Excel Hour - Join me as I share examples and my thoughts on Excel Agent

Foreign. Thank you, everybody. If you're out there, let me know where you're coming from. Sorry we're getting started a little late. Was having some technical issues. I think I sent out the wrong thing in the link as well. So it's been nothing but problems today. I had to send a second email. I apologize about that. And with that theme, I am having some challenges with Excel's agent. So we're going to try reinstalling it here and see if we can get it to work. If not, I do have a workaround to show some examples, so either way we'll be able to show some stuff. But let me do one other thing. Going to blur my background here so you don't see everything behind me. All right, we are now good to go today. You get my. I think you can see the chair in the background, but if you're out there, you can hear me. I do see a few of you out there. Please let me know where you're coming from. And again, apologize for the late start. It's been one of those technical days where everything's kind of gone wrong for me is the way I would explain it. So if few of you that just accidentally entered the stage, I gave you the law. Wrong link. Jo, Jody, what you're going to want to do is go to the streamyard link I put right here, this link and you can watch it there or LinkedIn. I accidentally tried to add you to the room by mistake, so you should be able to see that in the comments. I've also put it in the private chat, so you'll just want to join that way. So again, I apologize. By mistake. I tried to add a bunch of people to the. To the room. All righty. Oh, not my day. You'd think I knew what I'm doing. All right, we got Alexi from Jersey City, we got Justin from Florida. If others are out there, let me know where you're coming from and we'll get started here. Again, like I said, I apologize. Totally my mistake. I had. I've had one of those days where I just screwed a few things up. All right, I'm going to share my screen here, so I'd love to know, has anyone tried Agent mode in Excel yet? Has anyone had an opportunity to play with it? If you have, let me know in the chat. Sean, where can you get next FP A guy Cap? I'll see what I can do. At the moment, I. I don't have extras. I'll hopefully order some more in the future. I gave a Bunch of them away last year at a conference. But if you're going to be in Vegas, Sean, let me know and we'll. I'll see what I can do. If you're not going to be, let me know. But. All right. Thank you, Liz. I'm glad I'm not the only one who has those text tech days. I see Alexi says, you know, he hasn't tried it. So before we even touch Agent, I want to give some of my thoughts, some of my learnings I have now created with Inshore and Giles mail three videos. We've probably together spent 10 hours testing various agents inside of Excel. As we were testing them, Microsoft released its Excel AI agent. Hopefully we will be able to test it. If not, we'll try something else. Mine was broken yesterday and it's broke again today. I'm going to try uninstalling it and reinstalling it. Should only take a minute or two. Already started that and see if we can get it to work because it doesn't seem to be a time thing. But before we do that, let me just talk to agents. So basically, if you don't know who I am, I'm Paul Barnhurst. This is a little bit about me. I love podcasts. My podcast is now FPA Unlocked instead of FP and A Tomorrow, but very similar that to before. I have Future Finance about AI and technology, Financial Modelers Corner. I'm a Microsoft MVP. I'm married, been married for 14 years. Don't tell my wife. I almost got it wrong. And I have a daughter that's now 12, so the picture is a little outdated. I'm in Salt Lake. I love hiking, I love running, podcasting, reading outdoors. That's a little bit about me. What we're going to talk about today is the Excel AI agent. We're going to talk about some sample use cases, when not to use it, when should you use it, how should you think about it? There are a ton of hype videos out there and you know people like, well, this just kills Excel. You don't need to know anything. I'm here to tell you, pump the brakes on the idea that you still don't need to know how to model or you don't need to know Excel. If anyone's seen my posts over the last few days, I've really been pushing that, that you need to know Excel. It's not, it's not enough. And I'll explain why here in a minute. So let's first talk about what Agent is. Agent is basically they Brought the LLM where you could do natural language prompts in Excel. It was released September 29, 2025, basically correlating with the 40th birthday of Excel, which I believe was the 27th. Last week, Excel turned 40. It's currently in beta, and based on what a lot of people have said, it's definitely in beta. It works sometimes it doesn't work others. I know one guy who's tried multiple times and he's never been able to get it to work, and I still see a few joining the studio. If you're joining the studio, please go ahead and join the link in the private chat so that you're in the. In the webinar. That was a mistake. I didn't mean to add you to the stage. So, Hector, Andy, Katherine, Katherine, if you can all join the link in the private chat there, you should see it. If you can click on that and join through the webinar, that'd be great. I accidentally gave you the ability to come on stage with me, but it won't even let me bring you on stage. So that was a mistake. So thank you. I see you all dropping now. If you can join through that link, that would be ideal. All right. You have to have a Copilot license to use agent. Not the free version of Copilot, the paid version. So again, a little bit of a drawback, but I think what you'll see today is if you use Excel, a lot of could very well be worth it. Last but not least, it only works in the cloud. Sean, I know you're a hardcore modeler in the chat. Let me know the last time you build a model in cloud. I'd love to know, for any of you that are big modelers, when was the last time you built a model using the cloud? We'll see what they come back with. I'll give them a minute here because I know sometimes there's a little bit of a lag, so it may take a second. Alexi, Never. Yeah, same here. Not surprised. So that's the. The draw. That's a drawback to Agent. We'll talk more about that. Here's my I called it my hot take on LinkedIn. I will argue that 90% of accounting and finance professionals will get more benefit out of taking a course, whether it be general Excel or modeling and practice modeling, than spending hours trying to build a model with Gen AI. Other things, yes, you can do with Gen AI, but full models and I'm going to talk to models for a minute just because I want People to realize what they're working with. Not to say you can't save hours. There is no question Excel Agent will change the way we work and we can save hours. I am not disputing that. It is amazing. Not going to dispute that either. But I also saw something that said gen AI in Excel never desktop builds only. Yep, that's what I expected. Sean. The error rate. And feel free to take a guess. Anyone have any idea what the error rate is inside Excel right now with these agents that some early data shows a 43 error rate that's 57 accuracy. Would any of us give that to our boss Justin on Excel Agent? No, we'll try it today, but unless there's something I'm not aware of, it's still not working in desktop. But we'll test it here in a minute. Justin. Right. So that's the first thing I'm going to say. Why most people should not build a model using Excel Agent. Then let's talk about where it can be used. One generative AI is probabilistic. Genai is using probabilities to answer questions, which means that the answer can change the next time you ask for the same problem, meaning you get a totally different formula in Excel. It also means it can be right one, time, wrong the next. I've seen that. Okay, that's a concern. Second, hallucinations. We're all aware of those next validating and auditing the model. How many of you like to inherit somebody's model with no instructions and try to audit it? If you say you do, you're a liar or a masochist. It's one of the two. Nobody enjoys doing that. There's a reason why we call the broken spreadsheet the new employees problem. Because nobody wants to deal with somebody else's spreadsheet. And then complexity. The more complex the model, the harder it will be to get AI to provide a usable answer. That being said, despite all that, I've seen incredible results and I'm going to let Bill know you're going to want to join the link in the private chat, Bill, because I accidentally added you to the stage, which you don't want to be. So if you click on that link right there, that will bring you into the webinar. I accidentally brought you to the backstage. I sent the wrong link. So if you can do that, that would be great. Just exit here and go click on that link and rejoin us. All right. But he did say he loves to inherit models with no instructions, so he is a masochist. All right, now let's talk about when it comes to modeling. Let's talk about what the use cases are for Excel agent and Emmanuel. Sorry, people, I'm just gonna do this. Hopefully, last time you click in private chat, there's a link. You're going to want to click on that to join the webinar. I actually accidentally brought you in to the studio. You don't want to join the studio. You want to join the link right here. You can even see it on stage. I'll throw it up one more time for anybody. Please join this link here or join LinkedIn. But if you're in the studio, that's not where you want to be. All right, hopefully that's the last time I give that announcement. All right, back to this. So where I think when it comes to modeling, now when it comes to lighter tasks, writing formulas, building some very specific templates where you know what you're doing, it can be different. But I think it's great for giving you ideas around good structure. I think it's great for formatting, auditing and validating the model. Building your table of contents for the sheets, building your instructions. I've heard some people say, hey, I have it analyzed my. And it builds my instructions. Like, tell me the color coding I'm using. Tell me what sheets are here. Provide, you know, basic overviews of each sheets. Nobody likes documentation. How awesome is that? If I can pull up a model and have AI document it for me, do I have to tweak it? Probably, but it's a lot better than doing it myself. Writing formulas. Fabulous. Preparing for scoping meetings with your client. These are some of the use cases I think it's really good at. And I will show examples where it builds stuff because I want you to see what it builds, but I will also show where some of my concerns are. So before I actually jump into agent, I want to show one other thing because I think it's really interesting. So I'm going to go to a browser here for a minute and I'm actually going to show this in Google Sheets. So nobody, nobody tell Microsoft on me that I'm using Google Sheets for this. All right, So I did some testing last week, and I think this really highlights the point I'm trying to get at of why it's so important you understand Excel and why you proceed with caution when using the agent. Don't buy into all the hype out there. And there is a lot of it. So what I did, I used two different tools. I use one called Rosie AI and then I use Microsoft's Excel Agent that uses the copilot model. One uses the ChatGPT model. And what I did is I solved an Excel esports case. So if anyone's ever done an esports case, they'll know you do various levels. And then what I did is I did, I solved it twice and I recorded a bunch of information and I've done it with different tools and it can vary quite a bit between tools. Eventually I'll do it with more of them. But here's what I found. Level one, it solved the same way with both of them. Level two, it took a different approach. You can see these are complex formulas, right? Most of us would probably not be doing an array constant or let's using double negatives, a lot of those type of things. Here you can see one did a very simple approach solving the exact same thing. Both of these gave the same answer. Would any of us write that big huge long column? Probably not. How would you like it if in you added an extra year to a model and all of a sudden it used a totally different formula and you had to go figure that out? And so you can see used very different approaches. On top of that, I ran each of them twice and half the time, just two minutes later, it used a completely different way to solve the same problem. Right. We could all do that in Excel. There are some similarities, but that can be a problem if you're building something and, and auditing and making changes. If it, it's changing your formulas on you when you ask a new question, maybe a formula in the same column or whatever. So I share that more as just kind of a caution and a warning. But here's what's amazing. So I don't have the scores here. I have them hidden. But this case had 1250 points. That was a perfect score. Both these agents, Rosie AI and Excel, scored 1210 out of 1250. There were 12 questions on this Excel Esports case and they each got 11. Right. That's amazing. That. That is really pretty incredible. So now one person said that they thought it was available in the clouds or on the desktop. So we'll go to the desktop and test. I'm going to go to add ins. I deleted it. To re add it, what you have to do is you have to add Excel Labs. So if you have not added it, you can go to Add Ins and click Add. And we're going to add Excel Labs. We'll just take a minute to install it. Should show up here in a second. Let me Try again. All right. I'm not sure why it is not showing up now. I could always try from the the web as well. Let's try adding one more time. Weird. Let's try it from the web. If it doesn't work, I do have another tool I can use Microsoft Add Ins Store for Excel. Let's see. All right, Tells me to do it that way. That's not what I want. Yeah, okay, I know there's. Let's. Why is it not okay, we're going to try it in. We'll try it in the cloud here and see if we can get it to work. If not, we'll do some testing with Rosie AI which is similar, sometimes better, sometimes not. Okay, let's try this one more time. All right, so we're gonna go Excel Cloud now because it only works in the cloud and we'll see if we can add Excel Labs. But that's the first thing you have to have. I had it on here and it was working yesterday and then it stopped working and so I decided to re add it and it doesn't want to add. And this is what I mean by being that it's really fickle. See, it shows up here on my Google. So let me see. I think it's because. Let's see if it will look work here. Okay, so if I click on Excel Excel Labs, you can see you have Advanced Formula environment, you have Python Editor environment. We want the Agent mode. Agent Modes helps you build and edit workbooks automatically. It's part of Microsoft's Frontier program offering early access to AI innovations. It requires a supported Microsoft 365 license. We're going to try open and we're just going to try something. Please create a table with 10 cities. And what it's been doing to me is it's just been hanging up. It stopped working. So we'll give it a minute. What I like to do if it. If nothing happens here is I'm going to try. Are you there? Question. If this doesn't work, we'll do another tool very similar idea. Ah, it's actually working. It's saying, it's reasoning. Let's see what answer we get. It's planning a table creation so it looks like it's trying to answer the first one, not my second one, even though I stopped the first one. So what you'll. First thing you'll notice is it does give some instructions. So see saying hey, planning table creation, structuring and formatting table. We'll let it finish finalizing process Steps. So this is good. I haven't been able to get this to work for two days, so I can actually go through some examples now. We'll let this one finish. Who know it. So there you go. Right. It just created a table of 10 cities. Not impressive. No formulas, no nothing because I was really lazy there. So now let's bring in a file. So what I'm going to do is I'm going to open a file from my desktop. I think this is a good one to show. I'm going to delete this right here. All right, so what I have here is I'm going to walk you through the data, then I'm going to you show you the prompt I used. And this will show you what I mean by it's. It really is amazing. So I have this contracted revenue. How many of you have ever had to build a deferred revenue schedule? I know Sean has because we talked about that on my podcast. I'm sure several others have. Right. Deferred revenue schedules can be a pain. Is okay. You're writing all your formulas and building it all out and creating your waterfall. We're going to try to create one with Excel Agent. So first I just want to show you how powerful it is before we do anything else. So let's take my. I have my prompt. I've done this before. Let me just pull up the prompt I want to use. All right, So I have this prompt here. We're going to go to Excel Agent and I'm going to drop in my prompt. All right, so let me drop my prompt in here. I wish I could make this larger, but I can't. But basically it says I'm going to read it to you using the data on the raw billing data sheet. Built build a deferred revenue schedule that can be used to recognize the revenue by month. The column header should include customer name, product name, contract amount, contract start date, and contract end date. And then have columns for every month from January 1st through December. The row should include every dealership, the amount of revenue to recognize each month, and a check to make sure the amount recognized equals the contract amount over the period. This should be done in a new worksheet, right? It's a pretty simple prompt and let's just see how it does. We will know it's working in a minute if it start. There we go. I know it's working this time. So creating new worksheet schedule. This will probably take. I've done this a few times, probably about four minutes. So I could tell Some Excel jokes. If anyone wants to put some some questions in the comments, I can take them as well. I do know there's a little bit of a lag, so we'll do a couple Excel jokes while this is running. You know, ideally I really like to do where you guys work with me, but given most people probably don't have Copilot and Excel Labs and Beta, this was a really hard one to build where we work on them together. So it's more of the demonstrative type. So thank you for bearing with me on that front. But it's going through, it's creating it all. Let's see what's a good Excel joke. You probably if you've attended my sessions, everybody's heard all my jokes but what's the best column to store your Excel error messages in? Best column to store your Excel error messages in? I should say bestsell. It's a column in a row. Personally, when I'm frustrated with Excel I like to store my error messages in column fu row 2. So I know this is hard to to read but it reasoned for 47 seconds. Here's what it's built so far. So you can see it's saying hey now I'm creating the schedule. So it built the worksheet. One thing I don't like is it put the time. Would any of us when we're doing months put the time? Probably not. And it put it with a Z at the end. I think that's going to be a problem here in a minute, but we'll see. But you can see it used formulas so let's let it keep working. It's verifying its calculations and I've done this multiple times. First time I've got a date like this. Every time it's a little different. So Alexi, yes you could. You could do something like that. You'd probably need to tell it where to add it and then you need to validate it does it right. So Alexi just asked. I'm going to put his comment up here. Well that's finishing building and we'll see how it does. Let's say you have a three statement model with a workforce planning tab. Could you ask the agent to eg add two more developer Myers? I'm not sure. I'm sure that's hires in 326 at X compensation and have it updated. You could. Yes. Alexia probably would be as quick to do it yourself. That's the other thing like we had. We asked it to unhide some column. Unhide Some hidden sheets took it a minute. I can do it in less than a minute myself. There are definitely times that's the other thing people are finding. But keep in mind, this is beta version, right? It's not even version one. It's not even MVP of the product. It's just, hey, go test it so we can learn. It's going to get a lot quicker. It's going to get a lot better. All right, so it's close. It's a. It's exploring formula logic. We'll let it finish, but we can look at it while it's doing that. So what did it get right? We know it did my customer name. I could have said, hey, start in B2 if you wanted some people like, you know, column one to be open, I could also just come in and insert one column to the left if I needed to. Let's see if it insert one row above myself. Shouldn't impact anything. So customer name, product name. That's right. Contract amount. This should be 12. 12 million. Yep. 12.5 million. Has all the start dates. Has all the dates. Has a check at the end. Right now it's still thinking the formulas are not working. The reason for that appeared to me to either be the way the logic it used for the formula, but I think it's this date, so. So we'll see if it's smart enough to figure it out. But you can look at this formula if you guys can see that. It's a little hard to see, but probably not the way we would write it. If and if end, you know, divided by date diff. I've seen sometimes where it does year plus month. Often it will write the formulas in a syntax that's different than the way that we would think of or the common way we would do it. And so see, it's noticing, it picked it up. It says, investigating header issue. It now realizes I did something wrong with my headers haven't prompted it. It's correcting on its own. So let's give it a minute and see if we get there. All right, well, we're waiting it for the get there. Does anyone know why the spreadsheet hated its job? Why the spreadsheet hated its job? Too many functions. All right, it's close. This is the worst part of doing training like this is. I know for some of you this is probably boring to watch. So put your questions in the chat, but I will show some other examples. I think this is one where it's pretty good. Now, if you're really fast In Excel, you could probably build this in about the time it takes, right? The most complex part, all this you can just copy and paste is writing your formula, which is just a dates and an if formula. If you're not experienced, you decide to use this, you're going to have to be able to figure out when it's wrong. So it finished. It didn't correct its dates, it didn't figure out what it did wrong. It reasoned for 256 seconds. So I've done this several times, this type of thing, and you get something different every time. Hence why I'm leery to use Gen AI for modeling. It's probabilistic, it's going off probabilities, not deterministic. So let's go ahead and ask and see if it can fix itself. In row two, please include only the date, not the time. For example, column G2 should have a date of 0101 2026. I should say cell, not column. That should be cell G2. And then cell H2 should have a date of 0201 2026. And let's try that so on and so forth. Hopefully it likes my way of writing it. And so it's going to think for a minute. And so what I'll say while we're doing this is I interviewed Jeff Siebert. He was on the Social Dilemma. He was at one time the head of product for Twitter before it went private and was bought by Elon and became X. He runs a company called Digits which does accounting. It's trying to replace QuickBooks and has done some amazing work. He talked about, he's a big coder. He wrote his first coding program when he was like 12 and he referred, he said with Vibe coding early on it was pretty terrible because after a year or two I got where I could really use it. The other thing he said is the fact I knew how to code made me a much better Vibe coder. That gets back to my point at the beginning. Knowing how to use Excel, knowing how to model is critical. If you know those things, you're willing to check things, you're willing to be patient and know where and when to use it. It's freaking amazing. I mean, just think about when AI first came out. If you could do this, you know, you could also Claude and MCP servers and, and lots of different things. You know, Claude's going to become available in co pilot soon, so you'll have another model to reference. But let's see. It still seems like it's struggling a Little bit. But we'll. We'll let this one finish and then I'll show a couple other examples. While we're waiting for that, I just want to show that it's not as far as I know, I'm going to close my Excel and reopen it. And I believe my agent should. Excel Lab should come up in my desktop since it's showing up on my browser, hopefully here. So let's see. It is not. Let's see. I'm not sure why. There we go. So I got Excel Labs here. We can see Agent mode. Agent mode is currently only supported in Excel for the web. That's what I thought. So I think that answers the question. I think it was Justin who mentioned that. So, Justin, you can see that there it is only available in the web at the moment. I'm sure it won't be long before it's desktop. All right, so I'm going to close that. The other tools like Rosie and others I've mentioned, they are available in desktop. So that is one benefit of a lot of these third party. So let's see if it got it right. So here's my first one, 23,000. So I should see something from 6:1 to 5:31 and see my formula. You can see it starts in 61 and it goes for 24 months, ending in 5 1, the month of 5 of May 2028. And if I come over to my check, I can see it did a check to say hey is the absolute value. And what I like is it actually said less than $0.01, saying we'll allow a small rounding error. Very common. Okay. Or check it. I could ask it to put conditional formatting there and continue to work with it. I could say, hey, instead of an okay, I want a check mark, what, whatever it might be, I can give it those different instructions and it will do that. So there's one example, let's look at a few others. Let's just kind of play around and see a few things. So I'm going to go ahead and open a new file now. Go ahead and close that one. We'll come back here. Go to File open. All right, so Excel agent, let's go ahead and let's do a waterfall graph. Many of the tools I've seen have always struggled this. I haven't tested this with agent yet, so this will be a first for me. Let's see how long it takes to open the file. Come on. I did it. Not open. Open files from this device and I'll see. I'm clicking open, we're uploading this file. Okay. It's because it's. Let me do this. I'm going to. It's opening up from my OneDrive and it's probably taking a minute. So I'm going to do. I'm going to do a workaround here and I'm just going to copy the file from Excel Desktop because it's taking it so long to figure it out. All right, so here, I got this example here. One thing you'll notice with Excel Agent that's different than a lot of the other tools in the market is Excel Agent doesn't analyze the file. It waits for you to ask a question. Many of the others open it up and will analyze your file and say, hey, do you want me to try this? Do you want me to try that? Which sometimes can be annoying, sometimes can be really helpful. So pros and cons, both ways. But let's go ahead and try this. Please create a waterfall chart using the data. I'm going to fix that in cells B14 to B22. I'm gonna guess it has problems. We'll then continue to prompt it. Right. How quickly could one of us build a waterfall chart? Pretty fast. So this is a situation where my guess is, because I've done this a bunch of times, is with different tools. I haven't done it with Excel's agent yet. This is one where you could probably build it yourself manually quicker. So it's finalizing the setup, it's reasoning, implementing chart creation. Some tools will only create it with Python. All right, so pretty good, but not quite there. Right. Can't turn that into a boss. Definitely not. Right. So now let's see if we can get it to do the other things. Please set the 2023 actuals endpoint to total because it should be a total in a waterfall graph. So let's see if it can do that. It's thinking. And again, I know this isn't the most exciting episode because we're live and doing things. It looks like it created it twice. I'm going to delete one. Hopefully that's not the one it's trying to work on, but we'll find out. But even. Even if it doesn't totally get it right, like, I could obviously do this a lot quicker. I could right click here and set as a total. Okay, so I can't do set as doesn't show up. I forget I'm on web on desktop. I could just right click on this one element and go set as total. But that's not an option there. So I'd have to figure out where it is, which, I'll be honest, I have no idea. In cloud, because like many have said, I never use cloud. Anyone know where I can set this as a total? In cloud control, one doesn't work, so that shortcuts out. So these are, these are examples of the real world problems right now. All right, it's still investigating. Now let's go ahead and look at. Let's just look at desktop real quick so you can kind of do a comparison. Let's. Let me pull up my desktop. Well, that's thinking. So we're not just sitting here looking at the screen. If I come here, I got a waterfall graph, right? All right, so if I just come here, go insert recommended chart. It figures out I want a waterfall. I got it. I click this right here, the one element, and I need to set it as total. See how easy that was? All right, I want to get rid of my grid lines. Want to get rid of my label, my whole label if I can, right. If I don't want a chart title, could get rid of that, get rid of that. I have it labeled there and you can see how quickly I can start to do all this manually. In that time, I'm halfway there. In another minute, I could be all the way there, get rid of my connector lines because I don't like them. You know, enlarge my font, change the colors. In that time, this reason for 87 seconds and it still didn't get it right. So there, there's example two to. Again, I'm going to beat home the point where people are sick of me saying it. The best thing you can do right now for modeling or Excel is learn to be good in Excel. Those who know the tools they're using for AI will get more out of AI than those who think they can just write prompts and be done. Now, could you still potentially get there? Yes, if that's the way you want to do it. More power to you at this point. Year from now could be different. As of right now, I would recommend focusing on making sure you really know Excel well, then using AI to make you more efficient, not to build things for you, not to do all your work. That's going to be really dangerous if that's what you choose to do. All right, so there's example two. Let's. Let's pull in another example. That one didn't quite do what we wanted, but, you know, it wasn't terrible. Let me pull up another One here. So give me just one second. Feel free to ask any questions while I bring this up. Okay. And you'll have to forgive me if I'm not talking for a minute. My voice is starting to go. I've been going for about four hours. I did a three hour Excel training before this for a company. All right, I think this is a good example because I expect this one. All right, so there we. There we go. That's our. So now we're going to go here. I'm going to copy this data. 73,000 rows. I am going to copy it. And why is it not. It doesn't want to copy. Let's see if I can get it to copy. It's freezing. It's thinking probably because it's. Yeah, see, so too large of a data set won't even copy from desktop 2. So instead of opening it there, I just want to copy it in. But let's see if I can get a smaller data set to work. Let's give me one second here. But what I've found, and this has been true with pretty much all the tools, is if you have a large data set, your AI is typically going to bomb out on you. Let's try 5,000 rows, see if that will copy. There we go. It likes that. We'll take a minute. For simplicity, let's create a new chat window. And all I'm going to say, I'm going to get rid of sheet 2. Analyze. Analyze the data on sheet 1 and prepare an executive summary that I can share with my manager about current trends with any recommendations to improve gross profit. All right, let's see how it does. I haven't seen any questions come in for a while. If anyone has some, please put them in the chat. And I wouldn't be surprised if this one bombs out with even just 5,000 rows. Again, showing Excel can be great with small data sets. And this one is okay. It is planning so good. It is working with five. When I tried some other tools, I think I had to reduce it to under 10 or 20 to get it to work. So let's just see how it does. All right, let's see what's a good Excel joke or good joke. Does anyone know why a golfer takes two pairs of socks to the golf course? So why do you take two pairs of socks to the golf course? In case you get a hole in one. All right, we'll tell one more while we're waiting. Does anyone know what the Eiffel Tower and ticks have in common? Hey, Bill, I didn't think that one was way too easy. I liked it, but you're probably right. All right. Does anyone know what ticks and the Eiffel Tower has in common? They are both Paris sites. Get it? Parasites. Hahaha. All right. Does anyone know what a data analyst puts in their hair? Some product. And if anyone's interested, as we're watching this work in Schnorr, Giles, Mel and myself have been testing different AI tools. I'll be honest, one of them is already shut down and other ones change direction as Excel releasing Agent will probably kill most of these new tools. Some will survive, there are some that have done very unique things. But Agent is as good as many of these tools out there and you buy a co pilot license, you have agent. So we are doing a series on them, comparing and contrasting. I think many will find it fascinating. I think you'll, you'll hear a really good take from in and Giles, both modeling experts of hey, what does this mean? Yes, this is amazing. How do you need to think about it? How do you need to watch out Tom? And I'll be adding my opinions from time to time as the FPA guy. That's not a, a great three statement modeler like the two of them. All right, so it's determining the day mappings at the moment and so it's working through it all. We'll see what it comes back with. This will probably be the last one. We might have time for one more but hopefully this starts to give you an idea of agent how to think about it. And like I said, it is amazing. We had it build a three statement model yesterday. It was able to get it to balance the first time we had negative cash. We then said hey, build a revolver. Didn't care for the formula it used. It was clunky but it was right and it built a revolver. And so you know, it also put in all the assumptions. We could have got the assumptions to a separate sheet. So I do think, I don't think I know it can build a model. The question is should it build your models? And that's where I say pump the brakes. So Alexi, I think FP&A tools like a left, I think they have some great functionality they built where they're different than say a tool like Rowan that ended up shutting down. That just brought an agent into Excel is a left is built into the Aleph database where all your actuals and your budget and your forecast and stored. These Excel agents right now, they don't have access to that you haven't built that. A left has also integrated it into Slack and teams and different things. So I think the agents in your FPA tool are different than the agents in just Excel. And so here's another. Sean. Thank you, Sean. I, I want to share this while we're waiting for it. It's looked like it's done reasoning, but I don't see anything yet. So we'll take a look at it here in one second. It. You can see it gave us a plan. It reason for what, six minute? Five. Four minutes if I can do math. 240 divide 60. Yeah. Four minutes. I don't see where it gave me anything but we'll, we'll go through this here in one second, he says. I've been on a trial of X21 and spoke to the founder Yan. It's quite like Rosie. Got to say that my views are the same as yours, Paul. There are some use cases that let me do mass manual updates quite easily, but most of the time I can do things better and clearer myself. All right, so there's someone else who's giving kind of a similar opinion. So let's just go see. I'm looking into reviewing blah, blah, blah. I'm implementing the plan, I'm going to explore. I'm determining the mass, I'm breaking down the tasks. Well, that's nice. It didn't give me anything though. So I'm going to say where is my executive summary? Please put it on a new worksheet. Again, this kind of highlights what I just read from Sean. I could have thrown this into a pivot table, started to get TransAm and halfway done versus continuing to write. And so you really need to think what are the right use cases? I think scoping. So I've talked to a guy who's the head of modeling for Pete. I think PwC or KPMG one of them. I think it's PwC globally and they've looked at the entire modeling process for their clients and said, hey, there's 142 different kind of touch points, roughly 70 of those. We could use AI to be more efficient. But as they go through it and they try to calculate the time savings, what they found is either it doesn't save us enough time to justify the cost, it's not something the client will pay for more for, or it doesn't allow us to win more deals. 95 of the time across that entire supply chain. They're finding, hey, it doesn't make sense to implement it in their case they're finding most. The benefit they're getting from AI is not in the model build. It's in the stage where they're scoping and working with the customer. Hey, go review the meeting we had. You recorded it all. Write down what all the assumptions were, tell me what questions I need to ask, help me do that type of thing. And so that's where they're finding most, most their savings. I'll have them on the podcast here in a while to go deeper on that. But that's another example that kind of gets to the point of just because we can use Gen AI doesn't mean we should. And we have five minutes left. We'll see if we get any file here or not, and then we'll wrap up. So while we're waiting for that, if anyone has any comments or questions, I know most people, I know several have dropped. I can see that. I know this isn't like I said, doing this live is not the most exciting, but I really felt it's a conversation worth having. I want people to see this stuff and see how it works and how it doesn't work and how you need to start thinking about it. Because if people aren't learning and thinking about this right, the error rate's going to go up, not down for the files we build. And that is scary because, you know, 90% of files have a have an issue in them already. So with that note, we're waiting for that to run. I'm going to throw this up on the screen because I always do if anyone wants a kind of modern Excel course. Unfortunately it's not modern with AI. I hope to have some things coming on that front in the future, but I don't yet. I have one that covers Excel tables, power query and an intro to dynamic arrays. It's $60. You get 30% off with the code webinar. You're taking the same Excel stuff that's taught in the Wharton online FP and a program. It's the exact same videos. I know because I created both and I released the the same thing. So if any of you are interested, there's the code there. You can also find it on my website. And then here's my course bundle where you get my business partnering. Three courses, modeling, design, data analysis, modern Excel, about 10 hours. Again, doesn't have the AI stuff yet. Something at some point we'll add and there's 30 off that regular price. You can use the code webinar. So been highly reviewed. I haven't, you know, ever I, for the most part, people have been very happy with those, so feel free. They're also available on my website. Let's come back and see how it did. It's still writing the code. If you need a drop, feel free. I'll hang on until we see what this if this last one comes back or not. So feel free to stay with me. Feel free to ask any questions in the comments, but I'm going to stop talking for a moment unless I see a question or it finishes because my voice is gone. Ah, we do have an executive summary sheet, so we've at least made progress. Don't know what it's adjusting for consistency since it doesn't show anything. So you know Sean, as you mentioned, there are some mass, mass manual updates to let you do right. That might be worth the price. Some of these are 10, 20, $30 a month, even if it just saves you two, three hours and you're billing, especially if you're running your own business. If you're billing a client at $300 an hour and it allows you to build an extra $300, you know, an hour, three hours a month, that's $900. Is 50 bucks worth the tool? I'd argue yes. So I'm not saying these tools are not worth it. I think most of them are. I'm just saying you, you, you will get more value by getting better at Excel than spending forever trying to prompt. And right now you still have to do a lot of prompting, a lot of waiting. It's. It's not ready for prime time. For many use cases, it's supposedly finalizing recommendations and Insights hasn't shown anything yet. So let's see if it spits out something here. I'm going to give it two more minutes and then we're just going to kill it. Unless somebody tells me they're dying for me to complete it and watch the final version. I'm not sure there's a ton of benefit, but while I'm waiting for that, like I said, we'll get another minute or two. I just released an episode today with Janet with in Schnorr and Giles Mel, both global modelers. Okay, here we go. So we get. It's. It's coming. And here's what somebody said. AI will erase modeling. Just give it. Give a time. Most models are superficial or too off the consequent reality. Most of them are just intellectual exercise based on schematic thinking and nomenclature. AI will do it better. Now, if we're building poor models, that may be true. But if we know what we're doing, I would argue that. At least not in the short term. All right, so let's see how it did. Let's just take one quick look and then we'll finish. So it said, hey, you have data for nine months. Here's how many units you have. Your total revenue, your total cogs, your gross margin, average price. It's still working. It's investigating a formula. It didn't give us this yet. Daily revenue, but you can see formatting everything else, it's just. Okay, I don't know. Date revenue and gross profit. Why a put date on the axis? Beyond me what this represents, I have no idea why. It only captured a small portion of the data. So, anyway, let's go ahead and end there. It's still working. I think I've made my point. Thank you for joining me, and as always, super exciting time. Don't sleep on AI, but don't stop learning. That's my message for today.