341 | How to Automate Accounting with Claude AI (and Catch Errors Before They Cost You) with Maksims Sičs

0:00Isar Meitis Hello and welcome to Leveraging AI, the podcast that shares practical ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This is Isar Metis, your host, and we got an incredibly valuable and practical episode for you today. Every business, regardless big or small, has a lot of back-end work that has to happen for the business to run properly.

0:23Isar Meitis This is a lot of mundane kind of tasks that has to happen on regular basis, whether daily, weekly, monthly. And one of the big ones is accounting. And I-- unless you're an accountant, I think most people see accounting as a necessary evil that they have to do in their business, but not necessarily something that they like to do.

0:41Isar Meitis And it also requires being very tedious and meticulous about the process because you cannot have any mistakes when you're doing your accounting. AI is actually really good at these kind of tasks, meaning it is actually, if you know how to set it up correctly, it is actually really good at doing mundane, tedious tasks that are consistent, and it's very good at taking non-structured data, which accounting many times is, and structuring it in a way that will work to the needs of the business or accounting standards and so on.

1:13Isar Meitis So today, our guest, Maxim Six, his background is in accounting, but more than that, he is a ninja in old school Excel. To be more specific, he founded the World Championship of Excel. I did not know there is such a thing, but apparently it's a big deal with a lot of participants from all over the world, and it literally takes the best humans on the planet in using Excel and doing really amazing Excel wizardry and putting them into the challenge of really complex tasks.

1:48Isar Meitis Now When he saw AI agents being able to solve these really difficult problems that took the best people on the planet a long time, solve it in just a few minutes, he understood that there's a much bigger opportunity. Now today, he is the founder and he runs Flowbyte AI, which is a agency in which he helps companies build automations to solve different problems in their businesses.

2:13Isar Meitis And in today's episode, Maxim is going to show us step by step how he uses Claude in order to take a month or months worth of messy ledgers and pull them into a clean, ready-to-go, P&L and balance sheet, all while catching mistakes and so on. So taking, again, all the mundane, tedious work that is difficult and complex to do, that has to be done in a very particular way, and take all of that and fully automate it, giving you the exact things it found, including the mistakes, the gaps, the things you need to pay attention to, and so on.

2:49Isar Meitis Since I know that there is a lot of people out there who are like, "Yes, I would love somebody who help to help me automate my accounting," and I've done actually a really large workshop to a really big group of people from accounts payable in a really large organization, so a over a billion dollars in revenue, they have the same problem.

3:09Isar Meitis So you don't have to be a solopreneur or a small business say, "Oh, I have an issue with, doing these kind of processes." Every single business, large or small, it's a tedious work that is really complex, that takes a lot of effort to do, and being able to do this with AI takes a lot of that effort away and makes it, A, more accurate, B, significantly faster.

3:28Isar Meitis And so because this is applicable for practically any business out there, and by the way, the same is true beyond accounting, like you can say the same thing about other processes in businesses, I think there is a lot to learn from this process, and so I'm really excited to welcome Maxim to the show.

3:43Isar Meitis Maxim, welcome to Leveraging AI.

3:47Maksims Sičs Thank you, Isar. thank you so much for inviting.

3:50Isar Meitis Yeah, I'm really truly excited about this. Again, I've just done a workshop for about 40 people on, doing accounting with AI, but there's many different flavors and many different ways to approach this, and they had a very specific set of problems that they were trying to solve, which is not necessarily everybody's problems.

4:07Isar Meitis So I'm really excited to, learn, what you're gonna show us.

4:11Maksims Sičs Great. Great. so shall I just jump in or, or-

4:13Isar Meitis Yeah, let's dive right in. I think a little bit of a background. I think you have, again, I don't know any other person who participated or know about an Excel Olympics. So if you can tell us a little bit about that part- Sure and how that led you to AI, and then we can dive right in.

4:26Maksims Sičs Yeah, sure. So let me give a quick clarification. So actually, my majors is not in accounting. I'm actually, more a finance guy. is... there is a difference, right? So financial analysis and accounting are two different worlds, Fair although many people think that's the same. So I, I've never done accounting manually myself by, entering numbers and stuff like that.

4:46Maksims Sičs But I did a lot of finance stuff because I, I graduated from finance university. I used to work in a bank. I did, Microsoft Excel World Championship. I was one of the co-founders and, so basically with a team we was developing this as a global competition for Excel gurus. so yeah, I do have a little bit of experience in finance.

5:03Maksims Sičs So now what I'm gonna show is basically how you can automate some mundane tasks, with Claude. And there is, it's important to say it's nothing, extracurricular that you need to know to do this. This is pretty much, is available to the majority of the people. The difference is many people just don't know you can do this, with AI today.

5:23Isar Meitis Awesome. Perfect. great setup. Let's jump right in.

5:27Maksims Sičs Perfect. So let me, let me share my screen. We've got a few demos today, and first thing that I'm going to show would be basically the situation when you would receive, a bunch of general ledger for different, month. so this will be, a standard general ledger for a company where we have debit and credit, a bunch of accounts, de- description, et cetera, et cetera.

5:50Maksims Sičs And the idea is that, from this one, first of all, we would need to build a profit and loss statement and a balance sheet. That's sometimes what accountants needs to do, especially at the year-end when they have to prepare the annual report. And sometimes, the business manager or CEO may request it, in the middle of the year because they are receiving a banking loan, or they just wanna follow some results.

6:13Maksims Sičs So sometimes this will be already automated by some software that, companies are using, but if it is not or you need some specifics, you would just simply open Claude. what is important here, here you can give access to a specific folder and this is exactly what I'm gonna do now. So I'm gonna give access to that folder that I'm currently in, so that I don't need to drop all the documents by hand.

6:35Maksims Sičs Moreover, when a new general ledger arrives, so currently we have six months of general ledger up until May, but if there will be June data coming in, it pretty much takes as much as placing the file in the same folder and the Claude already has access to it. So now once I have this one, I will also pull up my prompt sheet so I don't need to actually place all the prompts myself.

7:00Maksims Sičs and we actually don't need this one. So it's six months of general ledger for my company. Build me a profit and loss statement, balance sheet month by month, January to June. actually January to May, Let me quickly double-check this. No, we've got June. Okay. As a new Excel file in the same folder, use formulas, not hard-coded numbers.

7:15Maksims Sičs Actually, this one is pretty important because sometimes, it can give you a really nice Excel, but it's gonna be just plain numbers, so you don't really know where this is actually coming from. And just, simply this prompt will just improve that, and that's it. let's just send it. So it asks us if we allow the access to the folder, so let's set that we do.

7:35Maksims Sičs And now basically it will take its time to work through the older files, compile the files into one bigger, larger file that will have all the six months, in one, and then it will need to create the P&L and balance sheets from these. And as you may know, if you've got the P&L and balance sheet, you can also build a cash flow statement, which is not asked in this case, but, could be asked in the future because this is something important for business owners specifically And if we're looking at these files, actually these files are pretty standard one.

8:08Maksims Sičs And as we see the totals for, for the debit and credit match, it means as we assume everything was inserted correctly. But as we go through, we will see that not sometimes everything is really correct because sometimes there might be mistakes even though the totals actually match because, the modern accounting software will actually not allow you to insert a transaction unless you have a counter account on the other side.

8:32Maksims Sičs So you may actually enter everything correctly, but there might still be mistakes. Okay. What we also have in these files, basically a chart of accounts, and there's nothing specific. It's just, the account numbers and account names and the categories, you might not actually have it, and that's fine.

8:49Maksims Sičs It will figure out it itself. here it just helps the model to understand straight away where this is actually going. Although, the models now are smart enough to do it by themselves as well.

9:01Speaker This episode is brought to you by the Multi-Agent Orchestration Course. It's a course that I have been teaching personally since April of this year, and every single one of the cohorts that we have opened have sold out. And the people who have been through these courses made incredible transformations for themselves, switching jobs, starting their own companies and so on, and definitely for their businesses by being able to build really complex processes by following our methodology and our infrastructure settings that we teach during the course.

9:29Speaker The course is four sessions of two to two and a half hours each over four consecutive weeks, and through it, you can go from knowing a little bit about how to use AI and using it in ways that most people are using it today to learning how to build really complex, sophisticated systems that can automate more or less everything in your business.

9:46Speaker The current cohort that is open starts on November 2nd, and it's gonna go on November 2nd, November 8th, November 16th, and November 23rd, and we are almost sold out for that cohort as well. It is very likely that we will not open another cohort in December, meaning the next one after that is gonna be in January.

10:02Speaker So if you still wanna learn how to build really incredible automations and learn everything that I've learned by doing this every single day for the last few years, don't miss this opportunity to come and join us on November 2nd. There's a link in the show notes, And if you're going to use the promo code LEVERAGINGAI100, you're going to get $100 off of the price that is on the website.

10:23Speaker Like I said, there's a few seats open, but they are selling out fast. So if you're interested in doing this still this year, go and click on the link right now on your phones and complete the registration and the payment so you can guarantee your seat. I'm looking forward to seeing you there. And now back to the episode

10:39Isar Meitis Yeah. So let's just go back to this to give people examples so they know what we're talking about. So when we're talking about, this part of it on the... no. So go back to the other tab on the chart of accounts. Okay. So this basically tells it the categories and what they fall into, right? So equipment and then accumulated depreciation and then inventory and trade receivable, bank accounts, share capital, earnings, trade payables, all these different things, and then which bigger categories they go into.

11:06Isar Meitis So they go into current assets, equity, liabilities, revenue, expenses, etc. So this helps the AI figure this out. But like Maxim says, the AI is now smart enough to know these things. So even if you don't give it to it, it will probably do a pretty good job. And you can actually ask in the prompt and say if you have any line item that you're not a hundred percent sure about, just ask or give me the options of what you think and I'll help you solve it, and then it will do that as well.

11:31Maksims Sičs Yeah, actually good point because the way these models actually work, they are forecasting the next word, which means that the, these models are not really deterministic. When I say really deterministic, I mean that, you know, sometimes they may just give you a little bit different output. Actually, if you would ask, 10 times in a chat the same question, it will answer in 10 different ways.

11:53Maksims Sičs The meaning of the answer will be the same, but the wording will be different. So here, as I provide the structure, I narrow it down to, to a specific account, a specific category because otherwise it will need to make assumptions. in this case, it's a little bit easier because, for example, this structure of accounts actually gives you, straight away the structure of the categories as well as, for example, everything that starts with five will actually go to liability, and this is the structure we have here.

12:21Maksims Sičs in my local country, like for example, everything that starts with six will actually go to revenue. So It can understand and, look for that information online itself, or we can just, give it straight away. Because otherwise, in some other countries, you will have a different structure of accounts and you will have different account numbers.

12:38Maksims Sičs so it will need to come up with a structure. And unless we give it the specific chart of accounts with specific categories, it will come up with its own. And actually, if you test it, five or 10 times, you might actually gef a little- get a little bit different structure depending, depending on, your accounts and the country, et cetera.

12:55Maksims Sičs now we're just providing this information, beforehand. We are avoiding this sort of mistake or... I should c- I call it mistakes, but it's really kind of internal assumption of a model, what the structure of accounts really is and what is the kind of the structure of the P&L and balance sheets we are looking to get.

13:12Isar Meitis Yeah, and I wanna broaden this concept because it's a very, very important concept. AI is really good, like really good at taking unstructured data. In this case, it's descriptions of expenses and incomes and so on- and structuring it, so knowing how to put it into relevant buckets. However, as Maxim said, it will make its own assumptions on how to create the buckets and which buckets to put it in.

13:39Isar Meitis So the biggest thing you can do for AI in general, and even more important these kind of scenarios, is give it the right context to do it the way you want to do it versus... Because there is-- Like, what Maxim is saying is that if you run it 10 times without giving it this, it's not gonna do anything wrong.

13:56Isar Meitis It will do different versions of right but many of them are not gonna be aligned with your specific need, and the only way it can know your specific need is if you give it access to that information one way or another, either as part of the prompt or as an example or as a tab in Excel or as a reference file or whatever the case may be, or as a skill.

14:15Isar Meitis there's multiple ways, to do that, but you gotta give it your way, how good looks like from your perspective, and then it's very good at following that and actually giving you, solid results.

14:27Maksims Sičs Yeah, very good points, Isar. I like your, your way of restructuring what I just said and just giving 10 ways of right instead of, just, giving random answers. it is exactly like that, But sometimes it misses some data. And actually, a good suggestion here as well, because, I've tested this example myself before the podcast and on, on other data and other examples.

14:47Maksims Sičs So it should have all the necessary data to make it right. However, if you are unsure what kind of data or information it... you need to give it to in order to make this right, you may just, add a, at the end of the prompt, add a single line, just, you know, "Ask me, ask me any questions you need or you need to know, in order to complete the task."

15:08Maksims Sičs Okay?

15:08Isar Meitis Yeah.

15:08Maksims Sičs So and then before it will start its job, it will actually ask you questions you might not even thought about, and this will actually help you to guide AI towards the specific, right outcome that you're looking for.

15:20Isar Meitis Okay. Yeah, I agree with you. That's... I do this a lot.

15:23Maksims Sičs Sure. So we've got the first result, so let's actually check this out. So it says, "I've saved the file in my folder." First of all, let's go and check if it really did. So now, as you can see, we've got not only, six files that we had, we also get this Nordbrew, which is the company we're, we're testing now, profit and loss and balance sheet, January to June.

15:43Maksims Sičs Which is perfect. let's quickly read what else it says. Your, so it has month by month P&L balance sheet. That's all good. How the tracing works. Okay, I'm not gonna read this part. What looks wrong? There are no opening balances. Okay, so it's just the turnover through the period, and then it will need to actually assume what is the starting balance.

16:04Maksims Sičs So okay, that's a good one. then it says customer payments looks like they're missing. No invoice is ever paid after the month it was issued. Okay, that's another cool finding because this is something, people will need to spend time on. And then it says, okay, the VAT is never paid. it tells me exactly what sum.

16:21Maksims Sičs So even those findings already, they give me a lot of information that is really valuable for accountant, especially if you're working with these data, you will quickly understand, oh, really? Okay, so what's wrong with this one? Okay, especially if you're even, checking if you're some, an intern or a junior accountant has entered the data, and you just, a senior is coming and checking how does this work, you will need to spend time finding, figuring this out.

16:44Maksims Sičs So let's take a look at this one. So now we've got the profit and loss statement. Let's actually check, so we've got the general ledger data on a specific sheet, and this is basically the input data that we gave, like six months. And what we actually did, it united the whole thing in one long table, which is really nice because now I have everything in one place.

17:03Maksims Sičs I could have done it myself. That's probably not a big deal. But anyways, it is good that it's there. And if I look at the profit and loss statement, as you can see, it even follows the, financial modeling practices, and these practices state that if we're using any formulas to the other sheets, we're actually using a green color.

17:20Maksims Sičs So green color actually here doesn't mean it's good. It means this is a link, or it's a, reference to, the other data. And as you can see here, it's not just, plain number in the formula. It's actually a form-- a plain number In a sh- in, in a cell. It's actually a formula that references the data that we just ask it to use.

17:40Maksims Sičs And it wrote all those formulas for me. We've got, all the data for the different revenue items. Then we got the couple of, cost of goods sold line items. it calculated the margin, which I didn't ask for. It calculated the first half of the year totals, which are great. I also pretty much like the formatting because this probably the, the formatting I would use, right?

18:00Maksims Sičs So we would have a header, something like this. We've got the accounts, we've got the account names, and then we've got all the numbers right in here with totals bolded, underneath. So this is amazing. It calculated the EBITDA for me, and then we got the net profit and loss. And this company, as we can see, is not doing well.

18:17Maksims Sičs Okay?

18:18Isar Meitis Yeah. I, I- It's not good wanna touch on two very important things in what you're showing. So again, those of you who are just listening and not watching, we're looking at a very well-structured P&L, page with the different standard components that go in there, very well laid out. Every single segment has a header, has a total of that segment, has the percentage, has the, profit or loss aspect of each and every one of this.

18:39Isar Meitis Again, very well structured. We didn't tell it how to build a P&L. So this goes back to the combination of two things that AI is very good at combining for us. One is the stuff that we need to tell it, like how do we want the general ledger to be combined into specific accounts? But then it knows what a P&L looks like.

19:02Isar Meitis It actually knows very well how a P&L looks like. So just the fact that we ask for a P&L, it knows how to build it. Now, if you have a special way you do your P&L, and there's always weird stuff. I've been in multiple companies that do things in different ways. Like as ex- I'll give you a great example from a company I worked at that actually did it in two different ways depending on the kind of revenue we had.

19:23Isar Meitis Some of our marketing expenses went above the operating expenses, and they went and they were a part of the, general profit, and then some of it was operating expenses depending on the kind of marketing activity. And that's very specific to that company that I worked at. That is something that you will have to tell the AI to do.

19:41Isar Meitis Otherwise, it would not know how to do this. But there is a lot of things it knows how to do extremely well. The other thing that I wanted to mention is to connect the dots from what Maxim said in the beginning. It said, "Don't give me the numbers, give me the actual formulas." And again, I don't wanna trouble you with the actual formulas, but they're very sophisticated formulas that I'd actually created here to actually create, this table, which is formulas I don't know how to create.

20:04Isar Meitis And so this is the other thing is just asking to give us the formulas allow us to see exactly what it's doing and to trace back. I will add one more thing that I really like doing when I do these kind of things, like really complex Excel with multiple tabs and tables. In general, one of the hardest things when it comes to tracking AI's work is tracing back what it did, especially when there's multiple inputs and outputs and so on.

20:27Isar Meitis In a text document, it is very easy to tell it, "Okay, give me a side note," telling it where did you bring this information? In an Excel, it's not that straightforward. So what I ask it to do, and I have a standard piece of prompt that I use every single time when I do this complex Excel work, is I ask it to add a comment to the headers of each column.

20:46Isar Meitis So in this particular case, we have January 2026, right? So everything is calculated the same way. It pulls it from the same different places, and I'm asking to add a comment explaining where did it pull the information and how does it get to the outcomes that is in this column. And it's gonna write it.

21:00Isar Meitis In Excel, it puts this little red triangle on the top corner of the cell, and then you can put your mouse over it, and it will show you, what's in the comment. And so if it explains to you exactly what it's doing, it is very easy for somebody when you wanna check the AI's work because looking at the formula like, "Holy shit, like I don't understand what's going on.

21:19Isar Meitis There's like 25 parameters and all these calculations that I don't know what they do." if it gives you a little comment there, it can explain to you what it's doing. So that's just another small tip. if you're asking a AI to build like a really sophisticated Excel for you, this is an awesome way for monkeys like me to understand what the AI is actually doing.

21:36Maksims Sičs yeah. here we don't have so many sophisticated formulas, so here are pretty straightforward SUMIF formula. But I should say I had a project, of a client that I was doing, and the client asked me to add one more thing to the project, basically like a model that I produced. And I thought about, "Okay, let me do this."

21:53Maksims Sičs And I sit for like 30 minutes, and I was trying to write this formula myself, and apparently I thought about, "Okay, I wasted 30 minutes," which will... should be like, a quick add. And then I was like, "No, let's AI do this for me." So as you can see here, I got a bunch of plugins, so you can pretty much, add a cl- your Claude or Chat to, Excel s- straight away and just use it here, in the Excel interface, so to say, and ask it to produce something for you.

22:18Maksims Sičs So I did And apparently it wrote me a such sophisticated formula. Oh my God, this was like, six lines of just writing like hieroglyphs and stuff. And I wanted like... You know, this was a client project, and I should have a full understanding of what's going on here, and I understand I, don't get it at all.

22:36Maksims Sičs Like, I, I literally have to sit for, more hours to understand what is the formula about. And sometimes, what I asked it about, I ask it to just to simplify. S- Make, just use simple formulas. Don't use, sophisticated, lambdas or new formulas that are added like dynamic arrays. Just let's use very, very basic ones.

22:53Maksims Sičs So it was able... So it said, "Okay. I will add, a couple more columns, so it won't be just, one clean formula, but then the formula will be just, one line or a couple lines that you can read and understand." And this was exactly what I needed because sometimes it can produce very, I say, very sophisticated formulas that even for pros would be hard to understand, and you would need to sit through and kind of read it.

23:13Maksims Sičs Yeah. So just ask it to simplify sometimes. that also helps. Awesome.

23:16Isar Meitis Yeah, great tip

23:18Maksims Sičs So let's take a look at the core, right? If balance sheet matches, then everything is fine, but there might be a problem in real life. Well, probably not with AI anymore. Like, you know, a few years back, there might have been issues when balance was not balancing, but really not anymore. So here, as we are looking at the balance sheet, the core numbers are really like total assets should be equal to total equity plus liability.

23:41Maksims Sičs And as we can see here, all the numbers are matching, and it means those are, those are exactly correct. So you might actually see how, or you might say how total assets can be negative, at all, right? And this is because we didn't have any opening balances, right? It just took the turnover and just put the turnover, assuming that the opening balance is zero, right?

24:03Maksims Sičs So that's kind of another assumption it has made. As you can see, there are no numbers, and just because, this month apparently it had, from the bank account, had more payments than it needed, it turns negative. And this is a no mistake from AI perspective because, a best financial analyst would probably do the same or just, ask you a question, "Hey, we don't have opening balance.

24:20Maksims Sičs What should we assume? Otherwise, assets are negative," or something like that.

24:23Isar Meitis Wh- which by the way, the AI did. It said, "Hey, I don't have any opening balances, just so that So we could have- Yeah when it gave us that comment say, "Oh, the opening balance is X," and then it would have continued from there.

24:33Maksims Sičs Exactly. and the interesting part, uh, I'm not sure. No, it's actually linked to the general ledger, I can't here update the opening balance because it still refers back the... It has, the opening balance, here. So if I put here, for example, the bank account, let's say we had, let's say 20,000. Uh, it's 200.

24:53Maksims Sičs Let's say 20,000, here. It automatically readjusts. You know- Yeah you see the formalized dynamic.

24:59Isar Meitis Because we built the formulas correctly. Yeah.

25:00Maksims Sičs Yeah, so it places the bank account, and now every other month, basically it readjusts given that we have this bank account plus sign. Once again, we go to negative at month three, right? that's again, that's not okay. It means the company doesn't have enough cash to pay, to pay. But anyways, and actually it shows now an error, that we have n- we have no balance, and the reason is because we added the assets 20,000, but we didn't add the liability side 20,000.

25:26Maksims Sičs Liability, yeah. So let me quickly fix that. Let's assume we've got the share capital 20,000, okay? And let's see if this balances out. okay, so let me see.

25:38Isar Meitis Yeah, I think this is less critical- Oh, it's negative because we're not trying to- It's negative. Oh, yeah. We're not trying to figure out- It's negative the accounting side of it. But the nice thing here- Yeah by the way, that you do see is that it immediately shows you there's an error. So you could have made that mistake saying, "Oh, let's give it the opening balance because that's what it asked for," and then not know that you actually have to add another parameter for it to actually be a clean balance sheet.

25:58Isar Meitis But it built, on its own, a line that is either green or red per month that is showing you- Yeah whether this thing is working correctly or... it's working correctly, whether the balance sheet- Yeah, it's working, is okay or not. Yeah.

26:11Maksims Sičs If we put it with the right sign so it's working. yeah. So 2020, now it's okay. Yeah. It even added those checks, which I really like. probably too many checks, right? We could have, gotten away with just one. But anyways, Yeah this is a funny thing. it actually gave me a little bit more other things, like for example, open items, that I probably...

26:28Maksims Sičs that I actually didn't ask for. So those are trade, account payabl- account receivables, basically our invoices to clients that were not paid yet, so outstanding. That's pretty cool. It's, so detailed, like even have the clients and the, invoice number. Everything is there. And some, account payables as well.

26:45Maksims Sičs Okay, so that's pretty cool. And of course, some review notes. again, the opening balance, there were clarifications about the VAT and stuff. but let's, let's leave this as is. and let's assume we're satisfied what we got, and let's assume we don't have any issues with, with the VAT calculation that was actually inside, which it told us to check.

27:04Maksims Sičs What we actually gonna do is we're gonna say... We're gonna go back to this chat, and now what happens is a new, So basically, a new month has arrived, so now we have July 2026. Okay, it's basically the same. It has the trial balance, but let's delete it because in the previous one we didn't have it.

27:21Maksims Sičs So we have another month. We just need to add another month. And it will be just as simple as, let's just drag it to the same folder. So now we've got the July 2026. Let me just copy/paste this one, and let me tell it the following. So I'll paste the number, here, and I will just add the following prompt.

27:41Maksims Sičs And let me just start it and read it for you because while I reading, it's working. So I added it to the s- the same folder, so essentially now it will go and check the folder again, and will find that file, and there is the prompt. So here's July. Before I close it, check it. Compare it against the six months you just worked with and against the source documents in the invoice folder.

27:58Maksims Sičs I want to know about anything that doesn't reconcile, VAT duplicates, amounts that don't match the paperworks, number that break the pattern of the previous months. So- it says it can't open invoices because I didn't give access to another folder that I was planning to do next time, which would be, taking the actual invoices and checking if the information that was actually placed in the general ledger was correctly and according to those invoices.

28:26Maksims Sičs Because basically, when we're talking about accounting, now the most sophisticated accounting companies, they will actually have AI read the invoices and place the numbers into the accounting system for you. So you don't even do that manually because once we do that manually, there is a human factor error that a human is tired and it will just mistype a number or something.

28:48Maksims Sičs Doesn't happen to AI. so in s-

28:51Isar Meitis in our case what we're gonna- Two, two things, two things about what you said that I think are important. Yeah. one is connected to what I said before. AI is really good at taking unstructured data, again, invoices are unstructured data, and understanding how to structure it, right? So yes, invoices are quote-unquote structured data because it has tables and numbers, but every one of them looks differently.

29:10Isar Meitis Like depending on the company that sends it, it's gonna look differently on how it's built and what it's written, and some are going to say shipping, some are going to say freight, some are going to say, whatever, delivery, fees. Yeah. Like everyone will have different names for different things and different structures and so on, and AI is very good at this.

29:25Isar Meitis And so you can apply this to many other aspects of your business because once you understand AI is good at taking unstructured data and structuring it, the same thing goes with analyzing customer calls. The same thing goes for analyzing documents that people have written, emails that people have... every one of those things is unstructured data that AI can structure and then deliver it in whatever format you want.

29:48Isar Meitis So it doesn't have to be a, general ledger to a P&L. This could be Phone calls analysis to recommendations to salespeople how to sell the product better. it's the same exact concept that we're applying here applied in a different field of the business. the other thing is a shameless plug that those of you who's been following, this show for a while, would know I actually have a software company called Data Breeze, and what Data Breeze does is exactly what Maxim says.

30:12Isar Meitis It knows how to take invoices, regardless what they are. It knows your accounting rules, your vouching reconciliation rules that are written in simple English because it's an agentic solutions, and the agents understand your rules and your SOPs, and they know how to enter and reconcile all your invoices straight into your accounting system.

30:31Isar Meitis So first of all, thank you for raising that topic because I wouldn't have thought of that. but you can do really amazing things when it comes to building... and, once you get to real life, when there's hundreds of difference of types of invoices and there's a gazillion, it's not just a simple prompt.

30:44Isar Meitis You need to actually build a serious infrastructure around it, but it is doable, and it was not doable two years ago.

30:51Maksims Sičs Oh, yeah. yeah, sure. So let's check the result. we see that it has finished, and it says the ledger itself balances, but there are four things that are actually wrong in it. So the loan interest is 12 times too high. July interest is books, books at 2,500. January to June it fell by four euros a month, so July should be 260, okay?

31:12Maksims Sičs So it probably entered the yearly number in just one month rather than just monthly number. There is an invoice that it was booked twice, and that could really happen sometimes. there is the wrong VAT number. so VAT was, booked, 261, against 118. And it says entry numbers restart at one. July uses numbers 142 instead of carrying on from June's 263.

31:38Maksims Sičs Okay. not sure what was, what was this. Together, items one and two underestimate profit by 2,600. And there are some other things to check out. So it actually took the newer file that is in the folder, and it actually, checked the file. It didn't place the file into our report yet because we didn't ask it to do so, but it could be easily added, and it could correct mistakes.

32:02Maksims Sičs Another thing is we can also add another folder. and for example now I can actually add this invoice folder that was right in here. So let me quickly do this. there is another folder, and I will ask, can you check if the invoices are If the invoices in the folder are correctly entered in the general ledger.

32:37Maksims Sičs Okay? So now it asks again to give another access to that folder. So I'm basically adding a bunch of folders where I store files about... this is the demo company, but I actually do the same for my company essentially. I would just give it data, I would give it invoices, I would ask it to recheck it.

32:53Maksims Sičs And actually, something that I was, really shocked, when I had this wow moment when my accountant actually sent me annual report for signature, and it actually sent me a bunch of files, not just annual. There is annual report, there is the general ledger, there is another file which is the detailed general ledger.

33:08Maksims Sičs Basically, it has all the transactions for the year, under each account. it sent me a trial balance, and there was, some other things, you know, inventory, document that I has to sign, and some other things. And basically, what happened is I was like, "I'm not gonna read the whole thing. I don't have time for that."

33:24Maksims Sičs So I was like, I already... To be honest, I already signed the documents. But before I did send it to the accountant for submission to the state re- register- state revenue service, I thought about, "Let's just quickly give it to Claude for checks." I gave it to Claude, and it actually, worked for about 10 minutes, and it told me, "Hey, did you actually notice your balance sheet doesn't balance?"

33:44Maksims Sičs I was like, I was looking at the numbers and this is the first thing that I checked, and I was like, the, the numbers really look the same. It's just, you got a, a total number. one, two, three, four, five, six, seven. And in the liabilities plus equity side, there was one, two, three, five, six, seven.

34:00Maksims Sičs So instead of four, there was five. And, it's just a small typo that I didn't notice and, state revenue service would reject that. But I, I even checked that myself, but I didn't notice because it's just such a small thing. So it told me, that's the priority number one. You have to fix that." And it actually told me why it didn't balance, which was a wow moment for me.

34:20Maksims Sičs Because usually to find this problem, you would literally have to go through, all your accounts and registers and check the transactions. But it actually, since it had all the detailed general ledger, it actually told me, "Hey, your, account balance on the bank is overstated, by €100 or something, and this is probably a typo."

34:40Maksims Sičs It actually told me this is probably a typo because I checked the general ledger numbers and they are correct. It's just in your annual report, probably accounting was, accountant was writing with, with hands, and he just made this typo. it literally told me exactly the place and I was like, wow, that would take a lot of time to find what was wrong.

34:58Maksims Sičs and this was a wow moment for me. So- And actually, w- another thing it found that, on page number four, my registration number, my company ID number was also wrong because it's a long, like 12, 12 letters code, so to say, numbers code, and it just mistyped one number. So again, it's like human factor.

35:14Maksims Sičs I would never have checked that. I would never notice that, but like a small thing.

35:19Isar Meitis No, I think these are amazing. I will say two things that, that relate to this. We always say check the AI's work, but it is as important to allow the AI check the human work, which is a- Oh, yeah great tip in this particular case, especially when you have so many details and so many things that are done by hand, which leads to my second tip.

35:39Isar Meitis Today, you don't need to do anything by hand anymore because all these tools know how to connect to APIs. They have MCPs. They have all these kind of solutions. So my Claude environment is connected to my accounting system, and it sees all the actual live data. I don't need to copy. I don't need to download.

35:54Isar Meitis I don't need to give it access to anything, and I can ask it any question about anything that's related to my finances or my company, and it can give me the answer in whatever format I want. Show me a dashboard of all my open invoices from the past three months, that is more than $2,000 and compare it to blah, blah, blah, blah, blah, and it will...

36:12Isar Meitis Like what- whatever format you want, it knows how to do because it knows how to, A, pull the data from the accounting system, and B, create whatever displays and output you want. And I will add one more thing that relates to what I just said because it is very important. We are very much living and analyzing our businesses, not just accounting, through an existing lens.

36:38Isar Meitis There is a standard format on how we deliver specific kinds of data. There is a report you write to your boss once a week. There's a PowerPoint you deliver to your board once a quarter. There is... whi-which means the data and the deliverable are the same thing, right? The data and the deliverable, it's a, it's here's a report, here's a PowerPoint, here's a P&L, right?

37:02Isar Meitis It's the same thing. The reality is, in today's world, we have a lot more flexibility. You should save the data as the data. Here are all the invoices, here are the general ledgers, here is our marketing, sales, whatever the information is, and then you generate the deliverable based on the need you have at that moment.

37:21Isar Meitis Which means the deliverable could be a PowerPoint, it could be a dashboard, it could be an Excel file, it could be a Word document, it could be a mix of the above. it could be an HTML page that is fully interactive, where you can move sliders and show stuff. a video. you can ask AI for whatever deliverable you want, and this gives us a lot more flexibility to generate the right output for the right scenario for the right people at the right time versus, "Oh, here's an Excel file.

37:45Isar Meitis Work with it." and this is another huge value that did not exist until a year ago, maybe even six months ago, the flexibility to generate really high-end deliverables on the fly for anything you need

37:58Maksims Sičs yeah. It just allows you to even, test and prototype and even build smaller apps. My other friend works in e-commerce business and basically on their internal meetings, he would just pull up a Claude and while people are talking, he would just build a quick prototype to test, during the call So how crazy is that?

38:16Maksims Sičs This would, take, weeks before and now it's just, a few minutes away. It just, and the thing is that the guy I was talking about, he's not IT guy. He's just regular salesperson- Yeah that is using Claude. That's it.

38:28Isar Meitis Yep.

38:30Maksims Sičs Okay,

38:30Isar Meitis So- No, go ahead.

38:32Maksims Sičs Yeah, let's just quickly, get back to this one. So we asked it to check if the invoices in the folder are correctly entered. It says okay. Actually, there are three problems that with the invoices and for some of them, once again, the invoice was dated and this amount posted twice, wrong interest, wrong VAT, right?

38:52Maksims Sičs So again, pretty cool that, we can check it manual because sometimes people would just make those mistakes, and I think we've seen that before. But, I wanted to show you one more demo if we have time, and this demo that helps me actually to gather monthly invoices because this is something I had to do before myself, and probably this is more for CEOs or people, working in the business that would supply the invoices to accounting.

39:17Maksims Sičs But this is something that I needed to do to my accountant as well. And for this one, I'm actually using, I'm actually using Claude Code, but you pretty much, you can use the regular Claude as well. I just have it saved here. And I have one skill that says monthly invoices. I'm gonna get back to this one.

39:37Maksims Sičs I'm gonna shoot it out, and I will show you what it does. But just before I do so, I need to open a folder where I want it to save my accounting invoices. And for this one, let me quickly pull up the folder, because basically, the way it works for me so for example, the way I structure it, I use OneDrive for this one.

39:58Maksims Sičs I will have it by month, and then I have every single month with the folder with the invoices. So now I'm gonna just use the September 2 as the sample folder, and I would ask it to, so save, all. So now we have new invoices in this folder. Actually, another thing, it will actually need to do, but it also already has access to, is my account bank statement.

40:25Maksims Sičs So the bank statement will have just a bunch of transactions and, I think everyone knows how it looks, so I'm not gonna show how it looks. but it has access to a specific folder where my bank statements are saved. So I'm giving it a bank statement. I give... I just pull this slash command. I give it a folder to save it, and then basically what it does now you guys, are going to see.

40:45Maksims Sičs It will, because some of the invoices are sent via email. Some of the invoices are accessible only through different apps, and you actually have to log in to access a specific invoice. some of the invoices are available on company platforms. So it's just, manual work. You have to go and log into different services to, to gather it, which I don't like to do

41:07Isar Meitis So, by the way, those of you who are not watching the screen, as soon as he ran the command, because it's a scale that has built in multiple steps and access and so on, it is opening a checklist of things that it needs to do in order to complete the process, and then it's just working through the plan, just going step by step, fetching information, aligning it, comparing it, dragging it, bringing it, and it's just doing the steps that, Maxim would have done manually, only fully automated.

41:33Maksims Sičs So if you see my screen now, for those who will see the screen, it's actually, this is my mouse. This is my mouse. And you... I'm not sure if you see there is this, orange mouse. Other

41:41Isar Meitis mouse.

41:42Maksims Sičs I'm not-

41:42Isar Meitis Yeah.

41:43Maksims Sičs So I'm not actually using my screen. It actually... Now it... Those are real invoices. This is my inbox, and it actually goes through each of the emails, and it opens the emails. It downloads the invoice because this was the first step. Some of the invoices are just sent to my inbox. Okay? So now it's, just goes...

42:01Maksims Sičs And as you can see right in here, it downloads the invoices. So if I open, so now it has downloaded, three invoices. Here we get another one. Now we'll have the invoice number four. so yeah, I actually... what I would do in this case, I would just shoot out the command. And by the way, I can still work in the other tabs, right?

42:18Maksims Sičs I can use, Google. I can go to other tabs. it's all working, because this guy is actually it opened a specific folder right in here and, Sorry, not folder, but the browser link or, tab, and it works in this particular tab. And it doesn't mean, you're paralyzed, you can't really work. I can work on my computer.

42:35Maksims Sičs I can do whatever I want to do. It just works in its own browser. It gathers the invoices. It actually saved the invoices in my computer, so it's not, As you can see now, this file already has a few files. So it has a few invoices that I've issued myself and a few invoices that it just saved. But what I also liked, from this guy is that it renames them Because this part, that was never done by me.

42:57Maksims Sičs For example, here you would see the names of those invoices, 10 April. What this invoice stands for? Like, how do I know what this invoice was for? How do accountant then quickly finds those invoices? This is so sophisticated. So instead of just this weird name, it will, put this name as the, supplier that supplies me something.

43:18Maksims Sičs Then it puts the name, the invoice number and the date of the invoice, which is very, useful for accountant because it's just pure clear structure. Otherwise, just a whole mess, right? And now it gathered, like, how many, I don't know, 10, 11 invoices for me, and then it goes to another app. So now this is a specific link, where you have to download the invoice.

43:38Maksims Sičs It's actually in Latvian here because it's, for my company right now. So it downloads that. Again, weird naming, so we... it has to rename that as well, and it just, just working. just working. another thing is pretty cool. it changes my accounts because I've got a bunch of different emails And I receive invoices to different emails, so it switches between different accounts for me.

44:01Maksims Sičs So now it's in a different account. So yeah, this is, This's been, like, a huge work relief for me because, in terms of timing, it's probably nothing very specific. It would take, 30 minutes or o- up to one hour to gather it, but you have to do it, every month, and that's, pure manual work. You don't need, brain to do that.

44:17Maksims Sičs so yeah. And,

44:18Isar Meitis No, but I think, I think the... There, there are a few very important takes from this. One is that once you figure out a process that you don't wanna do anymore manually, you have to teach the AI once how to do this, which Maxim had to do. what-- The magic you're seeing right now of the AI opening different Google accounts, going through them, finding ones that have invoices, opening the invoices, seeing who they came from, so they can save it with the right name, with the right supplier, with the right date, with, like, all these kind of things, the number of the invoice, is it because Maxim showed it once what the process is and explained what's coming and where it's coming and where to save

44:54Isar Meitis it, and so on. But once you do any one of these processes, and again, it doesn't have to be accounting, any other process. Once you teach it once how to do it, you literally ask it with simple English, say, "Now turn this into a scale I can reuse," and it will knows how to reuse it. The other thing that you can do is now that Maxim has it, let's say it's something you wanna do once a week or once a month, you can create a scheduled task and say, "Run this thing on the first of each month for the previous month," and then we'll just go ahead and do this.

45:21Isar Meitis And you can string them together. So you can just think about the two things we saw. The first thing we saw is after we built the general ledger and so on, but you can build the general ledger from the invoices. You can say, okay, step one, I have a scale that gets all my invoices from these 17 different places.

45:36Isar Meitis Now I have them in a folder. Step two, go read the invoices that are now named correctly from the folder, create a general ledger. Step three, here's another scale that knows how to take the general ledger and turn this into a P&L and a balance sheet and all these kind of things, and your entire accounting...

45:50Isar Meitis And step four is give me all the comments and all the things I need to check and pay attention to because they're not correct. and that's it, and now you've built an entire automated accounting process that will take you probably a few hours, maybe a couple of days to build and check and verify that it's actually working correctly.

46:07Isar Meitis But once it does, it runs on autopilot every single month

46:13Maksims Sičs Yeah, so exactly. and it's probably worth mentioning, it's actually in my Microsoft account, downloading invoices. It actually wasn't able to, connect to some service, so it asked me to log in. So that's sometimes that you would need to do still because, sometimes those services will log you out, and you will need to log in again.

46:30Maksims Sičs but it's worth mentioning what, is a skill. A skill is basically an SOP, right? It's standard operating procedure that would, you just describe how it works. And in this case, in order for it to know where to search for invoices... Actually, now I've got more invoices. I will show you how this actually looks.

46:48Maksims Sičs So let me go back, in here. So the thing that it needs to run this, it needs to know what kind of invoices and where to get them from. So basically, I've once created this file. Here's an example. So we... I've got a supplier. I'm sorry, it doesn't allow me. Yeah. So we've got a supplier, for example, here.

47:10Maksims Sičs It's actually not invoice. it's called invoice, but basically the supplier. And AI, all, every time it runs this, it, I will actually read this file because it says, okay, if I got this supplier, you will get this from my email. How would you get it? You get an email connector or MCP, right? from Mobilly, for example, you would get this from this email connector, MCP or Chrome cl- plugin if not connected.

47:31Maksims Sičs Then you have, for example, all my invoices that I have issued. So this invoice is actually located in my local folder. You don't need to search anywhere. If I, if you go to Google, the Google will not send an invoice. You'll have to go to this link, right? And it explains how to do this every single step.

47:47Maksims Sičs So next time, again, if there is a new, supplier comes in, it might struggle, but then I will just set it here and for future, especially for those suppliers that I have on a monthly basis, it will be there. So there is no problem. It knows what to actually use. And, you don't need to be like technical guy to know email connector and MCP.

48:04Maksims Sičs It's basically availability to connect. You can go to Claude, you can go to connectors, like customize, and, connectors. Connectors. And here you will just find, Gmail, for example, okay? And then- Yeah basically what you do, you just click Connect. That's it. You don't need to be a technical guy to do that, right?

48:21Maksims Sičs So you just log in, and once you log in, it will be able to read your Gmail. And this is what I actually did myself, right? So when you see- Yeah there is an email connector, that's pretty much, just logged in and connected my Gmail to AI

48:34Isar Meitis Fantastic. I will add one small last thing, and then I will let people-- let you tell people where they can find you. The one last thing that I will add, and it actually goes back to the very first thing we did, like the Excel work, where it's purely working within Excel. Ex- having AI work in Excel, definitely doable.

48:52Isar Meitis You just saw a great example how it created a P&L and balance sheet and so on. But there are two problems with-- three problems. problem number one, it's gonna cost you tokens every single time. So you're paying either through an API or paying for whatever, level of, license you have. You will have X number of tokens that you can consume, and running all the Excel work depends on how big the Excel is.

49:15Isar Meitis You give it more rows, more columns, it's gonna take it more time to go through the whole Excel. So that's problem number one. Problem number two is it takes a while to run. Sometimes it will take five, 10, 20 minutes. And again, it's not a big deal. You can do your emails or have a call or have a coffee at that time, but it still takes time.

49:33Isar Meitis Problem number three is if you're not alone in the company, anybody you give this to, so a set of skills that can do Excel for you, will have to have Claude or ChatGPT or Groq or Gemini or whatever, whatever tool you're using. The other option is once you figured it out, once the Excel work is very clear and you know how to go from this format to that format.

49:52Isar Meitis So in our example that we just did today is from a general ledger to a P&L and a, a balance sheet or a cash flow, report, you can ask it to create a Excel script for you. So Excel has its own internal scripting tool that can do the same exact process. Now, knowing how to script in Excel, you need to be an Excel ninja and participate in Maxim's, Olympics or, you can ask a- AI to write the script for you, and then it writes a script that runs inside of Excel.

50:23Isar Meitis The one last benefit over the regular writing it in Exce-- in, in AI itself, so yes, it's not gonna cost any tokens once the script is ready. It is-- It runs in literally instantly, like it-- you click the button and the other thing appears, and you can give it to anybody who has Excel that doesn't have AI.

50:39Isar Meitis But the last component is maybe the most important out of all of them, and it goes back to what Maxim said in the beginning. AI is not a deterministic engine It is extremely good. It makes less and less mistakes as the models gets better, but it still does make mistakes. A script runs a script. It will do exactly the same thing 10,000 times out of 10,000 times.

51:01Isar Meitis Excel will-- sorry, AI will get it wrong, let's say, one out of 100, because it will do something wrong in the Python code that it's running that particular time. so knowing how to take your work, and again, this is only relevant when you're working only in Excel, like you're taking one format of Excel and converting into something else, you can actually ask the AI to create a script for you, ask Claude to explain to you how to implement the script in Excel.

51:21Isar Meitis It will take you three minutes, and then you solve all these problems that I mentioned before. Maxim, this was- Yep an incredible conversation. We covered a lot of really helpful stuff. if people wanna work with you, follow you, learn from you, what are the best ways to connect with you?

51:36Maksims Sičs I do have a YouTube channel, so you can, go with Maxim's S-I-C-S, it's actually Sics, in, in local- but it's written like six, so so you can find me out here or my LinkedIn, that's the same name, basically Maxim Sics. You can find me over here. Or I do have my company website, which is flowbyte.ai. So there is an option to get acquainted what I'm doing, what's my company doing, and if you want to get in touch, let me know.

52:06Isar Meitis Awesome. Thank you so much. This was really valuable, really helpful, and very well explained.

52:13Maksims Sičs Thank you, Isar. Thank you so much for having me.