We Tested Claude Opus 4.6, and the Results were Super Impressive
In this episode of The ModSquad, Paul Barnhurst, Ian Schnoor, and Giles Male put Claude 4.6 to the test on real financial modeling accreditation cases. From three-statement forecasts to complex debt sculpting scenarios, the team examines just how far AI tools have come. The results are impressive, but not flawless. The discussion explores what this leap forward means for finance professionals and whether modeling is truly entering a new AI-assisted era.
Ian Schnoor is Executive Director of the Financial Modeling Institute (FMI), the global accreditation body for financial modeling professionals. He brings extensive experience in modeling, training, and industry standards. Giles Male is Co-Founder of Full Stack Modeller and a two-time Microsoft MVP. He specializes in Excel, financial modeling systems, and practical AI implementation.
Expect to Learn
How Claude 4.6 performs on real financial modeling accreditation cases
Where AI tools still make subtle but significant modeling errors
The difference between automation and augmentation in AI usage
Why strong modeling fundamentals remain essential
Practical ways to begin integrating AI into your modeling workflow
Here are a few quotes from the episode:
“I’m not even doing this from a testing perspective now. I’m just using it because it’s adding so much value.” – Giles Male
“Modeling is just as much about the process as it is about the end result.” – Ian Schnoor
Claude 4.6 marks a significant step forward in AI-assisted financial modeling, handling complex builds faster than ever before. However, subtle errors still highlight the need for strong technical knowledge and human oversight. The future of modeling isn’t replacement, it’s skilled professionals using AI to work smarter and deliver greater value.
Follow Ian:
LinkedIn - https://www.linkedin.com/in/ianschnoor/
Follow Giles Male:
LinkedIn - https://www.linkedin.com/in/giles-male-30643b15/
In today’s episode:
[00:00] – Trailer
[04:06] – Testing Claude
[11:14] – Augmentation vs. Automation in Modeling
[18:14] – The Value of Documentation in Modeling
[28:43] – Debt Modeling with AI
[33:20] – Transition from Manual to AI-Enhanced Modeling
[38:16] – Testing with New Tools
[41:24] – Debt and Equity Modeling with AI
[46:45] – Claude's Progress & Areas for Improvement
[57:33] – Final Thoughts
Full Show Transcript
Host: Paul Barnhurst (00:00:31):
The Mod Squad is back. It's been a while, but we're having another episode here on Financial modelers Corner and we're super excited for this episode. I'm excited to be joined again by Giles and Ian, we'll start with quick introductions. Giles.
Co-Host 1: Giles Male (00:00:44):
Hey, good to be back. It does feel like it's been a very long time since I saw the two of you. So my name's Giles, co-founder of Full Stack Modeller, two time Microsoft, MVP, one of Ian's master financial modelers and getting deeper into this AI rabbit hole. And I have to say, kind of loving it all right in,
Co-Host 2: Ian Schnoor (00:01:04):
Well, not my master financial modeler js, but you are a master financial modeler as recognised by the rest of the cohort at FMI, the Financial modeling Institute. And yes, it's great to be back with the two of you. Wow. I feel like it's been ages since we've chatted and wow, so much has changed even in the past six weeks, right? It's unbelievable. Yes. I'm also like Giles spent my life and my career in modeling, in Excel, in training, and in the last 10 years running FMI, financial modeling Institute as the world's only financial modeling accreditation body and love spending times with you two guys as we test our AI tools.
Host: Paul Barnhurst (00:01:39):
Well, thank you. And so as everybody knows, my name's Paul Barnhurst, the FP&A guy, and today we're going to catch up. First we're going to talk a little bit about what we've been up to and what we're seeing in the market and then we're going to jump into Claude new model Opus 4.6. It recently released its open Claude to Excel, and I think we all agree that we've reached a new level. Things have quickly changed. I can't help but think back to one of the first tests we did, won't mention the tool, won't mention the episode, but remember we were trying to build the model and we're like, do we even show this? It was so bad, it really struggled. The formatting was terrible. Compare that to what you're seeing right now. So what I'm seeing is really impressive. I'm seeing models that are pretty close to complete. Yeah, there's still little issues. There's some things and we'll talk about that, but Phil's leaps and bounds about where we were before we've had a breakthrough. So Giles, your thoughts, I know you've done some testing and some playing before we get into it, but how are you thinking about all this? Yeah,
Co-Host 1: Giles Male (00:02:44):
I have. I've been testing and using Claude a lot. To be honest, I'm not even doing this from a testing perspective now. I think Claude as the app and also Claude in Excel have been so valuable to me. I'm kind of beyond the mind space of testing this. I'm just using it because it's adding so much value to my life. I did a meetup for Ken Pulse recently and I was doing various examples of things that Claude could do that none of the other tools we tested last year would've been able to do Now they might be able to do it now because they're all based on LLMs that have improved. So I do want to give that caveat. It does feel like there's been a massive shift and you used to use that word augmentation or acceleration pool magnifier. Yeah, that's it. And I feel personally now that is applying to me the things that might have taken me a huge amount of time as a trainer or whatever it might be. Getting data sets ready, workbooks, pulling things together, getting ideas for core structures. I am leaning on Claude for all of that. So yeah, very impressed and kind of excited by what this means.
Host: Paul Barnhurst (00:03:55):
And I'll admit, I'm leaning on Claude for a lot of, I did a course recently and leaned on it pretty heavily and it really helps with bouncing ideas and all kinds of different places. So love that. Anything you want to share that you've been up to lately? You talk about using it a lot, but anything else? I think I heard you're going travelling or something. Any exciting announcement there?
Co-Host 1: Giles Male (00:04:15):
So I'll be travelling pretty much for the whole year. So this is Excel on the road version two, pretty much full time with a friend of mine in the Excel community called Faye. And again, that Faye's kind of three steps deeper into copilot than I have ever been. And we're going to be talking a lot about AI and copilot and implementation. And for me this is a really weird transition because I was so against the early hype, as you know on LinkedIn. I don't think it's hype anymore. And that's been a very tricky journey for me to go through mentally. It's like, okay, but hang on, I'm now starting to say really positive things about AI when three months ago I was kind of berating other people for doing it, but I think now is the right time. It feels different. The reason for the excitement is these things are starting to do things quite reliably.
Host: Paul Barnhurst (00:05:06):
Interesting. I would say I still think there's a lot of hype, but the models are starting to live up to the hype from earlier. I'd say four, six maybe is living up to the hype we saw five, six months ago. I don't know if it's living up quite to the hype of today, but it's definitely exciting in, I know you're waiting to share some things there, so let's let you jump in.
Co-Host 2: Ian Schnoor (00:05:25):
Oh gosh. Wow. It is unbelievable what has happened over the past six weeks. I mean I've done a lot of travel, I've been leading webinars all over the world and we just got back from China and Japan where people are, our first cohort is going through the first cohort that completed the AFM accreditation programme. And of course every time I go somewhere and I talk to people online or in person, I ask, how much are you using ai? How much are you using AI in your models, in your Excel work, in your day-to-day? And you probably won't be surprised that for the vast majority of people, well over 90%, the answer is not much. I'm not there yet. Most people are, I mean the three of us are probably on the forward edge, the leading edge because we need to be your average person is is still a little scared, not sure how to approach it and is not using it a lot.
(00:06:16):
But that is about to change and things are pretty, as you've both mentioned, change very radically. We are just a few days ago there was some of you may have even seen this or heard this, an article came out by a tech entrepreneur named Matt Schumer. It's called Something Big is Happening. It came out on February 9th and it's a blog. It came out as a blog and it's been an article and it's got a lot of attention, millions and millions and millions of reads. Basically his analogy was we're now at that point where he starts by saying, you remember that time in January or February of 2020 when everyone started saying, Hey, there's this virus thing and you're going to have to be, and some people started buying toilet paper and you're like, what are you doing? You're crazy. And then literally within a few weeks the world had changed and we're like, oh my gosh, now we're all at home and no one's going out and there's no food on the shelves.
(00:07:14):
He's made this analogy that we are sort of now in that window, that window and it might be longer than a few weeks, that window where some of us, some people are like, oh my gosh, this is really becoming different and big. And most people as I mentioned, are still pretty oblivious and his prediction is that big things are going to happen faster than anyone ever expected. So we'll see where that is, but absolutely 100% we tested Claude, I tested it a couple tests this morning on an A FM case and on a CFM case, which I did not do with any of the other tools, you'll see how it performed. But let's just say this is unlike anything that any other tool has been able to do.
Host: Paul Barnhurst (00:07:54):
Thank you for sharing that. So what I'd like to do before we start kind of jump into testing just a question or two, Giles, has this changed your opinion on the idea that we need to know modeling well, do you think we're at the point now where somebody who understands business well can go in Excel, give it a prompt and build a model and use those results?
Co-Host 1: Giles Male (00:08:15):
No. So I would still stand by the fact that certainly for significant modeling in the sense of project finance m and a modeling, anything like that, I think you still have to know what you're looking at because the value of the model being right is so significant and you still with stochastic outputs, you can't risk one in 20 times, it's going to give you completely the wrong answer. So no, I still think you have to look at it. What I'm starting to think more and more is I understand that for the non-expert user, we're getting to the stage where there's massive value add. And the example I used on the meetup with Ken was you imagine you're a young entrepreneur and you have a business idea and you have no network in finance or modeling or anything else, but you know that you want to raise some money and you want to go to a bank or whatever, speak to some angel investors, you don't know what a financial model is.
(00:09:11):
But you know that if you go into Claude and you start saying, Hey, this is kind of my idea, this is what's in my head. Claude could probably build you a financial model with everything you'd need to pitch and it might not be perfect, but it might be 95% there and that's a hell of a lot better than having no model. And you wouldn't have been able to do that a year ago, five years ago, whatever it might be, six months ago, three months ago since four six has come out, hundred percent. So I'm still kind of in that space that for I try to remember that we we're in a particular bubble of people that look at financial models and think about them and go to quite an advanced level. But actually I think the argument from Microsoft for things like copilot for since day one probably would've been, this isn't necessarily for you for everyone else and I'm starting to believe that quite a lot more now.
Host: Paul Barnhurst (00:10:05):
I'd love to get your thoughts in and then I'll share some. So what would you say to that?
Co-Host 2: Ian Schnoor (00:10:09):
So the tools, I'm of a couple minds, but yes, I mean I still do agree with Giles. I still believe that the human involvement in a different way. I love the fact that an entrepreneur can generate a model in 10 minutes, 15 minutes instead of spending a whole weekend on it, right? Giles, I love that, but I do believe that it's really going to be important to check, to double check and to understand. I talk about there's a lot, modeling is just as much about the process as it is about the end result. It's the process that gets you smart and intelligent and insightful. And so here's the thing, as long as humans want to talk to humans to help make a decision, let's start at the board level. All companies have boards as long as the board of directors wants to speak with a human who's the CEO, that human is going to have to be able to generate and share insightful ideas.
(00:11:04):
And that senior person, the CEO is going to want to talk to other humans. And as long as that happens, everyone down the chain is going to have to have good knowledge, good insights. If you can't answer questions and know what's going on in a project, you're really in trouble. But you can use tools to allow you to work faster, build more quickly, more efficiently, but you got to make sure that they're working. Last thing I'll say here before I turn it back to you is for anyone who didn't see it, anthropic came out with a research report a couple weeks ago where they say years. That's like century
Host: Paul Barnhurst (00:11:34):
Ago.
Co-Host 2: Ian Schnoor (00:11:34):
That's like two years is like 500 years in the rest of our lives. But two weeks ago they came out with a report talking about usage, how people are using ai. And they said all AI usage is now characterised and kind of falls into one of two buckets, one of two categories. People are either using it for automation purposes or for augmentation purposes. Automation meaning do it for me. I know nothing about the topic. Write me an essay that I will hand in and I want to know nothing about it. And yes, of course we all know students and people that are using it to generate something for which they know nothing about and they want to know nothing about, but the report suggests that now more usage is coming in the form of augmentation. Meaning you're already a strong, you're an expert in the topic and you want to use it to make you better faster, to be able to iterate with your new friend, your new resource, your partner, they will help you, they will brainstorm with you. And I still think that as long as we have to have human involvement to make decisions, that augmentation is going to be very, very important. So yeah, it's a bit of an interesting scary time because things are changing very, very quickly, but I'm still of the mindset that you have to understand what the models are doing because you have to communicate them and look someone in the eyes and share what's going on with the models.
Host: Paul Barnhurst (00:12:56):
And I agree with you, I still think it's a human in the loop. It's AI assisted. I caution the scenario you share Giles of hey, just building it in the weekend only because are they going to be able to validate everything? What if it hallucinates without knowing some things, you run a pretty high risk of making mistake. So I think, hey, could you have someone check it? Are there audit tools? Which is where we'll get. So I think we'll get to the point where your scenario can work with even a higher degree, but I'm not quite there yet that I would tell the entrepreneur, Hey, just do it and don't have someone check it or some general ideas if they know the assumptions well could definitely help. So I'm kind of mixed still a little bit there, so I'm probably kind of a little middle on that, but still,
Co-Host 1: Giles Male (00:13:39):
And just to clarify my position, I think every entrepreneur should just go and do that, but if you have the choice of you've got no support and you don't know a modeler and I can use Claude and get something, it's actually becoming quite probably a viable option but not perfect.
Host: Paul Barnhurst (00:13:55):
And I could agree with that. It's a viable, I just kind of caution, there's a lot of things you got to be careful about still, right? And so why don't we instead of talking about it, how about we jump into testing and we can continue some of this conversation. So we'll turn it over to you here first. Giles, let me know when you have your screen up and we'll share that and we can talk a little bit about the eSport cases and then I have just one I'm going to do and then we will probably turn the last half over to N as he did quite a bit of testing today and we had been talking about that and excited to share what we're seeing. It feels like a watershed moment in this whole idea of vibe working, modeling that we're at a level where things are getting good.
Co-Host 1: Giles Male (00:14:38):
I just hopefully left the studio with the screen, so bear with me. I'm coming back.
Host: Paul Barnhurst (00:14:42):
All right, so we'll tell a joke while Giles works on his technical challenges. So does anyone know why a spreadsheet can't budget in any ideas?
Co-Host 1: Giles Male (00:14:53):
I think you said this joke when we did our 10 hour road trip, Paul, and I've forgotten it. You had other one,
Co-Host 2: Ian Schnoor (00:14:59):
It had that much of an impact on you, Gilles, or you've forgotten
Host: Paul Barnhurst (00:15:01):
It. I probably did share this one. I do share a fair number of Excel jokes, but the reason and I told it wrong, I just start again. Start again. I got to start over. We'll give you a give me why was the spreadsheet constipated as supposed to be the joke?
Co-Host 1: Giles Male (00:15:16):
Go on, go on.
Host: Paul Barnhurst (00:15:18):
Because it couldn't budget. Alright,
Co-Host 1: Giles Male (00:15:21):
So what I think we've all broadly agreed is we, we've shown any viewers, we've got the process of setting a prompt, pausing, we come back, we talk about it. I think we've all kind of moved on from that. So one of the things I tested in the meetup last week with Ken was this, and ironically I got AI to build this, but in financial modeling worlds quite often you have model maps where groups of people get on a whiteboard and they draw arrows and segments going from one place to another. So I just thought one of the interesting things that I tested for this was to put this into Claude the app and just say turn that into some sort of an executive summary for me, which is what it did. So this was the output from this and I just thought again, just the quality of what this stuff is starting to look like for me now is impressive.
(00:16:16):
The detail in the map wasn't particularly comprehensive or anything, but it is just good from one. My vision for this was you imagine how many days we would've spent as modelers in a room with people trying to extract information, you end up with tonnes of notes and whiteboards and pictures and somebody had to convert that into, okay, what's the scope of the project and the model and I think now you could literally just put all of those pictures into Claude or another tool potentially just go, just summarise it for me, put it into an output. So I thought that was quite an interesting little experiment that I hadn't done before. The other thing, where would you put it? 70,
Host: Paul Barnhurst (00:16:56):
80%. You said the details weren't really good. How much time do you think it's saving on the overall if you had to do this from scratch versus
Co-Host 1: Giles Male (00:17:04):
This way? No, I thought it was good. So if you look at the picture, I just meant the actual depth of what was on the map
Host: Paul Barnhurst (00:17:10):
And that's what I mean by depth. So would you say this is 70, 80% of the way there? 60? I'm just curious.
Co-Host 1: Giles Male (00:17:17):
I mean this is a pretty bog standard set of notes to get to a three statement model. It's fine. I think my point is that there'll be nuances in any project that are quite specific
Host: Paul Barnhurst (00:17:29):
Complex, you're always going to have to do some level of work to it.
Co-Host 1: Giles Male (00:17:32):
But I was impressed and that for me, you think of all the stuff we talked about with Ian Bennett and this idea that you've got a life cycle on a modeling project and everybody's been focused on the build. Where my head's been with certain parts of testing over the last month is like, okay, well what are we actually talking about? And one of the areas for me is that early phase of scoping and when you are in dialogue with people. So I thought that was an interesting test. The other things that I've done, again, I haven't actually brought both of them up, but I've got clawed in Excel. So for anyone that hasn't seen this yet, I now have my clawed beta add in Excel. You just go to add-ins here on the home tab of the ribbon and it'll be in there and this is what you get. So this is Claude in Excel, it's linked to 4.6, opus 4.6 and I use the same prompt that we saw before when we were doing the eSports challenges and it just smashed it. I mean again, it's still got issues where I think you could do this in simpler ways so we haven't completely overcome. Sometimes it does things in a way you wouldn't choose to do as a human, but this is pretty good, to be fair. Again that
Host: Paul Barnhurst (00:18:45):
Did it get everything? Did it score 1250 or were there any
Co-Host 1: Giles Male (00:18:48):
Yeah, I don't think I asked it to do the bonuses. So it
Host: Paul Barnhurst (00:18:51):
Got
Co-Host 1: Giles Male (00:18:52):
A thousand out of a thousand and that was fine and I didn't make everybody watch for the humble MVP case, but I think it's done pretty well from my side.
Host: Paul Barnhurst (00:19:01):
Okay. Exciting. Any other thoughts or anything else you want to share on your testing before we move on? To me and then in
Co-Host 1: Giles Male (00:19:08):
Only that it's still not perfect. So whilst I'm really impressed with Claude and I have yet to have a single one of those moments that I've talked about before where I get frustrated and that thought hits my head where I go, I could have done this faster myself. That's always like a line for me of like now I'm annoyed. I haven't had that yet with cla, but I think you two may have or you have Paul
Host: Paul Barnhurst (00:19:32):
I have. So I'll share my example. Before I do testing, I was working on my case a course I'm coming out with on data visualisation and I decided to test the agents and I asked it to build some charts, did it an Excels agent? It gave me charts, not great charts, but it gave me charts. Claude gave me the data on the same sheet in just a little box as all words. So basically tables and said, here's your chart. I said, no, that's not chart. I want a visual. Gave me the same thing again. I said, no, you're not understanding. I want charts in Excel. I want this to be visually represented with lines and colours. Said I did that and I finally just closed it and gave up. I couldn't get it to build a visual. Now was part of that just timing hallucination, bad luck, sure.
(00:20:19):
But that was one of my first tests, so I was like, what's everybody talking about? I'm not quite seeing it with Claude. Since then I've done some further testing that's been very impressive. So I've definitely had those moments that you mentioned where you're like, I just finally said I'm done. I can build these on my own. So what I wanted to show is I changed my prompt a little bit from before to add a little more detail and in particular what I asked it to do for this deferred revenue schedule. If y'all remember our audience, if you've seen any other episodes, I give it a deferred revenue schedule, has invoice, customer name, a start, an end date, number of months amount, and then some details about the customer and I ask it to build a schedule to recognise that over all the months. What I did with this prompt is I went a step further and I talked to it a little bit about using fairly simple formulas, but I also asked that give me an instructions sheet, build all the documentation.
(00:21:16):
I wanted to see if we could push it a little further. So let's kind of take a look. The first thing I'll mention is I've done this exercise using Claude inside Excel agent. I've done it with Claude, I ran it like six times twice with chat GPT 5.2 twice with Claude in Excel twice with open Claude 4.6. 4.6 was the most consistent, so the least variance in formulas even though there was some and the best outputs is what I've found so far. So here's what we got this time it linked to the front page. What I find is sometimes it links, sometimes it uses an index, sometimes it will hard code, but I'm finding that less and less when I tell it not the hard code. So it's getting better at following that rule. So that's all fine. Good formula here works, although I do find it interesting, it's divided by the months on the front sheet.
(00:22:10):
I probably would've rather had a formula here just because I think it's cleaner. I've seen it do it both ways. I've seen it use date D, the year and the month and so that's one of the biggest things still is just the variability in the formulas and functions it uses and sometimes complexity. That's been a little bit of a concern for me, but they'll get better at figuring that out with giving it instructions and all kinds of different things. But outside of that clean format, put the information at the bottom. I mean it's a usable file. I could go ahead and add more and probably have an update and not have much concern like real job if next month I got my next 50, I am confident I could get it to do that. Would it save me a tonne of time? Probably not because once you've built this, it's pretty easy to drag it all down if you have new ones.
(00:23:00):
But impressive. What I found almost as much if not more impressive is how good of a job it's doing on documentation. Again, how it decides to do this, documentation is different every time. What it includes is for key assumptions and how it thinks about things, but it's still pretty complete every time. Right here we have overview sheet layout. It tells you, hey, this row is hidden. What's row four doing? What are all these rows doing? Gives you all the formulas used and tells you the logic. Then it breaks down the main formula, the monthly revenue, it breaks that down into detail. Hey, why does the dollar sign matter? Hey, what are we doing with all these other columns? And so even talks about, hey, why we use conditional formatting and tells me what colour. It used a lot better documentation than I ever got when I worked in FP&A when somebody built a model. Now I might have to clean it up a little bit, but just think that alone. If you can get AI to when people don't do a good job of documenting you inherit a model, document it for me, document the mistakes. I was really impressed. I thought we've hit a new level, but what's interesting is across the board even 5.2, which I tested in Excel agent was a lot better than when we first tested Excel agent.
Co-Host 2: Ian Schnoor (00:24:17):
Do you have agent in your desktop? I just got agent on desktop today. So I have,
Host: Paul Barnhurst (00:24:23):
I do, I mean let me pull up just so you can see the differences. Let me pull up one I did just last week. An agent, this is with chat GPT. So let's just kind of compare so you can see a little bit, they'd hard coded in this example here, but then it did do the formula, it did date diff, so a little different but not much here. It decided to use date and then e date. Okay, fine. Did the, okay here, put the totals down at the bottom 90. Just need to expand those out. But what I found when it did the logic, compare what we got on the previous one, Claude writing it all out step by step here, okay, here's some basic information. Here's your formulas, here's your logic. One's much more presentation ready and kind of easier to read of the two.
(00:25:14):
But it did a good job and that was one of the examples. It did a little better job when I used Claude within Excel's agent. So I'll just share one of those real quick. Look at the difference that was using Claude with Excel's agent. So not using open Claude to build it. You can see how much better that was than the last one for instructions. Much better laid out as far as this part goes, no real difference. Use different formulas a little bit like it did hard code here, which I didn't like, but pretty similar between the two. So what I found is all of them were pretty close on the build. There was definitely a difference in the documentation. Claude was quite a bit better, at least in this example. And then Claude bombed when I tried to build some graphs. I need to try some more with it. In fairness, I don't know if it was just a day where a hiccup, we've all been there where you like why is nothing working? But those are my thoughts. Impressive on the whole Any questions?
Co-Host 2: Ian Schnoor (00:26:17):
Gosh, I love the documentation. I love the, again, it gets to the whole augmentation piece, right? Use it to make you get better. I mean we've all been in a, this is what still keeps me optimistic about humanity and about having continued role for ourselves. How many people can relate to this being on a project, working on a tool, working on a model, and you're struggling and it doesn't work and it's not balancing and you're fighting and you're fighting and you finally finish it at midnight and you get it done and you get the numbers done, but it takes so long, so many days or so many hours that what do you not have time for? You never have time for the documentation. That always takes a backseat, right? The documentation, the summary, the graphs, the charts, the communication around it. And I've always believed if we can speed up the act of the mechanical build, it should leave us a lot more time to think and to do higher value work.
(00:27:11):
Make sure it's documented, make sure it's recorded. Because I still believe for a long time I don't think Claude presenting to the board of directors anytime soon. I don't think Claude is standing in the room. I don't think we're going to see an avatar with no human involved still presenting to the managing team which goes up to the board. So as long as I think a human's doing that, I love the fact that you can get the work done faster and then focus on other things. So I don't know if you guys have been in that boat yourself or you were up till midnight doing something and then didn't have time for the other things. If
Host: Paul Barnhurst (00:27:42):
You've done any modeling at all, you've been up till midnight doing something before every budget
Co-Host 2: Ian Schnoor (00:27:47):
Season, which means that there are things that you never got to, right?
Host: Paul Barnhurst (00:27:50):
You were hoping you could wing it during the presentation or whatever. I mean maybe exaggerating a little bit, but we've all been there where you're like, okay, I finally got the number to tie out. I think I'm okay. Hopefully I can get through the presentation.
Co-Host 1: Giles Male (00:28:03):
I'm going to feel like this soon. As somebody that trains the same way that I feel about Power Query now, which is it's such a no brainer to me that when you work with teams who are doing manual stuff and they've never had a query and you're like, okay, well I mean it's obvious that you should be using this and by not using it, you are just causing yourselves a hell of a lot of stress and wasting a lot of your time. It feels like we're kind of at that point now where ai, whether it's copilot or Claude or something else, we're very quickly going to get to the point where it's like, okay, you are seriously missing out. If you're not using this intelligently, you can still use it badly and that's not going to save you time, but it's getting to the point where you can use it in lots of ways. Like you said, to save yourself time to not be up till midnight. And if you're not doing that and you're not using Power Query, this is a double whammy during finance and you're not using AI or Power Query, you've got to get up to speed.
Host: Paul Barnhurst (00:29:00):
Funny you said that because last week I presented to about 80 portfolio companies for backed companies on Excel agent. One thing I said is all of you should at least be testing at this point, select a tool and start using it. It is rapidly getting better. Is it fully there yet? No. And I still believe you would get more benefit from learning the foundations first, but if you know what you're doing, start using it. So my advice to all of 'em was get a tool. There is benefit today but don't expect it to. It's still not press a button be done and kick your feet up and turn it into the boss. For me,
Co-Host 2: Ian Schnoor (00:29:40):
The exciting part is I agree, embrace it, test it, use it. Don't fear it. But again, I feel like it's not just about saving you time. Yeah, I mean I'm about to show you that it built a pretty good model in 15 minutes, but like I say, use the use your new free found time to do other things. Generate some additional supporting documentation, generate some research, come better prepared, learn some additional ancillary work. Do another model. Do something that makes you look like a real superstar when you have to present. Get it to make you better as a professional, as a person, make you smarter. And that's where I'm excited and I think those who embrace that will do well for the foreseeable future. Those who are afraid to or won't embrace it are going to have trouble sticking around staying on the sled. So that's kind of where my thinking is at, as is in my head, the major change is if and when the day comes that humans are not required to make critical decisions that could happen. Let's be clear that might happen, but if companies are completely run on autopilot with autopilot boards and autopilot CEOs all by computer, then it's a whole new world that could happen sooner rather than some people think pretty soon, but I don't think we're quite there yet.
Host: Paul Barnhurst (00:31:00):
When we're there, we're all out of jobs, so we'll be coming on the show and talking about how we like the breadlines.
Co-Host 2: Ian Schnoor (00:31:05):
We're just going to hang that all together. We're going to go find Giles in his truck and we're just going to join me and we're going to do daily Excel on the road episodes. No one will care because no one's using Excel and we will wonder why are you talking about Excel? The computers use Excel, but
Co-Host 1: Giles Male (00:31:22):
I've got two spare beds in the camp event. That's all I'm saying.
Co-Host 2: Ian Schnoor (00:31:25):
Absolutely.
Host: Paul Barnhurst (00:31:26):
Keep that in mind.
Co-Host 2: Ian Schnoor (00:31:27):
Absolutely.
Host: Paul Barnhurst (00:31:28):
Would you like to share your screen and share?
Co-Host 2: Ian Schnoor (00:31:30):
I would happily share my screen and walk through what I did. Now I knew where is the screen share here is, lemme get to this here to tell another
Host: Paul Barnhurst (00:31:40):
Terrible joke while we wait.
Co-Host 2: Ian Schnoor (00:31:42):
Yeah, I
Host: Paul Barnhurst (00:31:42):
Know. Please, please. You had one about gold diggers that I remember was what do cell references in gold diggers have in common
Co-Host 2: Ian Schnoor (00:31:52):
Sell references and gold diggers. My goodness.
Host: Paul Barnhurst (00:31:56):
If you want them to stay in place, you have to throw some money at them.
Co-Host 1: Giles Male (00:32:02):
I thought that was good road trip. I thought that was his best joke. That's
Co-Host 2: Ian Schnoor (00:32:06):
A good one. That's not bad. Yeah, that's right. I was going to say if the editing team is watching, I'm okay if you cut this. Just kidding. Let's move on then. Let's keep it real. Here is what I did this morning, knowing that Claude was going to perform pretty well, I asked it to build a three statement model, which is exactly what's required in the A FM programme on the A FM exam. On the A FM exam, people are given a case study, but what I did is I copied for this a FM test, I copied the case study into a sheet so it looks messy, but normally you're given it as a PDF that looks nice, but I just copied the data into the case here. You can see the case provides information about, this is the Henderson company. It's one that we have publicly available often it's a manufacturer of storage tanks.
(00:32:55):
We've been given three years of historical data most recently 2025. And then it says you've been asked to build a financial model for 20 26, 20 30, and it provides information about the sales. It's all buried. Here's the thing, it's all embedded. It's all just embedded into one sell. It's not in different sales, it's in one piece of text. Average selling prices for the next five years because that's how they would normally get it in A PDF, the capacity of the factory, the growth rates. We provided operating costs, variable and fixed and I just have a text label that shows that the variable costs are costs per unit, fixed costs are in aggregate dollars, millions. These are actual numbers. And then it tells you in a piece in a string that the sg and a, the selling general admin costs are going to be 3.9 million. We've got some information on the CapEx and on the working cap all embedded into pieces of text, income taxes, debt equity, other assumptions, and then there's instructions that say build a five year forecast model.
(00:33:54):
And then the other sheet that I gave to Claude, which is what people get in real life is this three years of historical financial statements, the income statement, the cashflow statement and the balance sheet and that's it. And then what I did, and I'm going to show you my prompting, I'm going to show you my prompts, you can see it on the screen. This is what I did. I went to Claude and said on the case tab you've been provided with information about a company called Henderson. You've also been provided with three years of historical financial statements. Build a five-year forecast model with all the required schedules. Now I told that the schedules, I had to know that it needs a revenue schedule, cost depreciation, tax working, capital debt and equity. So I knew that and said, oh by the way, there are instructions for you on the case sheet in B 60 to B 66.
(00:34:39):
And then my mind was a bit blown, I'm not going to lie. It started saying, okay, it actually did things that the other agents didn't do. It started saying, okay, I got it. I'll start by reading all the data from both sheets and then it started saying, I'm going to read this, I'm going to read that. Then it said, let me read the exact self, let me understand the self positioning. Okay, I got it. Now I have an understanding of the data. And it literally went down and said, I've got the data. Let me look at some historical metrics. And I was watching it. It took about two minutes or three minutes to do all this. Say, here's my plan, I'm going to do it, structure it with an assumption sheet. It's new. I'm going to give you a base best and worst. I told it to build some scenarios with a base best and worst.
(00:35:18):
I didn't tell it which variables to use. I said, choose, use your judgement and choose reasonable variables that we should build for a base best and worst. That's all I said. It said, I'm going to build these schedules, these are the schedules. What do you think? And it said, these are my key assumptions that I'm going to use. It said, I am going to build scenarios to have a best and a worst case around sales price, sales volume and inflation. Now that kind of blew my mind because that's what we do. That's how we did the, I didn't tell it to do that. I just said choose three variables and it decided it would go plus and minus $50 for the sales price plus and minus 1% for volume. And it told me what it would use for inflation and then it said, Hey, what do you think?
(00:36:01):
Should I proceed with this plan or would you like any adjustments? I'm like, ah, no, that's good, that's good. So I literally said, please proceed. And then it went on for another 10 minutes. It took about 10 more minutes. So 15 in total, I was watching it was talking to itself. It was building and talking to itself, building and talking to itself and I will show you and it was resizing. It says, now let me build this tab. Now let me, I'm not going to go through all of it now let me do this now I'm doing that. And it walked me through what it was doing step by step all the way down. This is all the work it did and it did a lot. It did a lot. I copied it out here so you can see it. It did a lot. It showed me everything it did and then it stopped and when I was stopped, I went back to the model.
(00:36:46):
Oops, this is the next one. And what it had built was this. It had, remember it started from the case. I'll get rid of the Claude prompt here. It put an assumption page. It's not the prettiest assumption page that you'll ever see, but pretty good. It would be very easy to clean it up. It used blue as the default font. It's got inflation, it's got tax rates, it's got the differential between government and accounting earnings. It has a sales volume. This sell interested me. This sell fascinated me. The historical sales volume because that's not in the case. We did not give it that. In order to derive that, you have to calculate that based on the revenue data and the pricing data in the historical year. So it's put a number in here. It obviously built a calculation and stuck a value and now it's very, very close to what we had.
(00:37:38):
It's 99% of the way there, but it figured that out. It's got sales volumes, it's got gross sales prices, it's got all the assumptions here looking pretty good CapEx. It made reasonable working capital assumptions and then it built the scenarios here. Look what it did. It showed for the base case. It's got the sales price, sales volume and inflation for the base case. It figured out to pull these numbers out of that one cell. It was typed in as a sentence. It pulled them out into a row for the base. It figured out to add $50 for the best case and then to take off $50 for the worst case. So my point is I can now see that on a few occasions it has made calculations like this one here and it has made calculations around a best case and a worst case, but just typed in the numbers so I know that it calculated it, it entered the value, fine.
(00:38:30):
Then it's deriving the live case by using a simple if statement. So probably not the function that most people would use. It's using a simple if statement and looking at B seven. B seven is the switch and it's used a very simple if statement again, not a big deal, it's working. I was again pretty blown away. That's the assumptions. And then the model up is here. Again, the formatting is not perfect, but the functionality is excellent. It's got links, so it built revenue schedule and it linked, it built a cost schedule and it linked it. The net income is very, very close to the net incomes that we have in our solution. Didn't love this. It's still using blue. The net income is linked, but it did not apply formula colouring black, it's using blue but still. So this is something I'd have to fix, but it's got the right link. It's doing it properly. It's got the cash flows and you can see there's some rows here where the formatting's not ideal. I'll come back to that. The balance sheet is built up. It's mostly links. It's working well, not as beautiful as we would like, but it's actually working and it's balanced. You can see it is balanced every single year and it does not have a plug. Unlike the other tool. We tested Jes. So
Host: Paul Barnhurst (00:39:44):
On the bank debt revolver, is that working properly?
Co-Host 2: Ian Schnoor (00:39:47):
I'll show you where it's working. It is. So I was wondering where it was. So look at this though. Look at this guy. It's amazing. It figured out to do a revenue schedule. It took the historical volumes and then it took the volume. It's made this green. When you're linking to assumptions, it used green. It understood that there's a gross selling price, a net for freight and warehousing, and then it calculated the net revenue and then it calculated the revenue as the gross revenue as gross price times the sales volume. It calculated the freight and warehousing. It did this beautifully. What I was most impressed is it actually understood how to calculate the costs variable and fixed it understood that a variable cost has to grow at inflation on a unit basis so that these two variable costs are growing at inflation on a unit basis.
(00:40:41):
And then it calculated the total variable cost here by taking the total variable cost per unit, multiplying it by the number of units. That blew my mind because that was just stuck into some text and in a table and it arrived at total cost. So it did a great job here again, I don't love that it's linking back, but these are fixable things that it's building formulas linking to another sheet, but it's unbelievable. It's unbelievable what it's done here. Working capital, it made some reasonable assumptions. It's built the formulas and then I'll show you what it did on debt. On the debt, it has the term loan at the very top. It actually built in functionality so that it doesn't overpay the debt. It has the revolver. Now, I was curious because when I came to the revolver, the revolver, the important revolver calculation is actually coming from above, so I'll show that to you, but it's doing interest.
(00:41:34):
I actually didn't even look at this yet. It calculated interest on the average. It's calculating and we know that that is a common methodology of calculating interest. I'm actually trying to see where it did. Oh, the interest on the revolver. Oh wow. It calculated interest on the average for the term loan, which it knows is okay, that will never create circularity, but it calculated interest on the opening balance for the revolver in order to avoid circularity. I think. And it did the same thing for the cash balance and what it did with the revolver is the revolver. It's linking it above to row 58, and so where it did the work, Paul was here, so I don't love that, but it's fine. It built a big formula on the cashflow statement. On the cashflow statement, it used an if statement. No one's going to love the way this looks, but it's actually generally directionally fine.
(00:42:30):
We like to kind of break everything out step by step, but it's basically saying if all these cash flows together are less than zero, then we have to borrow. Otherwise it will repay and it appears to be working. I did not dissect it perfectly, but as I say, I was pretty blown away. This took 15 minutes. I could spend another hour or half an hour and clean it up and check it. I'm checking one more thing here. I asked it to set up, one of the instructions was to set up print ranges. It did not do that. It did not set up headers and footers and print ranges even though we asked it to, but I think you would agree. I'm going to stop and let you guys comment on what you see. This is nothing like what we have seen in prior tools, guys.
Co-Host 1: Giles Male (00:43:11):
So yeah, I saw something very similar. I said I kind of tested it on Henderson as well. The only difference I had in which I dunno whether you tried was I just imported the PDF and it did a similar job just importing all the information from the PDF rather than putting it explicitly in the model.
Co-Host 2: Ian Schnoor (00:43:29):
I did that with the CCF M, the second level.
Co-Host 1: Giles Male (00:43:32):
Yeah, and it does it, isn't it?
Co-Host 2: Ian Schnoor (00:43:36):
Let me just show everybody. I mean if I open up the club,
Co-Host 1: Giles Male (00:43:38):
The other observation I was just going to say was even though we know these tools aren't understanding in the same conscious way, like a human understands the degree to which it's getting some of those modeling nuances, right? Like the min for the amortisation payments for the debt and stuff like that. It just feels like it's at another level of getting these things right? That's like a modeler's brain coming across in the output, which is amazing.
Co-Host 2: Ian Schnoor (00:44:04):
Yeah, exactly. Here where most people wouldn't do this, most people would just have a $25 million amortisation. It built in the backstop to say, well wait a second, it should only be the minimum of that 25 or the beginning balance here because we never want to overpay. So yeah, I agree with you. It was pretty mind blowing and as I said, the five-year forecast was pretty much spot on, very similar to what we did in ours. Paul, what are your thoughts about this?
Host: Paul Barnhurst (00:44:33):
Yeah, I mean it's very impressive. The testing I did on Henderson was more, Hey, just take the three years of data and build me a model, and it did good. I did it with a different tool that was using Claude 4.6 and it did a really good job. This is even better than I would've expected from what I've seen. But yeah, it's incredible how far it's come.
Co-Host 2: Ian Schnoor (00:44:53):
I mean, it's unbelievable. As we go down the balance sheet, as we go down, it adheres to all the principles we talk about, right? Just links, no big formulas, right? It understood how to link up cash and cash equivalents to the cash flow statement. It understand how to link receivables, inventory and prepaids to the working capital Down below it is just flatlining prepaid expenses. It knows that that's okay. It's got a total here. It's calculating the net pp e directly in the cell, which is just fine and it's other, right? That's perfect. It is linking the bank revolver to the debt schedule below. Same with the payables. So everything is, I knew that deferred taxes has got to be cumulative, but again, I'm recognising that I'm able to check this and understand it because I know modeling, and by the way, so again, I love this.
(00:45:42):
I love it, can do it, but just so everyone who's watching understands, I'm not suggesting you shouldn't know modeling. You need to because you still need to understand what it's doing. I'll tell the quick story to the viewers here. I was talking just by cheer coincidence two days ago with the head of finance at a big finance organisation here where I'm based, and he has been an early adopter of Claude. He was on the beta group, so he's been using Claude in Excel for many months since October, and he's been putting hours and hours into building and using it and refining it. And I asked him point blank, I said, would you agree with this statement? Do you agree that the only reason you are able to use Claude so effectively to build models is because you already know how to build models yourself? And he said, 100%, I'm only able to do this.
(00:46:33):
I already understand and know model building. He said, oh my gosh, the idea that someone is going to do this who does not understand modeling is scary and is going to lead to issues and problems because I promise you, anyone watching, I promise you, even if you're going to use Claude to get you this far, you will be, you guys can agree with this. You will be in a boardroom one day with a client sitting right beside you and the client's going to say, change this, modify that. Can you add this in? Can you change that? And you might be offline, the internet might go down, which happens all the time, or you might not have the ability for Claude to spend five or 10 minutes. You might have to make very quick changes on the fly and you will look very, very deficient and cause a lot of concern if you can't do it. So I really think it's important to understand what's going on here, but as I said, was blown away by this. So let's, any other comments and then I'll quickly show the CCF M case. What do you guys, anything else?
Co-Host 1: Giles Male (00:47:27):
I'm ready to see CFM. I'm invested now.
Co-Host 2: Ian Schnoor (00:47:30):
I
Co-Host 1: Giles Male (00:47:30):
Think I'm ready to take the CFM
Co-Host 2: Ian Schnoor (00:47:32):
Now if I can use
Host: Paul Barnhurst (00:47:33):
Quad.
Co-Host 2: Ian Schnoor (00:47:34):
Yeah, yeah, you'll be able to, so exactly. Well then again, just before we even wrap, I mean even so impressed it understood it understood from one sentence in our case study that there's an earnings differential between the EVT that gets reported on the accounting financial statements, which is this earning row. And what we tell it in the case is that we say assumes very simplistically that earnings for government purposes are going to be $5 million less. That's all we say, right? I think it's right here. We simply say, and for a lot of people, this doesn't make, they don't understand this. We tell people the tax rate is 35% and we say the aggregate reduction in government pre-tax earnings due to timing differences is expected to be 5 million per year for the next five years. We don't give it any more instruction. We just said that, and yet it understood, oh, okay, if government earnings is going to be $5 million less, then I guess it understood that the accounting tax is going to be the earnings times the tax rate.
(00:48:33):
It also put a max function there, which is actually not correct because that value can go negative. An accounting tax number can go negative, whereas it's forcing it to stay at zero. So that's actually not correct. You can have, and I know this because I know mata, you can have negative 45 million of accounting earnings, which should display a negative total tax number on your financial statements. So that's a mistake. I just saw that Claude shouldn't be doing that. But again, that's because I understand how this is working, but it did know that this one, it should have restricted, it understood that the differential, that the deferred tax expense would be the tax rate multiplied by the earnings differential. That number can't ever go negative. It then took the difference to get to the cash tax number, the current tax, this value technically can't go negative, so it restricted this one.
(00:49:23):
That's fine. It put a maximum function here, but there is a small subtle error that this one should not have been restricted. So in theory, if this was negative 45, then in theory, I'm just doing this on the fly, the actual, the accounting expense should have been whatever it was, the current number would've been zero, but the deferred tax should have been the differential. So it's not quite there. This should have been a negative deferred tax expense, which would've been reducing a liability or perhaps impacting an asset. But so that's not quite there, but it's still very, very impressive. So let's move on then. You guys want to see what it did for, again, it actually scares me a little bit. It scares me a little bit because I know that there are some people that will rely on this because they, oh my gosh, it's so good. And they'll miss very important subtle errors like that that are built into the file. Okay. I'm going to catch a breath and then move on. And are you guys ready to see here what happened? Are you still there? You said you were at 50% earlier. Are you still feeling okay, John? Is this,
Co-Host 1: Giles Male (00:50:26):
Oh, there's nothing like a good debt case study to bring my energy levels up from 40 to 90.
Co-Host 2: Ian Schnoor (00:50:33):
Let's go. I mean, we already told you that Paul and I combined are usually less than 50 anyway, so as long as you're still hanging at 50, we will be good. For those who are unfamiliar with the CFM programme, it's the second level of accreditation. This gets pretty complex. There are 12 topics. One of the topics is debt. And if a debt question is on the exam, people are required to build a pretty complex debt schedule. Now, this is one of the easier ones. The more current ones have been even a lot more difficult than this. This is still, but this, a lot of people, if you've never modeled debt before, this is still going to cause you, anyone who's never done this before, this is going to cause some issues. This is the Black Rat case and J to your point for this case, I did not copy the data into Excel.
(00:51:17):
I just gave it the word doc. I just said, Hey, take this Word doc. And the word doc says that black rat is a special purpose vehicle to operate a 10 year project beginning on December 31st, 2018 for 10 years ending 2028. Three sources of financing. The three sources of financing are senior debt, junior debt, and equity. Use the following information and the data in the case to complete these tasks and answer the questions below. So then what we do in the case is we tell people about the senior debt. It's determined their repayment is determined based on a specific debt service coverage ratio. So you have to maintain a debt service coverage ratio, which means that there's debt sculpting taking place here. Repayments are made at the end of every quarter. So this is a subtle nod to say, yeah, we told you it's a 10 year model, but you got to do something at the end of each quarter so a human would know that they have to model quarterly.
(00:52:12):
Did Claude know? We'll find out. Interest is paid quarterly and there's an opening balance. The debt service coverage ratio. The debt service coverage ratio is based on this formula, this calculation. There's a debt service reserve account, which is pre-funded that earns interest. It talks about the junior debt that gets repaid in 40 equal repayments. And then it talks about the equity, how the equity works, quarterly cashflow is received and ebitda. It walks through how we calculate the equity cashflow. That's it. And then there's a bunch of tasks, build schedules to track the different pieces of debt and it walks through it step by step with a lot of questions. Which scenario gives the highest IRR, which scenario gives the lowest NPV? What's the total interest rate on the junior facility? So there's a bunch of questions and so on an exam, candidates would've received this and they would've received, they would've received this. This is what you would've gotten on the exam. You would've been given this sheet that said, here's some information about the senior facility, the sculpting and interest rate scenarios, the debt service reserve account, the junior, the equity you're given all the data you need, and then a range of EBITDA profiles. Okay,
(00:53:27):
That's all that was provided in the worksheet. And on the exam, people would build a second sheet and build up schedules to calculate all the pieces of debt. Now, before I show you what it did, I'm also going to show you my prompts. So what I did is I said, using the Black GRAT case study in the Word document and the worksheet file, build a forecast for all the pieces of debt. All the information you need is in the case and the Excel file. Please answer the questions on page two. That's it. It said okay. It did the same thing. Well, actually it was faster. This whole case it did in 10 minutes, which was faster than the, it's a lot less rows. The Henderson one is three or 400 rows. This one is about a hundred. It said, I'll start by reviewing it. This is what it's going to do.
(00:54:10):
It told me that it was going to parse all the assumptions, create a new sheet. It said it's going to build schedules for the different pieces of debt. It was just kind of cranking through this. Then it said, let me start building it. Created a model sheet. It verified it. It kept going. It was talking to itself. It showed me every step along the way, and you did a lot. You can see it. And I love your point, Paul. I love this documentation. So I can see I know what it did. It's walking me through it. It's a lot, but it's clear, it's logical, and it's describing to me what it did step by step, and I will show you how far it got to. It got down to an IRR, it got down to an NPV. It did some formatting and then it was done.
(00:54:51):
Let's take a look at what it generated it. Is this the one here? Yeah. So it started with this worksheet file and this is what it built. And I have to tell you, it looks almost exactly the same as our solution file. It understood that it had to build the model from 2018 and to do it quarterly for four quarters. It understood that the ebitda, it use an index function. It understood that it would use an index function to pull the ebitda. It built sections for the debt service reserve account for the senior facility. It understood that the repayment, this is a critical formula. I won't walk through it. It understood that the repayment on the senior facility was going to be based on maintaining a sculpting ratio here, and then it's going on to check the debt service coverage ratio. So that's why the repayment is different every quarter.
(00:55:51):
It's calculating the repayment. This is one of the hard parts of the case based on maintaining a certain coverage ratio. So it did that correctly. It calculated the junior facility, it calculated the equity and it did it, and it got all the way down and it calculated the, it told us which quarter it would be able to repay the senior facility. It told us the total amount of junior repaid. That was one of our questions. And it arrived in IRR. It used the IRR function and it arrived, it knew it understood that it was a quarterly model and it calculated the IR is 10.16. What value did we get in our solution? 10.15. So it was basically not the right answer, but here's a problem. So this is what scares me. Now, I guess this could happen in real life too. It calculated an NPV, it calculated the NPV.
(00:56:41):
It used Excel's NPV function. I looked at it. The NPV function. The NPV function wants to know the discount rate and then the cash flows. So it looked at a discount rate. It took the discount rate here, which is an annual discount rate. It understood that that discount rate was an annual rate, so it divided by four, okay? And then it looked at these cash flows. The problem is it didn't understand, I understand because I've done this and I teach finance, I understand that the NPV function assumes cash flows are annual cash flows. So what it's doing is it's using a quarterly discount rate, but this stream of cash flows is being treated as if it was done over 40 years. That's a tiny error that had a major, major impact. This is what the decision would get based on. Your boss would make a decision based on the NPV and if you said the N PVS $80 million, or maybe that doesn't cut it or doesn't jive, actually the NPV should be calculated with an XNPV function or understanding that the time periods to say the discount rate is up there, these are the cash flows, and then these are the dates.
So anyway,
Co-Host 1: Giles Male (00:57:52):
I agree, and I'm guessing project finance like this probably won't be the area that starts fully leaning on AI modeling first. You'd hope not. Anyway, for that new bridge,
Co-Host 2: Ian Schnoor (00:58:00):
Any senior bank, any senior finance professionals I know are pretty particular, pretty anal, and they like things to be checked and they like people to know what they're doing. Listen, it's going to happen. But I agree with you. Maybe not tomorrow. Anyway, guys, that was great. That was fun. I don't know if you want to wrap things up or Paul. Yeah,
Host: Paul Barnhurst :
Let's go ahead. I think we wrap up here. I hope our audience is seeing that. The reality is we just have a leap forward. If you are not experimenting with agents, if you know what you're doing and if you're listening to the show, you probably know modeling well, it's time to start. It can save you time. As Giles mentioned earlier, I don't remember if you on the show or not, but he's upgraded his account to a max or one of the bigger ones for Claude. He's using it all the time. I mean, IN'S mentioned he's been wowed and super impressed. I'm finding more and more use cases. We would encourage you to jump on board, but don't advocate everything to the computer. That would be kind of my thoughts. It's getting better. Still going to make mistakes. It's still human in the loop, but use AI so you can focus more and more on the strategy. Start to find the right use cases. A really complex project finance model is probably not the first place to start. In fact, I'll say it's not, but there are many cases where you can get a lot of benefit. Last thoughts from you in, and then we'll give Giles the last word.
Co-Host 2: Ian Schnoor :
Yeah, I've said a lot. I'll just keep it brief. I mean, I agree with you, Paul. This stuff is mind bending. It's unbelievable. It's doing some great work, but stay ahead of the curve, play with it, and continue to elevate your own skills. I really believe that that will lead to success.
Co-Host 1: Giles Male (00:58:23):
And I'll be quick as well. I agree with both of you. I think we've got to start playing with it. One of the things, for example, for me in the Claude app is you've got this concept of projects, so you can add additional context within a project area, which is a bit like I think in chat GPT world, creating your own. So these are these nuances that I'm finding where if you learn how the tool works, like Claude specifically, you are actually then learning how to get better answers back from the tool. So it is now a case I think of really start playing with it. Learn how to use it properly, but be cautious.
Host: Paul Barnhurst (00:58:00):
It's amazing. Go out and learn. We will have another episode coming up in another month. We have an exciting guest for you there, and I'm sure we'll do some more testing in the future. But wow, just think of the journey. Those of you who watched our first episode to today, it feels night and day. It doesn't feel like six months. It feels like five years, 10 years of progress. So thanks guys. Thanks
