What Happens When the AI Tools Fail Basic Math and More with Ian and Giles
In this episode of The Mod Squad, hosts Paul Barnhurst, Ian Schnoor, and Giles Male continue their hands-on testing of AI tools for financial modeling. This time, they put Subset, an AI-powered spreadsheet tool still in beta, through its paces. The hosts explore whether Subset can realistically handle core financial modeling tasks, including importing Excel files, building three-statement models, and applying basic accounting logic. Along the way, they uncover significant limitations, bugs, and logical errors that highlight the risks of relying on unsupported or immature tools.
Expect to Learn
What Subset promises to do and how it performs in real-world testing
The challenges of importing Excel files into non-Excel environments
Why basic accounting logic still breaks many AI modeling tools
The risks of using outdated or unsupported AI tools found online
What it would actually take for professionals to move away from Excel
Here are a few quotes from the episode:
“There’s no AI on the planet that should tell you gross profit is revenue plus costs.” – Ian Schnoor
“It’s clever, but massively flawed and unreliable in lots of areas right now.” – Giles Male
Subset shows ambition in trying to act as a full AI spreadsheet, but the testing reveals serious issues, from incorrect formulas to flawed financial logic and unstable performance. While the tool demonstrates how far AI experimentation has come, it also serves as a cautionary example of why finance professionals must validate outputs and maintain strong technical foundations.
Follow Ian:
LinkedIn - https://www.linkedin.com/in/ianschnoor/?originalSubdomain=ca
Follow Giles Male:
LinkedIn - https://www.linkedin.com/in/giles-male-30643b15/
In today’s episode:
[02:40] – Welcome back to The Mod Squad
[05:04] – Introducing Subset and its promises
[08:38] – Importing Excel files into Subset
[11:27] – Errors, bugs, and beta limitations
[13:50] – Building a three-statement model from scratch
[19:25] – A Basic Revenue Reality Check
[22:37] – Why Excel Is Hard to Replace
[27:10] – Lessons learned from testing multiple tools
[30:01] – Why Structured Data Matters
Full Show Transcript
The Mod Squad. We're the Mod Squad. The Mod Squad featuring Ian Schnoor, executive director of Financial Modelling Institute, Giles Male, humble MVP and co-founder of Full Stack Modeller and Paul Barnhurst, the FP&A guy.
Co-host 2: Ian Schnoor (00:31):
I think it's still worthwhile to see it because it just goes to show people that there's, you got to be careful. There's stuff out there. You might find stuff out there that is old or not being supported or is offering to promising to do something. But here's the thing, Paul, that might be true, but I got to tell you. Can you scroll back to the top please? Js. I don't care what version of AI you're using, there's no AI on the planet version one or minus 10. That should tell you that gross profit is revenue plus costs.
Host: Paul Barnhurst (01:03):
All right, we're back. This is our next episode. We've decided to be the Beanie and Glasses people. Giles, what's with the outfits today?
Co-Host 1: Giles Male (01:13):
Well, we didn't want people to realise that we may or may not, it's unconfirmed. Be recording two episodes back to back. I mean, I've told them now, but we've changed. We are in disguise. We are so professional.
Co-host 2: Ian Schnoor (01:28):
It's unreal. We would never, ever record multiple episodes back to back, so obviously we just happen to be randomly wearing different outfits.
Co-Host 1: Giles Male (01:37):
I think we fooled everyone. I think we will fool everybody.
Co-host 2: Ian Schnoor (01:39):
Fooled everyone. Nobody would ever think this is happening right after a prior episode. Good job.
Co-Host 1: Giles Male (01:45):
Yeah.
Host: Paul Barnhurst (01:46):
Alrighty. We fooled them all. Can
Co-host 2: Ian Schnoor (01:47):
I take 'em off now? I
Host: Paul Barnhurst (01:48):
Pick off my glasses so I can see, but I'll keep the beanie. We got to have that going.
Co-Host 1: Giles Male (01:55):
It's cold here.
Host: Paul Barnhurst (01:55):
I'm keeping, and I want to say mine was made by my wife, so I got one. Would she make one
Co-Host 1: Giles Male (02:00):
For me?
Co-host 2: Ian Schnoor (02:02):
She would. Can I get her to make one for me too around here? Just let me know. It's getting cold in Canada, but I'm going to dha.
Host: Paul Barnhurst (02:10):
All right. I'm going to stick with it. Are you going to be brave and go with the hat throughout Giles?
Co-Host 1: Giles Male (02:15):
Yeah. I dunno how brave it is, but I'm just a bit chilly so I'm keeping it on. I'll keep it on.
Host: Paul Barnhurst (02:21):
Why don't we get started? We'll do quick intros here. Welcome to another episode of the Mod Squad. We're a minute and a half in and now we're introducing ourselves having a little fun. This is what episode nine or so I think we're getting up there. We're nearing the end of testing all these tools and so we have one more we're going to look at, we'll talk a little bit about different tools we've considered testing and what's next with the Mod Squad. So why don't we start with intros of everybody. My name Paul Barnhurst. I'm known as the FP&A guy, been hosting Financial Modeler’s Corner for a couple years and thrilled to have with me Giles and in as co-hosts, I run my own business. I do a lot of podcasting and we're going to have another fun episode today in why don't we go to you next.
Co-host 2: Ian Schnoor (03:05):
Sure. Thanks Paul and great to see you both again for another episode. Yeah, Ian Schnoor, executive director of Financial Modelling Institute. My life and career have been focused on financial modelling and it's a real pleasure and thrill to be doing these episodes with the two of you as we really learn a lot and test and hopefully share some interesting insights and knowledge with people all over the world around where we're heading and what's going on in the world of ai. Alrighty, Giles.
Co-Host 1: Giles Male (03:33):
Hello. Yeah, Giles Mail, Amateur Excel gangster rapper as you sort of found out last episode and in my spare time, co-founder of Full Stack Modeller and we help individuals and teams get better at Excel and financial modelling. Really happy to be here. It's been a hell of a journey. It feels like we've been doing this for ages nine episodes. Doesn't actually sound like that much, but I feel like I've learned a huge amount over the course of those.
Host: Paul Barnhurst (04:00):
I totally agree. We've learned some. So let's do a quick recap, what we've tested so far and then what's on cap for today. So Excel agent we've done, we've done Trace light, rosy ai, rest in peace cabs, AI shortcut. Yeah, can't forget shortcut. That's Giles's favourite truffle pig, Melder and today, elk, elk, A thank you and Elk a. And today we're going to show one of the tools that's out there that we had on our list to test. It's still in beta. We'll talk a little bit more about that and then we're going to talk about just some of the tools we've seen out there that hey, here's why we haven't tested them, their wait list or the different reasons, and then talk about what's next As we feel like we're near the end of testing these tools for now, we may bring some others back on.
(04:48):
We have some potentially on the list when they're ready, but we're talking about some guests and different things. So why don't we start with the tool we had scheduled for today is subset and I'm going to bring up their website here and we'll talk a little bit about them. I think they're a unique one and I love what they're trying to do. Very difficult to do, but let me just bring 'em up here. So I think the first thing is subset is one I found when I was searching is we're trying to find all the tools. It wasn't someone reached out to me, they didn't complete the survey like, okay, an AI spreadsheet editor build to make finance professionals more productive, generate a p and L from A PDF, build a DCF or update a returns model. So that's thought. All right, this makes sense to test kind of right up our alley or what you'll notice first is there's no pricing.
(05:37):
You can try it 100% free. It's its own spreadsheet. I do find interesting. It says it's backed by Merrill and Site Partners Index Wharton and very similar to all the other sites. A lot of white space generate a p and L from A PDF, build a DCF update, a return model, automate grunt work, and talks about how some things are on the roadmap for these different exercises. Not sure I fully understand, receiving a, creating a teaser data room access. So it looks like kind of steps in some kind of m and a activity it's showing in what it's building. So thoughts on the website before we jump into the tool Giles?
Co-Host 1: Giles Male (06:19):
Yeah, I mean, so it's interesting that it says build A DCF, so I appreciate it's in beta, but if it can build a DCF, that's quite a thing. Similar to what we've said before, the website looks pretty clean. It's giving some examples of what it could do. Format input in yellow fill. I love it already. That's what I do. That's what I do. Definitely. Right.
Host: Paul Barnhurst (06:41):
They probably were spying on one of your models when they created that part.
Co-Host 1: Giles Male (06:45):
Formatting does look unbelievably similar with the dark blue headers and the white. Anyway, I'm sure it's fine. It's scraped all of my models off the internet to build an AI tool. Shocking.
Host: Paul Barnhurst (06:55):
There we go. Ian, any thoughts on the website?
Co-host 2: Ian Schnoor (06:59):
Yeah, it feels a little bit more sort of bootstrapped, homegrown at this point. I mean the formatting of it, the other ones seem to have embedded gifs or videos or they're just a little slicker. This feels similar but it feels like it's a little bit as homemade so to speak. I don't know if that's fair or not. I dunno how if you think it just, yeah, Paul.
Host: Paul Barnhurst (07:23):
No, I was going to say I think that's fair is I was going to say yeah
Co-host 2: Ian Schnoor (07:25):
Anyway, but we'll see. And we took a look and the one key thing about this tool, as we noticed, this is not built into Excel. This is its own separate piece of cloud-based software that we need to look at by logging, going in and opening a new tab in a browser
Host: Paul Barnhurst (07:42):
And I spent a little bit of time looking at the websites, a couple interesting things. If I click here, let's open in a new tab that brings up the spreadsheet with the chat. We'll get to that here in a minute. We can see it's integrated a bunch of different models. The next thing that I thought was interesting is it has quite a bit of help docs, but it says we're building the best AI spreadsheet. We spent three plus years building our spreadsheet and now we're focused on designing the best user interface for building financial models. So learn more about how to use subset and it shows what functions it includes, what shortcuts it includes because it'ss own tool, it doesn't include all the functions. So we'll we're going to try with some of the Excel files and maybe we'll just try asking it a question and see how it builds on its own. If we have any kind of challenges as we've seen so far. We had a little bit of challenges loading some Excel stuff. Why don't we start, do you want to try loading one of your Excel cases here, Giles, and let's see what happens.
Co-Host 1: Giles Male (08:46):
My personal preference is to be in Excel, but I mean if you're a Google Sheets user, you obviously don't care what you're using anyway, so maybe you would
Co-host 2: Ian Schnoor (08:53):
Switch.
Co-Host 1: Giles Male (08:53):
I caught it
Co-host 2: Ian Schnoor (08:54):
Loud and clear.
Co-Host 1: Giles Male (08:56):
Didn't have to worry about Oh, I took you seriously. I was Oh, you didn't hear? I'll say it
Co-host 2: Ian Schnoor (09:02):
Again. No, I'm never serious. No, I got it loud. I just thought you wanted a chance to repeat it a second time, but I'm actually a fan of Google Sheets. I think there's a place for it. I think it's a great product for the right, even with our team when we're collaborating together on pretty low simple files, it can be great, but I hear you
Co-Host 1: Giles Male (09:23):
Straight away a bit weird because it's kind of broaden in. So I had to import the file and it's certainly put the tab names in, but the formatting is, it just doesn't look like my spreadsheet. So the data is
Host: Paul Barnhurst (09:37):
There. Yeah, a little bit of truffle pig to a lesser extent where when you brought in a sheet, they didn't quite bring in everything.
Co-host 2: Ian Schnoor (09:43):
Yeah, this is the challenge of opening up an Excel file into a non Excel environment, right? Is it things get missing, so what happens when you run it?
Co-Host 1: Giles Male (09:52):
Please log in to continue. Okay.
Co-host 2: Ian Schnoor (09:54):
Okay, we're back hitting.
Co-Host 1: Giles Male (09:57):
Go. Can I keep, wait, you know what?
Host: Paul Barnhurst (10:00):
You have to reopen your workbook. You got to
Co-Host 1: Giles Male (10:01):
Reopen your file. Good lord. Okay. Okay. But
Co-host 2: Ian Schnoor (10:06):
It is working to determine the formula requirements. You see, it's trying to now calculate your request on a blank.
Host: Paul Barnhurst (10:12):
It says the green cells are there and I assume we're dealing with, yeah, you might want to kill that one. Why don't you try? Can you go up to the different models? Try llama. We've had no tool. That's a metas open source one. I'd just be curious
Co-Host 1: Giles Male (10:27):
Why not? Oh, okay. Proposed. Change it to please answer. Oh, here is the solution for level one.
Host: Paul Barnhurst (10:37):
Go ahead and hit accept. Let's see if it brings it in.
Co-Host 1: Giles Male (10:41):
I just can't see it. It says here is the solution.
Host: Paul Barnhurst (10:43):
I know, but you have that accept button, click accept and see if it,
Co-Host 1: Giles Male (10:47):
I don't understand.
Host: Paul Barnhurst (10:48):
Try again. I guess please solve level one. Maybe tell it where to insert it in cell E 77 or something
Co-host 2: Ian Schnoor (10:56):
You didn't ask nicely. Giles, can you maybe it wants a please. Let's see here. Accept. Accept. So, oh, oh, hang on. Is it doing something somewhere?
Host: Paul Barnhurst (11:08):
Give it the specific sell and see what, try a different model.
Co-host 2: Ian Schnoor (11:11):
Let's go back to
Host: Paul Barnhurst (11:11):
The typical lines.
Co-host 2: Ian Schnoor (11:12):
Maybe we didn't understand what the green cells, again, I think it's helpful and healthy to be able to see us struggling like this, right? I mean this struggle is an important part of the process. All right, so
Host: Paul Barnhurst (11:23):
Clearly this tool is still, like you said in betas try for free.
Co-Host 1: Giles Male (11:27):
Yeah, it doesn't
Host: Paul Barnhurst (11:28):
Seem
Co-host 2: Ian Schnoor (11:29):
To struggle bringing in an accelerator. What would be your personal threshold for patients before you get frustrated and Huck it? I mean,
Co-Host 1: Giles Male (11:38):
Oh, similar to my patients with people that overused vlookup.
Co-host 2: Ian Schnoor (11:43):
Be careful. I'm a big VLOOKUP prop here.
Host: Paul Barnhurst (11:50):
VLOOKUP is my favourite. Giles, what are you talking about?
Co-host 2: Ian Schnoor (11:54):
The point you're not going to have a lot of patients to spend a few hours working with this at this point?
Co-Host 1: Giles Male (11:59):
Yeah, to be honest, I dunno what it's doing. So I dunno if it's put the answers somewhere. I can't see anything in the green cells
Co-host 2: Ian Schnoor (12:07):
Certainly not responding the way the other tools did. Is it?
Host: Paul Barnhurst (12:11):
Yeah. So clearly at least it's not ready there. If you're importing Excel files, can
Co-host 2: Ian Schnoor (12:17):
You try one of the opus, one of the Claude
Host: Paul Barnhurst (12:21):
Sonet or
Co-host 2: Ian Schnoor (12:22):
Try one of the anthropic tools there? Yep, this is anthropic here. Try again. You, what do you ask?
Co-Host 1: Giles Male (12:31):
Oh, an error. It just said error. Okay, please try again. Error. Okay, I think I'm probably at the point where
Co-host 2: Ian Schnoor (12:43):
Does it say error? I didn't miss that where it said error
Co-Host 1: Giles Male (12:45):
At the top. So try
Co-host 2: Ian Schnoor (12:48):
Try again. Lemme see what it says. Error I missed. Oh, error there. Error on error occurred. Yeah. Okay,
Host: Paul Barnhurst (12:55):
Why don't we scrap bringing in Excel files and just try it. Lists examples saying help us build remember with I think was truffle pig. We asked it to build Nvidia. DC, F. Yeah,
Co-host 2: Ian Schnoor (13:09):
Try that Js, you got it open, why don't you try?
Co-Host 1: Giles Male (13:12):
I had a three statement button so I
Host: Paul Barnhurst (13:14):
That's fine. Three statement state. So click that. So let's see what it does when it builds on its own so we can at least get a bill for the tool.
Co-host 2: Ian Schnoor (13:19):
Okay, so, oh, this is cool. Alright, this is some life we we're watching now but we didn't tell it which, how does it know what company or
Host: Paul Barnhurst (13:29):
It didn't? It just said it's a manufacturing company. Use the data for the inputs forecast five years use R one,
Co-Host 1: Giles Male (13:36):
R one, Q1 references.
Co-host 2: Ian Schnoor (13:38):
So the magic is happening now this is nice to see. It's from scratch. It's building something here from scratch. You know what, I'd almost love you to run it again and pick Amazon or Tesla or Microsoft or something. I mean I don't know what it's building here and why is it, I can't quite see, but is it showing us in our one C one format is that's what's going? Yeah, yeah,
Co-Host 1: Giles Male (14:01):
It's exactly what it's doing. I dunno whether it's going to convert that
Co-host 2: Ian Schnoor (14:04):
At the end. Is it going to convert all that? I mean let's see if it does anything with that. It's a BC manufacturing company three statement financial model.
Co-Host 1: Giles Male (14:13):
I don't really understand the reference though if it's just saying, oh
Host: Paul Barnhurst (14:18):
And now it's joint numbers. So it looks like we've got some value errors. Now we are using Claude
Co-Host 1: Giles Male (14:25):
Now,
Host: Paul Barnhurst (14:26):
We'll just give it a minute here. Supposedly it's done
Co-Host 1: Giles Male (14:30):
Well this makes no sense. Revenue is a hardcoded number. COGS is linked to the revenue number, so it's the same number. Sure,
Co-host 2: Ian Schnoor (14:37):
That's exactly how it works in real life. Perfect. So revenues equals costs
Co-Host 1: Giles Male (14:41):
And then gross profit is just revenues plus cogs.
Co-host 2: Ian Schnoor (14:44):
Brilliant. Wow. It's genius. Look at that. I love that. The gross profit is double the revenue. Wow.
Co-Host 1: Giles Male (14:50):
So from revenue of $150,000, let's call it we end up with 2.1 million net income. That's incredible. Sorry, how do I start? Something like that.
Host: Paul Barnhurst (15:01):
What I wonder because go to the list of models for a second. If you do the dropdown where it says clot opus and everything,
Co-host 2: Ian Schnoor (15:07):
Most
Host: Paul Barnhurst (15:08):
Of these are quite old versions. There's no GPT five. I'm wondering if they left it out here but kind of stopped development on this thing. They didn't respond to any of the emails, older models. It feels like somebody spent a long time on it, maybe saw all the AI stuff and I don't know or
Co-host 2: Ian Schnoor (15:26):
Take down. I mean
Host: Paul Barnhurst (15:28):
It's just stop
Co-host 2: Ian Schnoor (15:28):
There. But again, I think it's still worthwhile to see it because it just goes to show people that you got to be careful. There's stuff out there. You might find stuff out there that is old or not being supported or is offering to promising to do something. But here's the thing, Paul, that might be true, but I got to tell you, can you scroll back to the top please? JI don't care what version of AI you're using, there's no AI on the planet version one or minus 10 that should tell you that gross profit is revenue plus costs, right? So I can't argue with that at all.
Host: Paul Barnhurst (16:05):
That's
Co-host 2: Ian Schnoor (16:06):
A new invention that's not newly invented with GPE, the latest GPT, right?
Host: Paul Barnhurst (16:10):
So what you're telling me is in no world can I add my cogs to my revenue and use that for my revenue base. I'm
Co-host 2: Ian Schnoor (16:19):
Pretty sure that's not a new invention.
Host: Paul Barnhurst (16:21):
I'm sorry for my gross profit. Yeah,
Co-Host 1: Giles Male (16:22):
I don't know. It gets worse the further you go down. So depreciation, sorry, SG and a is just equal to the gross profit Depreciation is equal to the sg and a operating income just adds them all together I think including gross profit are and just
Host: Paul Barnhurst (16:38):
Expenses equal to operating income
Co-host 2: Ian Schnoor (16:40):
Like
Host: Paul Barnhurst (16:41):
It's
Co-Host 1: Giles Male (16:41):
Just linking to the robot.
Co-host 2: Ian Schnoor (16:42):
This concerns me because clearly you didn't write this prompt. This was a default prompt built in that said,
Co-Host 1: Giles Male (16:49):
Hey,
Co-host 2: Ian Schnoor (16:50):
Click this button and it'll run a model. So stack plus income equals net income.
(16:55):
I would love, can you try this? Can you write one more thing I just can you say build a three statement model is that button there again, I'd love to say build a three statement model for Tesla. Try that. Try to say build a three statement model for Tesla. I want to see build a three statement model for Tesla. Make reasonable assumptions. Let's see what it does make reasonable. I want to see, again, if it doesn't understand the accounting logic and follow up as Sure, great. Let's see what it does. I mean it was able to build something by just when you clicked on a made up manufacturing company model, but let's see if this is connected. Let's see if it has the ability to pull historical data, it's doing something right. Okay, I want to see if it buggers up the calculation of revenue cost, profit.
Host: Paul Barnhurst (17:54):
You know what another key takeaway I'm taking from this so far,
Co-host 2: Ian Schnoor (17:57):
Right? That
Host: Paul Barnhurst (17:58):
Is all the work that these tools are doing on top of the LLM.
Co-host 2: Ian Schnoor (18:01):
Yeah.
Host: Paul Barnhurst (18:03):
Now I would expect if we just went the Claude, we'd get something better. So it's interesting how much of this is their instructions, their logic versus the LLM, but it does show there is some value in what each of the tools are doing for sure.
Co-host 2: Ian Schnoor (18:19):
Now what it's got, yeah, they're trying absolutely. I value it's got to, it's trying to do 20 24, 20 25, 26 forecast.
Host: Paul Barnhurst (18:30):
Looks like it took two years of actuals and four years of estimates. I'm wondering if it's not going out to the web, if it's one of an AI that only has data through 24 or why it didn't. Let's see if it's getting
Co-host 2: Ian Schnoor (18:43):
Revenue, cost of revenue, gross profit. So it's showing revenue and then it's showing costs which are negative and so the gross profit is mathematically working correctly here, right?
Host: Paul Barnhurst (18:58):
Can you enlarge it a little bit? Giles your screen. I'm having our time reading these numbers. I'm squinting constantly. Thank you. That's better. Feel much better now. My old eyes, I mean I'm not 40 like you Giles.
Co-Host 1: Giles Male (19:13):
Okay. Yeah. Okay,
Host: Paul Barnhurst (19:15):
First stop real quick. This is just do a simple sniff test to the numbers. The sniff test automotive revenue is the same as energy and services in 22 and it's bigger in 23. That's not right.
Co-host 2: Ian Schnoor (19:27):
It's the exact same. Yeah,
Host: Paul Barnhurst (19:29):
Yeah. Exact same in 22. And then energy and services is much larger in 23. These numbers are just can't be right? I
Co-host 2: Ian Schnoor (19:38):
Mean that's possible, but what's highly unlikely is that the numbers were identical to each other. Well
Host: Paul Barnhurst (19:43):
Yes, but I also mean just knowing Tesla, 80 to 90% of their revenue comes from automotive. So clearly these numbers are not right.
Co-host 2: Ian Schnoor (19:52):
So there's something that's not right.
Co-Host 1: Giles Male (19:55):
I mean that's all we need to know, isn't it? I just think it's, as you said, it's not there.
Co-host 2: Ian Schnoor (19:59):
So you're right Paul, maybe this is not even being used or supported and it's just up there, but the danger. But I think it's very valuable to recognise the danger of trolling the internet for AI tools because there is stuff out there for free and it will do stuff and maybe you'll find something free that's better than this but still not good enough and still making mistakes. I think it continues to demonstrate that you have to understand what you're doing, what you're looking at. You can't sacrifice your own knowledge and your own skills. You need to be able to check, validate, and it is still basing itself on the same underlying LLMs. So I think you need strong diligence in all of this and it was really interesting to see that this is out there anyway.
Host: Paul Barnhurst (20:40):
And then I want to take a minute just so our audience sees, we've spent a tonne of time trying to make sure we've looked at every tool that's out there. So I'm going to share two things on the screen and I think one is more so people are aware. So let me just share my screen here. As I went through this, one thing I looked at is just what are all the alternative spreadsheets out there beyond Google Sheets and Excel? So jazz, no, don't tell me. And we didn't necessarily test these outside. We only tested those that really were advertising themselves as AI and for finance, but there are several out there. Rose is a really interesting one that I've heard a lot of really good praise about. I've played with it some. It's really good in marketing. Great at bringing in reports. So what we're saying is there's a lot of stuff out there and you'll find some of these paradigm is by a 21-year-old, 22-year-old college student that's raised millions of dollars.
(21:34):
She graduated from Harvard or one of the big Ivy league schools and it's a really interesting one, the spreadsheet to automate manuals, data collection and they bring in a tonne of agents and so it's kind of an interesting approach as they've delivered, said, we're not going to try to be Excel. We're going to try to focus on really automating your data collection. As you could see, hey, bring in a company and let us bring in the revenue and categories and all kinds of stuff from the web. So not modelling, but just to show people equals is one out there. Sheet rocks. Quadratic is one that considered testing, but they're not really designed for a finance. They started as the data science spreadsheet and then they pivoted the ai. So quadratic makes your spreadsheet, could work easy with brilliant results. Add data, add questions. I've looked at it before. It has 95% of the formulas that Excel does and has some new ones. I'm curious as I just show some of these, Giles, what would it take for you to step away from Excel and use another spreadsheet?
Co-Host 1: Giles Male (22:34):
Lots of money that I'm not sure there is anything that would get me out of Excel you would've to pay me to do that because
Host: Paul Barnhurst (22:43):
Basically we going to have to pry it out of your cold dead end.
Co-host 2: Ian Schnoor (22:45):
You'd have to be tied up with someone stepping on his back and forcing him disabling Excel from your computer. It would be pretty be pretty severe, wouldn't it?
Co-Host 1: Giles Male (22:57):
I guess the serious answer to that is there was a period Google Sheets, as much as I joke about Google Sheets, it is actually pretty good in lots of areas and there's certain features that have been ahead of Excel tables now has got features in Google Sheets that's better than Excel. They were way ahead on the array kind of logic and they still are in a sense, they've got array of arrays working. So I think the serious answer is I'd have to see something that was so significantly more advanced than Excel that it was worth the trouble me learning it and getting familiar with it to warrant stepping away.
Host: Paul Barnhurst (23:30):
How about you Ian? What would it take for you to, on a serious, we know Giles, we'd have to give 'em a butt load of money, but what would it take for you in
Co-host 2: Ian Schnoor (23:38):
Well, so I mean in all seriousness, and I will show this on purpose. I mean people talk about why there has not been as much, I mean Excel does have a cloud version of Excel. It's not very, to my knowledge, it's not very widely used. I'm not aware of people that use it. I've played with it. I think that there's a hesitation to move and away from a desktop software to a cloud software for spreadsheets. Yeah, some people talk about the security. I'm not sure if that's a real answer or just a made up answer because other types of data, people are comfortable in the cloud. I am comfortable with the work environment and again, I'm going to show you, I mean when I run my webinars and I taught, I'm a big advocate of using the keyboard and this is the keyboard that I use and I joke around, but this is my keyboard, it's got no letters on it.
(24:23):
And I do that on purpose because I want people to work quickly and efficiently and I keep my hands on the keyboard, but I don't look at the keyboard. My point is I'm really getting into the tool and working it and playing with column width and changing visual dynamics and getting really into it. And that's difficult in a cloud environment. Everything is slower and drags and you don't have the same delicate ability to modify and tweak things when you're working in a cloud environment. That's my experience. And so a hardcore Excel user modeller wants to work right in the software, in the keyboard work fast. And I don't know that that's easily replaceable in a cloud environment. So I'm not going anywhere anytime soon either. Giles, would you agree with some of that?
Co-Host 1: Giles Male (25:10):
Yeah, absolutely.
Host: Paul Barnhurst (25:12):
Alright, so back to here. You could see these are some of the others, like I mentioned that are out there equals has got a little bit of traction. They're kind of an interesting one. They started kind of saying, Hey, we're building the next spreadsheet, but a couple years in, they found their product market fit, right? Go to market analytics. You trust insights that drive revenue today, even though it's a spreadsheet type solution, they very much focused on, hey, you can do SQL, writing your own data, building the reports right in there versus trying to compete with Excel. And so what it tells me is you have to have unique use cases if you really want to compete, right? I mean nobody can just build a spreadsheet and like, all right, I'm good to go. And then here I wanted to show these are all the tools, this is the list of different tools that basically found out there that are AI in some way.
(26:07):
And then I went through all these and said, okay, are they at a point we could test them or are they on a wait list? Several are on a wait list. Some were like, oh, they're their own software, or they're just talking about building something weren't even in beta yet, still kind of in concept. Some were different enough to like, oh, all they're doing is helping you write formulas in Excel even though they say ai. And so we just want you to show, I mean, look at this list. We got 38 things we put on this list. We went out and looked at everything we could find out there. I did quite a bit of searching and looking to just get an idea of what's there and we've brought you at this point what we think is a very good list. We may test some others later on as some of them come out of wait list. I think Index would be an interesting one since they're backed by OpenAI Mosaic we've talked about. So we'll continue, but I think we're coming kind of near the end of going out and testing the tools. Agree Giles in thoughts on that?
Co-Host 1: Giles Male (27:10):
I think certainly the third party ones, I think we've got a collective interest in looking at some of the LLMs to within whatever reason and sensible capacity we can. I think that could be interesting. I suspect we'll come back to agent. I mean we might come back and review all of these tools again in six months, but it's really interesting for me that I had a kind of feeling in my head of what these would be able to do. And I'm not sure actually that's a lie. There's certain things they've done that I have been impressed with, but my position is exactly the same as it was at the start, just with more evidence now, which is that it's really clever, massively flawed at the moment. Unreliable in lots of areas. But I love it. I keep supporting it. I want to be using these tools. I use AI in lots of parts of my life already. Yeah, I hope it progresses
Co-host 2: Ian Schnoor (28:04):
Right. Ian, your thoughts there? Listen, I mean obviously AI is evolving by the minute. I don't know if either of you've played with the latest release of Nano Banana for, have you tried that at all? I played around with that a little bit yesterday. It seems like the earliest entree into AI was research, right? Crawling through the internet, aggregating and doing and actually generating research results for you so you didn't have to click on various links. It feels like it's a bit of a step up and a leap for them to actually build spreadsheets. Now it's working and they're doing things and I have no doubt that they are going to keep progressing. And I highly encourage everybody, especially young professionals, you don't want to miss this boat. I think now is the time. Here's the thing, I strongly believe now is the time to watch our videos to start playing around because it's safe.
(29:03):
Because I don't think anyone's expecting you to be a world renowned expert at this point because I think generally people know that there's the good, there's the bad, there's the ugly. So I think it's a good safe time to be learning, trialling, piloting, staying afloat, making sure you're aware of what's going on. You do not want to get left behind. And there will be a time when these tools get better and better, but I can continue to believe that they will be your partner if you know what you're doing, they will be your partner. In the same way that when Excel came around, people either adopted it or didn't and you can't really be working in business these days in finance or accounting if you don't know how to use a spreadsheet. And so I encourage people to kind of roll up their sleeves, try these things out, they're going to get better so that you can continue to use them as they get better in your own work and support you
Host: Paul Barnhurst (29:48):
And something I'll add. That's interesting. I've seen several of these fp and a tools with, they're starting to bring, they bring in the AI and obviously you do some modelling there and integrated three statement and they do a better job. Like you connect NetSuite and it builds out the full three statement. But the difference is that's all structured. There's the chart of accounts all there. They're working within a very detailed tool where they can give very strict instructions. And so they can get templates where, hey, we'll do a three statement or we'll do a headcount or we'll do your CRM data. So the more structured we can make this stuff, the more AI can help with templates and with help with use cases. We were really showing a lot of, I don't want to say unstructured, but semi and unstructured. And it also shows the importance of knowing what you're doing regardless. But the tools can,
Co-host 2: Ian Schnoor (30:41):
I mean, Paul, we found that, right? Paul, we found that normally in the eSports cases, even as they got complex by saying, here are the instructions, here are the green cells build a solution, super impressive, super difficult, but it is structured and sort of programmatic, right? It's able to kind of figure out a solution sometimes better than others and do it. But when you say, Hey, here's three years of historical financial statements, go wild and build me a beautiful three statement, five-year forecast model. We're not there yet. The
Host: Paul Barnhurst (31:10):
More instructions you can give it, the closer it will get, the more structure you can give it. I think that's one of the key things if anyone's wanting to try with ai, the more detail, the more structure, the more framework you can give it, the better the results are going to be. You still have to know what you're doing doesn't fundamentally change that. And then as far as what's next, maybe we'll talk a little bit, I know you mentioned LLM, Claude, we want to get our hands on Claude for Excel. We've tested Excel agent, which is basically copilot. And so we've done that. And then I know we've also as a group, we've discussed potentially bringing on some guests or things. So maybe talk a little bit of how you're thinking as we continue to go forward here, Giles, any of your thoughts of what people can expect?
Co-Host 1: Giles Male (31:53):
Yeah. Well, I think just personally, I really enjoy it because this is helping me learn so much and obviously I think it's nice to share and bounce ideas with the two of you. So I am not sure how much value there is right now in US just continually going, okay, there's another third party tool. What is really interesting to me is seeing how the LLMs stack up themselves. All of these tools, as far as we're aware, lean on an NLM or multiple L lms and some of them will have layers of code and additional instructions that may do something. But I just have a gut feeling that when we now look at chat GBT and clawed that they're going to perform really well. So I'm interested in that. And I'm also really interested because I don't think any of the three of us would call ourselves AI experts coming at this with a decade of AI experience. I would love to speak to people that do have more insight into this technically, or maybe people in the industry that have been leading the way themselves actually and have done even more digging than us.
Host: Paul Barnhurst (33:07):
So I think the message there and we'll get thoughts is this isn't the end of the series. We still have stuff we want to do, maybe pause on the third party tools, but we feel like there's something here long-term for us to continue to learn and bring our thoughts on how AI is impacting financial modelling. I think we all believe it's going to have a fundamental impact, a big impact. There's no question there. It's just trying to separate hype and reality. So in thoughts,
Co-host 2: Ian Schnoor (33:39):
Listen, I think it's been fascinating. I'd love to continue doing the following. I'd love to continue to try new tools that come available that look like they're impressive and worth trying. I'd love to come back to the tools we did test at some point to see how they have evolved and improved over time. I would love to bring in guests as we talked about, both on the technical AI side and on, we know a lot of people or that are bankers, finance professionals, accountants that are using it to talk about their own experiences. I think there's a lot of things we can do to support people all over the world as they themselves try to learn about what's happening in the world of AI and finance and spreadsheets in particular. So yeah, I look forward to continuing our discussions.
Host: Paul Barnhurst (34:20):
Great, and thank you so much for joining us on this journey. If you've been through all nine episodes, I feel sorry for you. No, I mean, thank you for joining us. We really appreciate it and we're excited to bring you more episodes. So please stay tuned for more from the mod squad as we continue to see how AI is changing modelling and spreadsheets and the way we work.
Co-host 2: Ian Schnoor (34:43):
Now, are you going to include a link at the bottom to how people can buy your wife's hats or is that They're
Host: Paul Barnhurst (34:49):
Not quite ready for sale yet, but I'll have to talk to her. We'll see.
Co-host 2: Ian Schnoor (34:54):
I think you're going to get massive. I might even do it for
Host: Paul Barnhurst (34:56):
Three. You never know.
Co-host 2: Ian Schnoor (34:57):
No, you're going to get massive demand. I already want to put an order of five, so she's going to need to get busy. They're great. I would like a couple. I'll go talk to her right now and let her know she needs to get busy on. I hope I can afford them, but I would like a few. So anyway, it has been great to be here and I look forward to continue our conversation.
Host: Paul Barnhurst (35:17):
All right. Until next time. Thank you for joining us.