going to accept the AI stuff that's in QuickBooks and have a big mess, and you're going to have to fix it. Yes. So it's going to make you more efficient on one hand, but you're going to get lots of new business in the other hand.

Blake Oliver: [00:11:22] If you like. If you like cleaning up messy QuickBooks files, which is a great business, which I did for a long time, then you're going to be just fine. Given the way that, uh, all these developers are progressing. We've also got Amy Sexton here, CPA in the chat, Amy says AI is upgrading exponentially, so it's probably best to roll with the AI even if it's rough. I see your argument, Amy, and that's kind of been the argument of the AI tech founders Sam Altman saying, oh, we're going to get to AGI in like 2 to 3 years. So don't worry about that. We're going to have like an AI that can just do everything. And we're actually finding that that is not happening. Like AGI, this artificial general intelligence that's just as smart as a human that acts like a human. That time frame keeps getting pushed out further and further. And I think I know why. I think it's because llms are very similar in many ways to how the human brain works, and that's why they're so effective. But they're also missing one key element that makes us human. And it's the ability for our internal model, the LLM in your head, if you will, to adapt and change over time. Llms, once they are trained, are not changeable. So imagine if you took a snapshot of somebody's brain And you somehow put that into a computer and you could talk with it, but it can't change. It would be like having a conversation with somebody who has no memory. Or if they do have a memory, the way it works is not that they update their model, it's that they write down notes on a pad of paper, and they have to go look at those notes every time you talk to them, so they can remember who you are.

Blake Oliver: [00:13:14] It's like the guy from memento, you know, that Christopher Nolan movie memento who has no memory. And so he comes up, he comes up with this strategy of like tattooing himself and writing notes everywhere to, to to have a memory. That's kind of what these llms are like. So if you approach it that way, then you see there are some serious limitations. But if you can overcome those limitations by giving them tasks that they're capable of doing, given that lack of memory or very low context window. Then you can create really, really powerful tools. Like last week I showed how I created this tool that is basically like a mini Hubdoc auto entry. Next, I created an AI agent and I didn't have to use any code to do it. I did it in Zapier agents, and I've been debugging it a bit this week, and I'm proud to say that this morning I got it to successfully enter a multi line invoice into Xero. I forwarded a PDF. It entered the bill into Xero multiple lines with sales tax, and it allocated the sales tax to each line item on the bill proportionally, and it coded the bill and entered it as a draft, like a human accounts payable specialist doing that one task. That's the sort of thing that AI is really, really going to get good at doing in a reliable way.

David Leary: [00:14:44] And inarguably the QBO bill scan and receipt scan are working very good. I'd rather just send my receipts to them than a third party app, and then have to go to matching and all the other stuff. So I push everything straight, all my receipts and bills. Now I just put right into QuickBooks directly through the. It's funny because everything gets his label of AI now the AI agent, even though it's probably just old OCR stuff that it might know what is AI and not.

Blake Oliver: [00:15:12] Well, that's.

Blake Oliver: [00:15:12] The thing is, we don't know, right? Like it's AI washing.

David Leary: [00:15:16] Yeah. And I don't care. I want to send an email with the, with the receipt. And then the next time I open QuickBooks it's there and just matches in the bank feed. Like I don't really care how it's being done. It could be a human doing it. I just don't care. I want the output. That's what I care about.

Blake Oliver: [00:15:31] Ashley says finally made it to a live show. Welcome to the live stream, Ashley. Great to have you here. And Amy asks as a follow up. So mapping has to be done every time. I assume, Amy, that you're talking about the way my AI agent enters the bill. Um, so. And how it extracts the data from the PDF. The power of an agent is that you can give it any PDF. It can extract the text from the PDF into a structured format. And then it can look at that, that, that block of text and figure out how to map all of the different items in that invoice to where it needs to go in zero. And that extraction, structuring and mapping was basically the value of an app like a Hubdoc, which was acquired by Xero for $70 million. That those rules and in order to get it to be reliable, they would actually have to go and like get every invoice template in the world that they could find and In train with rules and a little bit of rudimentary AI, but not generative AI. They'd have to train their models to like, do the mapping, and that took years and was very expensive. And now a gen AI model that's just plugged into like the generic ChatGPT LLM can do it. That's how capable the Llms have gotten. But if you think about it, that's not a very sophisticated task for a human to do. It's pretty basic. Here's a PDF. Now go and enter it into the accounting system. Very hard to do with code. Deterministic code, but not that hard to explain to a human. To write it all down. To write it all down was challenging, I'm not going to lie. To actually, like, if you had to actually teach somebody how to enter a bill into an accounting system who has never touched one before. Like they've never seen a bill and they've never seen an accounting system. To actually explain that is a little challenging. But I was able to do that and it only took me like a morning, right? It didn't take me years of development work.

David Leary: [00:17:45] Yeah, I think I.

Blake Oliver: [00:17:46] Had.

Speaker3: [00:17:46] A.

David Leary: [00:17:47] Slide like that in a deck once where I was just illustrating the concept, and I had a bill, and then I had the bill UI of QuickBooks, and I had 50 lines across the PowerPoint pointing out all these fields from the bill. It's there's a there's a lot happening on a bill you don't realize.

Blake Oliver: [00:18:01] So those kind of tasks are really good for AI. But like identifying those and then like creating the the the instructions is is kind of challenging. But I really want to encourage more accountants to start trying to use these, uh, agent tools. Zapier agents is a great way to start get access to the beta. I'm going to share my template with everyone. I just I just recorded a video this morning, and I'm going to share that out along with the template. So you could try it yourself and you could customize it for QuickBooks. The great thing is that once you have this installed, I mean, you're not paying a third party tool to do this anymore. You're just paying for Zapier and and um, whatever. Like additional prompt prompts you do to ChatGPT. If you go through the API. Adam Zaki, welcome to the show. Cfo.com is here. Hey, Adam. Adam says future accountants will be trained to check these agents rather than do actual accounting. Good or bad, I see both sides. I do question the ability of accountants who have never done the work to check the work. I think that's like the fundamental problem in auditing right now and why audit quality suffers. It's because you've got all these like, kids right out of school who've studied accounting in theory but have never done the work, and they're supposed to be checking the work of experienced accountants and.

Speaker3: [00:19:16] Somehow.

David Leary: [00:19:16] Does not make it easy to check either, because the AI like they're trying to disclose, right? Even QuickBooks does this on the bank feed. There's like a paragraph that kind of explains its logic, and you need AI to understand what the hell that it gives you like three sentences on why it posted something somewhere, and you have to read it 12 times to like, figure out, okay, I think it's doing this. It's just not even the, Me. Um. You're right. Without the knowledge of you knowing instinctively where something should be posted, the context clues you get back from the eye aren't going to help you figure it out.

Blake Oliver: [00:19:49] Nightlight says, Will we still need accountants with AI? Well, given how difficult it was for me to create this agent, I mean, in some ways it wasn't, but in other ways it is. And get it to work reliably. I mean, okay, I can build an agent that enters a bill into an accounting system. Very narrow task. But you could not build at this point an agent that can do everything that a staff accountant does or like a manager does, or honestly, even just an accounts payable specialist. There are so many things, little tasks that they do throughout their day. You would have to build an agent for every single one of those discrete tasks, and figure out how to plug it in to the data for that company, both the inputs and the outputs. I mean, that's a massive, massive undertaking. Just think if you had to in your job, write down every single task you do and a detailed description of exactly how you do it so that somebody else could copy it who's never done your job before. How would you do that? Can you even do that? Because if you can't, you can't hand that off to AI. You have to be able to write down how to do the job. And a lot of jobs, there's parts that are very rules based and you can write them down. But a lot of jobs like there's an art to it and art cannot be expressed in terms of logic.

David Leary: [00:21:19] And I don't know, I agree about the art, but I also think context, right. We had a situation this week where a customer paid us the an app, auto charged their credit card, and they and the customer also happened to send us an ACH, right? So we got paid twice from that customer. And I don't think AI could figure out what to do with that. They would be very confused by this. But because there's no context, the AI for AI do the bookkeeping. They probably need the 50 email thread with the customer as well to figure out what to happen. It just doesn't have all the context.

Speaker3: [00:21:51] Mhm.

Blake Oliver: [00:21:52] So you'd have to somehow get an AI to be able to write all the instructions down. And that is how you get to like artificial general intelligence is when you can have an agent that makes other agents and then fixes them, and then they just kind of grow like that. That's what you need. But I don't know if you can do that. I don't know if an AI could actually make other AI agents.

Speaker3: [00:22:20] You probably.

David Leary: [00:22:21] Could just.

Speaker3: [00:22:21] Reproduce.

Blake Oliver: [00:22:22] Debug.

Speaker3: [00:22:22] Them.

David Leary: [00:22:22] And train humans 18 years from now to do the work.

Speaker3: [00:22:26] And the other problem.

Blake Oliver: [00:22:27] Is the cost, right? So like even this bill entry agent, if I go through the log of the prompts because it's, it's it's a series of prompts, it's prompting itself. That's how these agents work, right? They have a goal and they have tools and they have instructions. And then they they follow through the instructions. And the way they do it is they read the instructions, and then they think to themselves, they think they have this inner monologue, and they think about what to do next based on the instructions. And then they go and try to do that thing, and then they come back and look and see, did I do it? And then they keep going. And it's a lot of prompts. It's a lot of stuff that I can see on the screen. And I know there's more behind the screen or on the back end and like that's probably expensive. And so the question becomes, even if you can create an agent for every discrete task of a job, how much will it cost to run that symphony of AI agents? If it's like a hundred of them and they're all chugging along doing stuff all day long, it might be cheaper to hire a human being because our brains are very efficient compared to AI gen AI right now is incredibly energy inefficient. Now that might get better. I mean, it will definitely get better, but at the moment it's not. Reminds me of like the that idea from the matrix. Like the whole the whole, uh, the movie The Matrix where the humans are kept around because we're like little nine volt batteries. The energy that we produce about. That's the crazy thing, is your brain runs on about nine volts. A nine volt battery. And if you think about how much energy it takes just to run one of these little AI prompts, I'm really curious, actually, what the comparison is.

David Leary: [00:24:16] So what would you estimate is the cost to do one bill based on what you built using AI for dollars?

Blake Oliver: [00:24:23] You know, that would be a good experiment. I should, I should basically I should try using, uh, a prompt to like, like a deep research project to figure that out. I know that when we. So when we create courses on earmark, this is my big AI accomplishment that I like to brag about. And this is why earmark can add 100 courses a month, and why our costs are so low is we can take a transcript of a podcast episode and produce a CPE course draft, and it costs about $2 in API credits. So that's definitely like a lot of prompts. It's not an agent that does that. It's a prompt series because agents weren't available to me when I built it. But it's like dozens of prompts. So, you know, maybe, maybe a prompt is like $0.10 or something, I guess.

David Leary: [00:25:17] I mean, you have the cost to having the human do it. You have the cost of building or having AI do it. Ai in the form that we're currently all interacting with right now, or like in your bill scan thing, it's at best you're probably doing it for $5 a bill. It's probably not going to get much cheaper than that for a long time.

Blake Oliver: [00:25:33] Well, I think it's less than that. I think it's less than $5 a bill. But it might be more than what these current tools charge you.

David Leary: [00:25:40] Which is what I'm saying.

Blake Oliver: [00:25:41] Ten $0.20 a bill.

David Leary: [00:25:43] Yeah, $0.12 a bill. Probably on Dex. I just looked at their cheapest plan. Yeah.

Blake Oliver: [00:25:46] So? So at this point, the AI agent is probably more expensive than just paying Dex to do it. But at what point does it flip over? And also the flexibility of being able to create all your own instructions for the agent. That's pretty cool versus having to, like, buy off the shelf software. It's an interesting experiment. I don't know, like I don't know where it's going to go, but I'm going to keep building these things and sharing them on the show as examples of like effective ways to implement AI and back office. The MIT study that we talked about last week says back office is the number one place where you can implement AI. That means that accounting firms doing back office type work have an enormous opportunity to free up their people from doing that kind of work for clients, use AI to do it, and the margins become like software margins you could make if you had AI agents doing all that work for your clients, you could make 80% margins on accounting and bookkeeping and and even tax prep services someday. That's nuts. That's like inverse of what it is now, right? Like a professional services firm. Well run generates about 20% profit to the owners. But it could be like 80%. All right. We have to move on. And I have another AI story that I teased. Excel has added copilot. Do you want to say something else, David?

David Leary: [00:27:07] I was going to read that.

Blake Oliver: [00:27:09] Oh, yeah. Let's do that. Let's thank our next sponsor.

David Leary: [00:27:11] While you prepare your story team up. Are you turning away clients because you can't find qualified accountants? You're not alone. Accounting firms everywhere are struggling to find talent even when they're willing to pay top dollar. The stress of being understaffed leads to overwork, sleepless nights, and burnout. Here's what most firms don't realize. The best accounting talent in the Philippines doesn't want to work for outsourcing companies. They want to work directly with firms like yours. Think about it. These are accountants. They can do the math. They know how much you're paying that BPO, way more than they're receiving. And that's where Team Up comes in. They help you hire accountants in the Philippines directly. No middleman, no corporate politics, no endless turnover. You build a lasting relationship with your talented people who become part of your team. Follow your processes and share your culture. It's like hiring somebody, someone remotely in another state. Plus, you'll save money with with your accountants even more. Everyone wins. Want to learn more about why forward thinking accounting firms are making the switch to direct hiring? Head to The Accounting Podcast. That is The Accounting Podcast forward slash.

Blake Oliver: [00:28:19] Excel. Now has AI built in. There's a new function called copilot. You just type copilot and that'll activate it. And you can enter natural language prompts into your function and have those prompts run on your data in Excel. An example classify this feedback. So you have a sheet that has customer feedback that you've collected, and you point the copilot function at those cells or range of cells. And then it can classify the type of feedback based on your criteria. You can just write whatever prompt you want. And I suppose you could just store prompts in cells in another tab and just apply the prompts to the the worksheet.

David Leary: [00:29:07] Now does it when you hit. So you're going to hit equal copilot and you're going to type in what you want. Then does it replace that with a formula and save that in the cell. Or is it just the AI thing. Because if just leaves the AI prompt in the cell, won't that just continue to maybe not be super reliable?

Blake Oliver: [00:29:25] It. Right. So the ideal thing is that you use AI to, If possible, come up with a deterministic formula and not use AI. But if you can't, then you can use the AI and you bring up a good point, David, because every time the cells update, you have to. There's another call to the LM, right? Which gets expensive. So Microsoft is limiting how many calls you can do. They're only giving you 100 calls every ten minutes, but it only counts as one call if you reference a whole data range instead of dragging the formula across multiple cells. So in my spreadsheet, I could have 100 columns and I could have a copilot function in each of them, and I could update my spreadsheet every ten minutes. And that's the limit. So if I wanted to update it every ten minutes, I guess I could have, or every minute I could have ten columns that use copilot functions. But that's kind of cool because let's say I had one column of customer feedback, like just narrative, like comments. Let's say this is like, uh, I don't know, something people type into, like my comment form or maybe a lead form on my website. I let them, like, type in a description of what they're looking for. I could then pull all that into Excel and that's a column. And then I could create a second column that like asks, I don't know, like identify the tone of this. Right. Whatever. I want to figure out what from this free text that I couldn't do before. And I could have ten of those columns doing that analysis on this submission, and I could update that every minute. That's pretty powerful. But I don't know if that's like across your entire, like all your workbooks or what. Because I can imagine that, like every time you update a field or a cell, then the AI has to run again. This is going to get really expensive for Microsoft.

David Leary: [00:31:22] I remember this might save you time is you're always going to have those situations where you know there's the formula. Like, you know, we've all done this. You have first and last name in one column, and you want to put first name in one column and last name in the next column, and you know how to do it, but you never remember it. So you have to go search on Google and you find the answer like things like that's where it's going to save time is these formulas that you know, but you don't know. And it'll just type the formula for you hopefully. Yeah.

Blake Oliver: [00:31:47] It yeah. If you don't know the formula or like you just we might not have to build those complex formulas anymore for stuff that is like not like super critical. So here's some examples from Microsoft blog post about how you could use this, uh, performing sentiment analysis, classifying text without exporting to other tools, generating marketing campaign ideas and SEO keywords, rewriting messages for different tones or clarity, creating lists and tables that fit into existing models. Distilling large data ranges into narratives. Highlighting trends in complex data sets. Converting technical calculations into plain language explanations that would actually be interesting is like, uh, if you have a cell that has a formula, you could use the copilot function to explain this formula in plain English. That's always the hard part. When you like, go look at an Excel sheet that's been built by somebody else with all these nested formulas, like, how do you ever figure out if you didn't make it, what it's doing? It's really challenging to unwind that. But maybe that's a solution. That'd be really neat. Basically doing technical documentation for you. You have a maximum.

Blake Oliver: [00:33:00] Oh, I said it's 100 calls every ten minutes with a maximum of 300 calls an hour. Excuse me. Uh, there's no direct access to live web data or internal business documents, but that's coming. Dates are currently returned as text rather than Excel's date format. And I guess there's some issues with large array support, which makes sense, because if you have a really large array, that's going to be like a really giant prompt. And probably there's issues with like the size of the context window there. Uh, you have to have a Microsoft 365 Copilot license. And good news, the information sent through the copilot function is never used to train the AI models. So, hey, if you're using the copilot function, I want to hear about it. Let us know how well it works for you going into the comments. Emily says if my employer is prohibited from using AI, should I go somewhere that allows it? Or is the technology not there yet in the government? We aren't allowed to use AI even though we'd save time analyzing docs. That's a tough question to answer.

David Leary: [00:34:06] Story two weeks ago where a huge percentage of employees are just ignoring the rules of their employer and just doing AI anyways. Yeah, so I say do it at your own risk. But pretty much employees everywhere are doing it regardless of the policy. They just don't care.

Blake Oliver: [00:34:22] Yeah. Um, I mean, like, I personally, I can't tell you what to do. Amil. But, like, personally, I could not work for an employer that wouldn't allow me to use AI. Like, I'd. I'd go crazy. But I also can't work for an employer that makes me use Microsoft Outlook. So that really limits my options. Uh, nightlight says how much has using AI impacted your business to not hire a CPA or accountant? Well, I think the reason that David and I haven't hired an accountant to do the accounting is because we actually like it, and it keeps us in the biz.

David Leary: [00:34:56] Like, well, I can't host a podcast if I'm not using these things, these tools and messing around with it. But yes, ideally we should hire an outsourced bookkeeping firm to do our bookkeeping.

Blake Oliver: [00:35:07] But but we don't.

David Leary: [00:35:08] Do it because we get from doing it is.

Blake Oliver: [00:35:10] Yeah, like we can't. If you would you listen to a show where people are talking about like how to do accounting if they never do any accounting? No. That's why we do our own accounting. But we'll probably get help when it becomes unmanageable. Um, the dialect says key disclosure for Emma's copilot is that miss is warning against using it for anything material or important for now. Yeah, of course they are, right? They're trying to cover CYA. Um. All right, moving on. Well, we were just talking about Excel. So let's talk about the professionals that make a living from Excel whose entire jobs exist basically because of Excel. That is fpna professionals, financial planning and analysis professionals. And if you go back in time and you look at a chart of that job growth over time, it was basically the birth of like Microsoft Excel, Lotus one, two, three that created those millions of jobs started to anyway. And it's funny because there's like a trend where you see like fpna professionals going up and bookkeepers going down at the same time, like at the same moment. And it was the birth of the electronic spreadsheet.

Blake Oliver: [00:36:19] Um, there's a new survey of Fpna professionals, 50% believe AI will reduce headcount requirements, with 65% saying junior and entry level roles face the biggest automation risk. They also say, though, that even though AI is going to do the data prep and basic modeling, which it totally can do, if you want to build a financial model, just try ChatGPT. Agent mode it'll make you an Excel sheet with the financial model with all the formulas. It's wild. So even though it's going to do that, the valuable skill is still data storytelling. So being able to communicate, explain the model, explain the numbers. That is still critical. 87% say that those skills, communication and storytelling skills are going to be critical within three years. Finance teams are expected to explain numbers, not just crunch them. Oh, and this goes to your point, David, about security and the question that Emily had about employers allowing you to use AI. So these professionals, 93% of them are using ChatGPT for work, but only 22% have AI policies to protect sensitive financial data.

David Leary: [00:37:39] Yeah.

Blake Oliver: [00:37:40] David, you said you had a story about Barstool Sports.

David Leary: [00:37:45] Yeah. Barstool sports. Um, those of you, it's a very popular came up through the social media world of the internet. Um, some sometimes controversial Barstool Sports, but. But it's what is this huge business empire where they. They're sponsoring college football bowl games. They have their own social media channels, YouTube channels, podcasts. I mean, it's a media empire at this point.

Blake Oliver: [00:38:09] It started as like a gambling app, right? It's a sports gambling.

David Leary: [00:38:11] Then it became a gambling app to now. Yeah, they have their own gambling app. It's this massive media empire.

Blake Oliver: [00:38:17] Um, and there's a location. They have a they have a like a bar.

David Leary: [00:38:21] They have bars now to.

Blake Oliver: [00:38:22] Scottsdale.

David Leary: [00:38:23] Bars and restaurants.

Blake Oliver: [00:38:24] Yeah, yeah. Okay.

David Leary: [00:38:26] The reason the story came here is Barstool Sports is suing an accounting firm. So Barstool Sports has filed a federal lawsuit against Omega Accounting Solutions and O'Brien Sales and Marketing, accusing them of failing to pay millions in advertising and sponsorship fees. So this all centers on a $6 million contract they signed in 2022 that included sponsorship of the Barstool Arizona Bowl, rough and rowdy boxing events, programing around the Super Bowl and March Madness. Um. Barstool claims it fulfilled the obligations, even bringing on Omega's CEO on its platforms, but they have only received a partial payment so far. So they were the agreement was Omega was going to pay that $6 million in four quarterly payments of 1.5 million each, and they haven't paid it all. And so they still owe 4.4 million. They maybe, maybe have done one payment or partial payment so far.

Blake Oliver: [00:39:20] Do we know why this. This is an accounting firm.

David Leary: [00:39:22] This is an accounting firm. Amiga accounting. Com and they actually have a link to their actual blog post. So they omega when they announce this partnership created a blog post. So Omega Accounting Solutions, they're in the IRC game essentially they're not they're a mill. They don't come off as a mill. But in 2022, everybody was trying to get a piece of this IRC action, right. And they they basically wanted to. The way they frame it is how they're partnering with Barstool Sports to raise IRC credit awareness like that. That's how they thought that was going to happen. So but they fully embed. And Omega is interesting because they have all their other partners and they partner with a lot of uh, um, trendy is not the right word. How do I like I don't know the word I want to use on who their partners are. Um, like like a golf X golf, which is like an indoor golf simulator. Orange County Soccer club. Um.

Blake Oliver: [00:40:22] They're like they're like fads or like these businesses that are like, uh, like you said, trendy. Trendy.

David Leary: [00:40:28] Well, so. So what I'm thinking is I here, here's here's my brain. I'm a business owner. I own an accounting firm. I like golf. How do I do? I like sports, how do I get how do I do partnerships to help me with my lifestyle? Oh, the wine country allowance. The brewers. Right. Like they're all their partners are like people you'd want to be associated with to as a lifestyle, right?

Blake Oliver: [00:40:50] Yeah. So so it's a it's a business expense, but, uh, you know, it's it's really for the perks.

David Leary: [00:40:55] It could be. Yeah, it feels very perks. That's the word I wanted. Yeah, it feels very perk ish. So. Yeah.

Blake Oliver: [00:41:00] So that's a new word for me. Perks.

David Leary: [00:41:02] Barstool sports accounting firm.

Blake Oliver: [00:41:04] Suing an accounting firm. Well, I wouldn't, I mean, it's kind of insulting to call an irk mill an accounting firm, don't you think?

David Leary: [00:41:10] Well, it's not clear that any irk mill. But I guess if you go to somebody's an accounting firm's website and IRC is like one of the.

Blake Oliver: [00:41:18] Top.

David Leary: [00:41:19] Thing for menus. Instead of it saying industries we serve and it lists 500 industries like yeah.

Blake Oliver: [00:41:25] All right, all right. Well, we won't judge them until we know more. Let's talk about tariffs. David. It's been a little while when.

David Leary: [00:41:34] Just came out today right.

Blake Oliver: [00:41:36] Oh did it I don't know.

David Leary: [00:41:38] The shipping of goods. Right. If it's under 800 bucks or $600. Used to not have to pay I don't know.

Blake Oliver: [00:41:45] You tell me.

David Leary: [00:41:47] Okay. My understanding is before you could get goods shipped to you from China, if it was under $600, there's no tariff tax. And then Trump changed that to have a tax on China for that. But now he's putting on all countries. I think this as of this morning I heard about this. Yeah.

Blake Oliver: [00:42:03] But I'll.

David Leary: [00:42:04] Let you continue on.

Blake Oliver: [00:42:05] Your I'm gonna I'm gonna ask perplexity what's going on with this. Oh yeah. You're right. The de minimis exception, David. The de minimis exception Perception is that imported packages valued at $800 or less. Were considered de minimis or under the threshold, and they weren't looked at for tariffs. That policy has ended, meaning all shipments to the US are now subject to new tariffs based on their country of origin. So basically this is the death of those like Chinese sites. Temu. Temu is done. Yeah. Direct shipping from China. Uh, other countries. That's how they overcame the. They didn't have to worry about tariffs. So it looks like there's over a billion packages a year that come into the US duty free. And that's not.

David Leary: [00:42:57] Bypassing their shipping one package at a time to bypass the tariffs. Yeah.

Blake Oliver: [00:43:02] Well it's under $800 right. Yeah. So for the next six months, these packages will face a flat fee of 80 to $200 per item or a tariff rate of 10 to 40%, depending on their country of origin. After the adjustment period, all small packages will be charged tariffs based on the new reciprocal rates announced by the Trump administration, which vary by country and are up to 50% for shipments from India or Brazil. Wow. This is actually like this could be huge because, you know, you 80 to $200 per item on an $800 maximum price package is enormous.

David Leary: [00:43:38] And a lot of these packages like my my kid's done this. He's ordered like a little light for his computer and it cost like 12. Look, dad, I got this. It was for $4 and it cost him $4 in shipping, but now it's going to be. That's $8 and be another $80 in a tariff to do the same type of purchase now.

Blake Oliver: [00:43:54] So my question is who who's going to like how is the how is border control custom? How's customs going to collect the tariff on these small packages. Because it's going to individuals mostly. Right. So as everyone know on this show, everyone who listens to this show knows that tariffs are taxes paid by US businesses or individuals, not by the exporter in the foreign country. It's a tax on Americans. It's the individual receiving the package that's legally responsible for paying the new tariffs, not the foreign seller or shipper. So how is that going to work with like a billion packages coming in? How is customs going to collect a billion tariff payments from a billion people every year? Are they just going to hold it there at customs. Like how is this actually going to be implemented? Is my question. So I guess I'm going to have to research this. This seems like impossible to enforce unless you just like, like, are we going to have a billion packages stacking up in the mail somewhere? Like, and I wondered, like when they come in, when the small packages come in off the plane from China, where do they go? They must go through customs somehow. Is like customs just going to like, keep them there until they get a like a credit card payment from an individual or a check or like what?

David Leary: [00:45:20] And some of these items are $2 items. $1.86 like these are cheap, cheap cheap items.

Blake Oliver: [00:45:27] Boring. Accountant says Fedex and UPS will collect prepaid tariffs or refuse to receive the package for shipment. Oh man, it's going to be a giant mess. Frank says what do you think about an AI? Or maybe just an RPA that could go into hundreds of Excel files and make changes? I'm on a large audit where making a change in one spot means you have to do it in hundreds. Interesting question Frank. Um, well, my first question is why aren't those workbooks all linked together already so that you only have to make the change one time and then it waterfalls into the other workbooks like, I'm no Excel master, but David, I know this is possible because you've done it.

David Leary: [00:46:14] But if it's coming from many to one. So if you have 100 spreadsheets all tying back to a parent spreadsheet, that's. And you have to change the 100 other ones, that's the problem.

Blake Oliver: [00:46:28] No, I'm saying you you have one value, right? You change one value in one sheet. And Frank is saying you then have to change it in hundreds of others. But why not just link that value in that one sheet?

David Leary: [00:46:40] The value I'm thinking like a formula. Like the template or something.

Blake Oliver: [00:46:43] Okay, well anything. Right? Like it's this concept of global variables. You always if you're going to use a variable. Yeah. In multiple sheets or multiple workbooks you want to make it global. So you have a tab of like global variables. And then you map your sheets to use those global variables. So Frank I'd be curious to know more about that so that we can answer your question better. Um, because I think that would make a lot more sense, honestly, then trying to use AI to do those changes. Like that's a process problem, not necessarily an AI. Like, you want to fix the process problem, but you theoretically could use AI to go in and make the changes in all the workbooks. But the problem would be, since it's probabilistic and statistical, it might only be 98% accurate. And then how do you know where the 2% errors are, where it didn't update? Now compare that to a human. Maybe a human is also not 100% reliable. And then you might be better off just using the ChatGPT agent mode or something to do that. But I think the problem right now is that I don't know of a way to use an AI agent on the desktop. Nobody's come out with that yet that I've, I've seen that's like being talked about. But as soon as you can deploy an AI agent locally on your computer and then have it chug away. I mean, you could theoretically give it tasks like this.

David Leary: [00:48:09] If I had 100 spreadsheets that are all intertwined and connected to a master spreadsheet, I'm probably not anytime soon. An AI agent going and try to muck with those spreadsheets. It's just so much risk of things going wrong.

Blake Oliver: [00:48:21] Yeah, I would just rather like I think, I think the issue is that your documentation system is broken and it could just be improved without AI. And there's actually like tons of like audit software out there that does this kind of thing where they like, link all the workpapers together so you don't have to deal with this junk. So I said I was talking about tariffs. Oh yeah. Here's what I wanted to talk about. And we have four minutes and then I have to go. So quick tariff update. Um why haven't tariffs boosted inflation? I'm going to admit we were a little bit doom and gloom about this on the show. 18% announced average tariff rates. Now I didn't think that was going to crush the economy. But I did think that if it goes any higher, we're really, really screwed. But the economic news has actually been way better than anyone expected who was worried about tariffs. And it turns out the reason is that the actual effect of tariffs have only been about half of that doom and gloom 18% number. So companies so far, at least as of mid-August, only paid around 9% in weighted average tariffs. That was in May. This is according to a Wall Street Journal analysis. So 9% economists were estimating anywhere from like 12 to 18% that I saw. And the reason is that more than half of US imports have exemptions to the tariffs. So it's really tough to figure out what is the effect going to be when you've got these rates that have been announced.

Blake Oliver: [00:49:53] But then there's all these exemptions. And apparently it's about half. And the good news for the economy is that when tariffs are 10%, the cost can be absorbed. Retailers can absorb a little bit, consumers can absorb a little bit in the form of higher prices, and the exporters can also even absorb a little bit if the retailers push back on the exporters, the importers say exporter. You got to absorb some of this too. And so everybody takes a few percentage point hit and it doesn't kill the the whole model. It doesn't shut down. It doesn't reduce demand enough on the consumer side and profits enough on the retailer side and the manufacturer side to be a huge issue for the economy. It's possible to absorb. So that's that's really it. And what's interesting is that, um, the biggest retailers have chosen to absorb the tariffs and not pass them on to the customers, and they're actually gaining market share as a result. Walmart has absorbed most tariff costs instead of pushing them onto consumers. They've only raised prices on 10% of imported goods and their sales jumped 4.6%, but target sales fell 1.9% as they are battling the perception that their prices are higher than other competitors. So tariffs taxes in this case are actually better for the biggest retailers like Walmart because it puts pressure on their smaller competitors who cannot absorb the hit.

David Leary: [00:51:28] This data you said was from May data.

Blake Oliver: [00:51:30] Yeah.

David Leary: [00:51:32] So I think still in May the baseline tariff was still 10%, but the other ones didn't aren't really going into effect until August 1st. And on like the big the big tariffs on Mexico and Canada etc..

Blake Oliver: [00:51:45] So but those big tariffs also have huge numbers of exemptions. So it might just stay around there. Yeah.

David Leary: [00:51:50] But it makes sense why their average was 10%. Because that's what the that's what the rate was back in May.

Blake Oliver: [00:51:57] I don't know. It's the problem with this whole tariff regime is it's so complex that nobody can really figure out what it is until it actually gets implemented, and by then.

David Leary: [00:52:04] It's whack a mole. Workable.

Blake Oliver: [00:52:05] Just just it might be too low.

David Leary: [00:52:06] The summary on Google.

Blake Oliver: [00:52:07] Or too.

David Leary: [00:52:07] High?

Blake Oliver: [00:52:08] Yeah, yeah. So anyway, that's where we're at. Um, that's why tariffs haven't crushed us and hopefully they won't. David I have to jump to a meeting. This was so much fun. Thank you to all our livestream viewers. Uh, David, if I jump, can you finish out the episode? Read the episode.

David Leary: [00:52:23] Wrap up the digits out here. I'll let you jump in.

Blake Oliver: [00:52:25] All right. Thanks, everyone. See you next week. Bye.

David Leary: [00:52:29] Let's be honest, accounting software hasn't changed much in decades, except for rising costs and declining service. Now there's finally a reason to switch and never look back. You can now do your bookkeeping on digits. Digits is AI native accounting software that works for you and not the other way around. While other platforms just slap ChatGPT on old workflows and call it AI, bookkeeping demands more than a chatbot. It demands precision, auditability, and trust. Digits rebuilt ledger software from the ground up with probabilistic probabilistic categorization and human review prompts the result. Over 95% of transactions auto booked with unmatched accuracy. 54% better than ChatGPT style models so you can close faster, stay in control, and finally stop wrestling with your accounting software. Low value clients are now high value for years worth of cleanups. Now just take two hours all while your clients get visually stunning reports, streamlined collaboration, and insights they need to make better decisions. To see why hundreds of firms are making the switch to digits, head over to The Accounting Podcast. That's The Accounting Podcast. Thank you. Digits. And I might wrap up with one more story here. Let's see. We talked about gusto. I know we talked about Barstool. Gusto. Gusto is buying, uh, the retirement app guideline. So they've been partners since 2016.

David Leary: [00:53:52] So guideline provides retirement 401 (K) type of benefits. Gusto has not really had that as a feature. They've just been partnering. But now they own a company to provide those services as gusto wants to expand its customer base past the 150,000 small businesses that they have and just keep growing up market, and they need to get bigger businesses. You have to offer those retirement plans and healthcare and things like that. So they did purchase a guideline, and guideline has about 400 employees that will all wind up joining the Gusto team. And I think that's it. Perfect. All right. Well we'll see you guys next week. Oh, if you want CPE for this episode, you want to head out to Earmark app. And you can get a free CPE for listening to this episode. And then if you want to get CPE for other shows, we have that as well. You can listen to Tax in Action and get, uh, IRS CE credits. We have she counts. We have, uh, unofficial QuickBooks accountants podcast. So whatever it is you need to listen to podcast on get Credit. We do have those available. You just want to head to earmark app and you get that CPE credit. All right. Thanks everybody.