We have 60 minutes, so it's not a lot of time, but it's just enough to get us through our agenda for this webinar. So a quick introduction of myself. My name is Juan Cabrera.

I'm a finance instructor, for Financial Edge and Wall Street Prep.

I cover mostly the investment banking division of our business, as a trainer.

And, that's where I started my career, in banking.

And then I moved on to doing a quantitative finance PhD, and I slowly got into consulting and training as well. So, I have been doing a lot of work as you would expect, right, on using AI in financial workflows. Right? Given that clients want that, they expect that as well.

So our session today, as you can see on the screen, is Financial Modeling using Claude. All right? So Claude is our one tool for today.

There is a bit of onboarding here, but it really depends on how you want to follow on this session. Right? So I am going to be using Claude, of course.

To be exact, I'm going to be using Claude Cowork as opposed to Claude Chat. Okay? You might have access, depending on what kind of license you are using, you might have access only to Chat or to both. Okay? If you have access to both, I strongly recommend that you follow the session using Cowork.

But if you're using only Chat, then you're going to have to adjust a little bit.

Right? And I'm happy to give you some quick tips on how are things different. Just know that the way we prompt, and the way we sort of navigate Claude when we're using Chat, is different, right? In a way, you might say it's a bit more manual than if you're using Cowork.

Okay? In any case, we'll discuss this.

It is, as I said, only 60 minutes, right? So not a lot of time to be dealing with the tech issues that you are facing, right? Worst case scenario, feel free to just follow on the session without doing the work. Right? This is getting recorded, so you have access to this in the future. So maybe you can sort of look at the video later and do the work then. Okay? Whichever way you prefer. Very quickly, we have a Q&A window where feel free to put your questions there. Okay? You can also directly message me on the chat as a couple of you already did.

Just type your question there. Honestly, I will be answering your questions verbally. Right? So when I see them pop up in front of me, I will address your questions. Okay? Great.

So let me get started here. Let me start with this.

This is our agenda for the day. Okay? So it's called Financial Modeling with Claude, right? We're going to be doing four things in these next 56 minutes we have left.

The first one is we're going to set up our workspace.

So I'll spend maybe 5 to 10 minutes on that one. Okay? Then we're going to be moving to our first block of actual modeling, which is building a three-statement model.

After that, we are going to be building a WACC or cost of capital estimate. And then the last bit is, once we have the cost of capital estimate, we're going to be building a DCF valuation model. Okay? And that's really going to take us through that entire hour. Okay? Now, you might be thinking, well, that's a lot of steps.

Yeah, part of it is that we're doing modeling, and we can do modeling across different tasks or different exercises, like three-statement models and DCF models.

But as we go through this, we're going to have a different approach. So each of these blocks in the session will show you a different approach on how to interact with Claude, how to build financial models with Claude.

So very quickly, you can see here on the three-statement model here, we're going to start by using a prompt, then we're going to iterate, and then based on our chat history, we're going to build the skill. Okay? So this is more of a hands-on, not really automated process. We're actually going to build the skill from scratch. You might say that. Okay? That's the first bit. Okay? And then on the second bit, when we look at the cost of capital estimate, we're going to now take a skill that is already built, and we're going to run it. Okay? So it's obviously going to be a little quicker.

And then the last bit, the DCF valuation bit, is going to show us two things. One is how to use specialized skills. That means these are smaller skills that have a very narrow scope and that we can then chain link. Okay? Which is a very strong best practice, right? To have very specialized skills that you can chain link.

A DCF valuation model is a perfect example of a place where you can apply this method. Right? And the other thing about DCF valuation is that we are going to actually be jumping on Excel and using Claude as an add-in. So when we look at the three-statement model and the cost of capital, we're going to use the Claude App.

And when we get to DCF valuation, we're going to be full on Excel. Okay? Great.

As I said before, and I'm going to repeat it just to make sure you are aware of this, any questions you have at any time, just put them on the Q&A window, put them on the chat. Okay, great.

So, let's start with the first bit here, number one, which is setting up the workspace.

Okay, so I have a slide for everything, okay? But that's what we're going to be doing.

So if you're doing this with me, let's jump in together.

To me, this is the main step.

Whenever you are working on a deal, whenever you're working on a project, whenever you're working on a task, this really is the step that should consume your brain power. It should consume a lot of your effort.

It should consume a lot of your time because if you get this set up properly, the rest of the work, the actual work, is going to be more efficient in terms of speed, but also much higher quality, right? So a lot of people think, oh, this setting up your workspace is just create a project, right? It's a lot more than that, because you got to do it very well. And the other thing I should say is that these workspaces are not static tools. They are living tools. You should update them as you learn more about your deal or project, okay? So we're going to do three things here.

We're going to create the project in Claude. We're going to use a template today, and then once we add the template, or make the template available, I should say, we're going to paste project instructions.

Of course, because of the nature of the webinar, it's only an hour, a lot of this I have pre-built, but you might have to do this from scratch, and build these kind of things from scratch. So, we're going to see plenty of examples.

A quick question, are the slides going to be shared after the session? I can share the slides right now. So let me put them on the chat. Okay, I should be able to add the slides. Just give me a moment.

All right. So I have just now sent out the slides on the chat.

I will hope you can download the slides from the chat.

Please confirm that you can. Otherwise, we can always make them available after the session. Okay, great.

So let's start by jumping onto Claude.

And at the beginning, I'm going to be focusing a little bit on Claude, and then as we move through the hour, I'm going to be focusing more on the model building task, okay? So here's my Claude.

It's ready to work. So let me start with this simple thing.

I'm working with the Claude app right now.

And you can see here I have two options, chat and co-work.

Some enterprise licenses don't have the co-work option for obvious reasons.

Thank you for confirming that.

So, again, I'm going to do everything on co-work.

It's a great way to do work. And let me tell you the key difference.

Co-work allows you to link local folders directly to Claude. So you can have a project built, and that project can be directly linked to a folder in your local drive. And you can have Claude then read and write onto those folders. So file management it's just the easiest thing, okay? If you're doing chat, then you're going to have to upload files manually onto the project folder, and then you have to delete files because you don't want to overload your context window.

Okay? So let me start by creating a new project.

So I'm going to go up here where it says Projects, click on Projects, and then I'm going to type New Project.

I have a few things here that I've done before, things that I probably have to clean up myself. Okay. And here we are.

Now, I'm going to call this project Nike Valuation and then Live. And the reason I call it Nike Valuation Live is because our case study company for this session is Nike.

It is technically a valuation kind of modeling session because we're ending up with a DCF at the end, a discounted cash flow analysis, and we're doing this live. Okay? So I will type here a quick description, build a three statement model, and let me type this, WACC and DCF for Nike.

It's just a description so that when you look at the project, you can see what the project is about in the future, okay? It's not a big deal.

Now, here is one of the key steps.

One of the key steps is that if you're in co-work, or at least in general, projects allow you to use a folder. So I've already, and I'm going to show you this, I've already created a folder. It's right here.

You can see here the folder listed. It says 04.

I like to number my folders.

04_Nike_Valuation_Webinar_Live.

That's the folder that is going to have all of my work inside.

So I'm going to link that folder. That's what I like to refer to as my root folder, the root folder for that project. Okay? So let me go here.

Oops, I clicked away from it, but that's great. Let me try again.

Nike Valuation Live, build a 3SM WACC and DCF model for Nike. Okay, great. I'm going to click on Use a Folder.

I'm going to do a bit of work finding the folder. There it is. That's my folder.

Okay, I'm going to click on Select Folder.

It's going to say, "Hey, do you want Claude to change the files in this folder?" And I'm like, "Yes, that's exactly the reason I have that folder created." So always allow.

Okay? And that's it. The folder is being linked. That's me, that's the folder.

I can now create a project, and I am ready to work.

Now, of course, I'm not done setting up the workspace, but I now have a canvas to work from. Okay? The first thing I want you to be aware of is that every chat that I create within this project has access to that folder.

Okay? In fact, you can see here on the right side, there is basically all of my context. This is everything that really goes into my project on the right, and you see folder here and you see the folder link. Okay? Quick question here. Will we gain access to this folder? So, the folder is something you want to create on your own local drive.

You create it, then when you create your project, you link it to the folder, and now Claude has full access to that folder and only that folder.

And it can write files to it, it can read files from it.

Now, why do I want to read files from it? Because if you look at the folder here, that's the folder.

That's just any folder. I just create a new folder, right? Double-click and look what's inside.

Okay, maybe this thing shouldn't be here. Let me move it out, okay? I don't want to confuse you here.

Okay, that's the only file that is inside my folder. And what is it? Let me open it up. Okay? It's this. I already opened it, actually.

It's our template. Now, working with templates when you are doing financial modeling, it's great for multiple reasons. But let me start with the reason that most people, I think, care about, and that is token efficiency.

If you have templates, the template is going to provide Claude with a lot of context and instructions that would guide Claude and get Claude to the output faster and more efficiently without searching as much, without making as many assumptions, without asking as many questions, and it would allow Claude to simply generate the output faster and better.

Okay, so it's just better. So, I have this template here, which is our Financial Edge style.

Well, what about it? It's got the colors of Financial Edge.

It has some information about the model we're about to build.

Okay? It's got, if you notice, the 3SM tab, it's basically empty. But guess what? Claude now will know where to build this model. And when I say, "Hey, Claude, build a DCF," it will know to build it in the DCF tab. It won't have to create a sheet.

It won't have to format everything from scratch.

So even though I took some time myself to build this template, which wasn't much time because the template is the template that we use in general within our company.

I just adapted it briefly. Right? It took me maybe 15 minutes to build it. Right? Now, that 15 minutes is going to represent huge amounts of token savings in the future. And also it's going to make my work look better, be better, layout, formatting, structure, everything looks better. Okay? So we're working off of a template, especially financial modeling. Now, just so you know, this template has three empty tabs, 3SM, that's a three-statement model, the WACC, the DCF.

Those are the three steps that we're going to model here, and it's got also an extra tab called Financials. Just for you to be aware, the Financials tab already contains the income statement and balance sheet of Nike historically. Okay? So you can see here the last three years of the income statement here, down here, the balance sheet. This is a really, really easy export from the filings of Nike. Right? It also has some management forecasts, okay, for only revenue, and that's it. Right? And then I'm going to let Claude sort of help me with the rest here. Okay? So Claude will know to take the data from here when it's building the 3SM model, okay? Great. I'm ready to now jump into doing the work.

So I'm going to close this down, right, because you know what Claude is going to be working with, right? You know what the file looks like now.

Let me close this down.

Let me show you the folder one more time. Why? Because notice how I call this Version 00.

Right? Version 00. So when I work with Claude, any time Claude will do something to this file, it will never overwrite it.

It will always create a new version.

It doesn't matter how small the change is. That's what I want.

I don't care if I have 100 versions. Okay? I don't want to overwrite in case I need to go back and check something. All right? Great. So that's the folder.

Let me lower this, and that's my project. Now, first step.

It's already done. Created the project, linked the folder.

We know what's in the folder. Step two is extremely important. It is to include or add project instructions.

Now, let me just say this thing. Whenever you run a prompt within this project, the project instructions will be added to that prompt on the back end. That's what's called the context window.

The context window that is processed every time you put a prompt will contain the full project instructions.

So project instructions are persistent instructions that impact every single chat within a project. Okay? Just keep that in mind, right? So if you want instructions that are going to be consistent across all of your chats, put them into the project instructions.

Now, I've already created this for you.

One of the files that you should have Okay? And if you don't have it, let me know.

I can share it with you.

Is this prompt library. Okay? It's only three prompts, but they're very sophisticated prompts.

So I just built that so that we can actually go through it and copy-paste this as opposed to type things in.

We only have 41 minutes left, so let's get to it, okay? So here is the prompt library.

Yeah, we'll share that file, sure.

I thought you had access to this elsewhere, but that's not a problem, right? Okay. And I'm going to actually share a few more files since now I'm already doing this, okay? What I'm about to share with you are seven skills. Okay? Seven skills we're going to use toward the second half of this session.

All right. So you should see now, I just shared with you seven skills. They should all go through.

It looks like two of them might have not gone through.

Let me see if it did go through or not, but there are seven, basically, ZIP files. Okay, great.

Now, let me keep going. Now, go to your prompt library and you will see here your first prompt. These are big prompts, and I'm going to say this quickly, but these prompts, I don't ever type word by word myself. I always go into Claude and I say, "Hey, Claude, I need to generate project instructions for this project," and I explain things.

And then I say, "Create it for me," and then I start editing, iterating through it until I get project instructions that I'm happy with.

So it's not something that I type line by line, and that's the way you should work. Leverage AI as much as you can, even when you're creating instructions for AI itself.

So take this and copy-paste.

Okay? Just take this whole thing and copy-paste. I'm going to copy-paste it from here.

Again, you will have all the materials, so if you cannot, for some technical issues, get everything here, this is going to be recorded for you. Okay? Great.

So let me go back to Claude and let me go to Instructions, open this up, and all I'm doing now is I'm taking those project instructions I just showed, and I'm going to paste them here.

And now, I'm not going to go through every word here, but I want you to just get a taste for what these kind of instructions look like.

So the first thing is this, which is extremely important.

This is basically setting the persona and what the task at hand is.

So I am an investment banking analyst in this specific team.

I'm building a valuation model for Nike, and the valuation consists of this, this, and that. There you go.

Now the project knows every time the context, the key context.

But besides that, I have some additional rules.

Now, this, again, let me say this one more time.

These are rules that apply to every single piece of work, to every single task. It's going to read these rules and it's going to process them.

Things like formatting. I want blue for inputs, black for formulas, green for cross-sheet links.

Things like inputs. If the input is something you can observe, then you have an answer, use it.

If it's a judgment input, like what is the growth rate of the company in the long term, then I want you to propose a range, but do not just make it up. Right? And then very important, versioning. Something I talked about earlier.

Never overwrite a file. Any time I ask you to take a file and do something to it, create a new version.

So this thing will be read every time.

That's how I ensure that we never overwrite, okay? Save instructions.

We have instructions. We have the folder.

In case you want to know quickly what else is here, there is memory. Memory is project specific, and memory is built automatically by Claude. As Claude observes your work, it starts creating memories by topic. Okay? Now, and you can start seeing them. After a few chats, it will pop up. Okay? It takes a few chats. Now, you can also say, "Hey, Claude, take this piece of information and save it in memory," and it will do that for you. Okay? So it's another source of context for Claude to help you do your work.

Finally, the context window. This is where you can upload files.

Again, those files get read every time you do anything in Claude, so be mindful of that. Don't have this filled with 100 very long PDFs that have tons of pictures. Okay? Because that's going to slow you down.

That's going to make Claude work a lot, right, and potentially not get really the information you need. So manage that very well.

In my case, all of my files are going to be embedded in the folder, in the Nike folder, so I'm good. All right? Great.

Let's get started. So if you go to your prompt library, now that we have our project set up, let's look at our first task.

Our first task is build an operating model, build a three-statement model, okay? You are an analyst. You are in the consumer and retail coverage team.

Your MD says, "Hey, we need to value this company," Nike in this example.

Go ahead and build an operating model. So I'm going to use AI as a first pass.

Always verify, never trust AI to give you a final product, right? In fact, you might use it for pieces of your work, but always verify. It is your work at the end.

You're accountable, you're responsible.

AI is only helping you speed up steps that might be time-consuming. Okay? Great. So what's the source of information? It's going to be historical financials and management forecast, which I showed you on the Excel template. And the deliverable is what we're going to build a three-statement model. Okay? So go to your- ... library, and there is this one big section.

This is a huge prompt, so you're going to have to paste it in pieces.

Build a three-statement. Okay? Now, if you ask me, "Whoa, that's a big prompt.

That's two pages long." Right? Yes, because to build this prompt, I iterated through AI until I'm like, "Okay, this covers everything I need." Normally, that prompt building itself is a task. Okay? But once you have the prompt, you can build the skill from it, and now that becomes a repeatable procedure that you can use very quickly in the future. You'll see that in a few minutes. Okay? So for now, let's pick this prompt, and this prompt, what is it doing for me? It's saying, "Hey, build a three-statement integrated forecast model for Nike.

Start with the valuation template.

Build it on the 3SM tab," which we know is the right tab.

And then it's giving additional information.

It's saying, "Hey, the source data that is in the Financials tab." Okay? And then it gets more into the actual modeling. For example, revenue fade.

I want you to take the revenue of Nike, and I want to fade it toward 2.5% long-term growth rate. Okay? The layout of the tab, I want you to build the income statement, the balance sheet, the cash flow statement, and so on.

Let's copy-paste it, and then I'll continue with this because it's going to take a few minutes to run. Okay? So copy-paste this. I'm going to do it on my end.

Let's go to Claude, and let's start with the first chat.

Okay? For this task, I would use something I haven't talked about.

Typically, you want to make Sonnet, you can see here the model choice, you want to make Sonnet your sort of default model. Okay? It's powerful enough, but it's not as token heavy as Opus. You want Opus for more demanding tasks. Okay? So I'm going to use it.

By the way, this one is a demanding task. It's a big build.

So I'm going to use Opus on this one. Okay? And I'm going to paste that huge prompt.

That's the huge prompt. I just pasted the whole thing.

All right? There it is.

I'm using Opus. I'm using Cowork. The folder is linked.

The template is available. I'm going to press go.

And that's going to take-- Oh, by the way, saying, "Hey, do you want me to access the folder?" So it gives you another layer of security, right? I normally just say okay or allow. I never say don't ask again, just in case. Right? I want to be on top of what Claude is doing. Right? So, okay.

And let it work. Okay? Let it work. Now, while this is going to take about five, six minutes, so I'm going to use this time to talk about the three-statement modeling piece and answer some questions. Feel free to put questions on the screen now.

It's a perfect time.

Let's see. I have a question here that it says: Could you describe your iteration process? Did you ask it to create a sample model and looked for error or missing items, or was it more of a discussion? Okay. Great question.

A really good question. So, I'll tell you my iteration process to create this prompt. Okay? What I did is I actually built the model in steps.

Right? So, for example, I said, "Okay, build the income statement." Right? And then it would do that. Right? And then I checked the income statement carefully, and I said, "I don't like this. This needs to be moved here.

This calculation should be different." I just picked on everything.

Of course, and this is the beauty of using AI as a tool, right? I am the creator. AI is helping me with this. Right? It's not doing it for me.

So without the knowledge, then your output doesn't come out as good, your prompts don't come out as good, your skills don't come out as good.

So I went, and I picked on every single thing that it did. So once I felt fully comfortable with the income statement, then I said, "Okay, now let's go and build the balance sheet." It gave me a first pass.

I looked at it. I'm like, "Oh, fix that.

I don't like this." And I'm not talking about only mechanical errors in the model. For example, my cash should be linked to the cash flow statement forecast, right? Or my accounts receivable should be based on a percentage of sales or a days ratio.

And it's not doing that for me, right? Just going through every line, it takes time.

Then as I finish doing that, I would then, at the very end, and we're going to see an example of that in a moment. Okay? At the very end, I would say, "Okay, take all the work that I've done with you and create a super prompt from this, a prompt that sort of takes all the knowledge that I've given you in this chat," and it creates the prompt. And you can do an extra step.

You can say, "Now, take all the work we've done in this chat to arrive at this final version of the model," and do what? Create a skill.

In the future, I can just run the skill, and it's going to build me something that is going to be really close to the finished product. Okay? So good question. Do you know when I put the effort on high, even though this is a pretty complex task? I would say this.

It is a complex task.

I am very mindful of my token usage, right? And I think that Opus itself, even at medium, is extremely powerful for this kind of work. Okay? Which is it's a complex task. Right? So I think that would just do it. I don't want to increase it to high or max because it would eat up tokens, I believe, that are not going to give me that much added value. Okay? Another question here: What checks do you perform after each model Claude builds to ensure that it hasn't made any mistakes? Oh, amazing question.

So it depends on the model.

But if I'm building a three-statement model, guess what? I go straight into the balance sheet. Check.

And I'm like, "Is that thing balanced?" Now, Claude is really good at doing this.

What it might not be really good at is in actually balancing it properly.

It might hard code a number here and there.

So what I actually do is I basically go first through the key areas. So for example, I check for the balance sheet being balanced, I check for my interest line on the income statement, having a circular switch, and that there's no error coming from it.

I check that my cash is linked to the cash flow statement.

I check that all of my schedules, I mean roll-forward schedules, whether I have a debt model or it could be a PP&E roll forward or an equity roll forward, that they're actually taking the last year's value into next year's beginning value, and it's actually rolling forward the data properly.

At the end of my check, I have basically checked 95% of the model.

Now, I'm not including on this assumptions, because Claude, you will see on the output, Claude will build a model that has assumptions built, which is amazing. That it can just say, "You know what? I'm just going to assume this." But it's going to give me the entire debrief. I did this, I did this. Because I'm telling it, don't just assume.

Assumptions is not an issue for me, because they're hard-coded numbers.

I'm going to go in and I'm going to replace them with whatever I want.

Claude is just giving me a first pass.

So I'm more concerned about the mechanical build of the model, the formulas, the links, than about the assumptions.

Oh, value of tokens. So I cannot give you a definite answer on that because it depends on multiple factors. But I can tell you this.

So basically, the question is how many tokens? That's what the question is.

So what I can tell you is this, though.

If you run a heavy task in Claude, you definitely should be going to your settings.

Under settings, you should see usage, and you should start checking your usage movements. So if you're running a 20-minute task, check before, check after, what percentage is being used for any given task. Does that make sense? And if you look at the current session, that's more accurate because it's only for this session before the reset.

So I can see, honestly, this is going to update later, but it might have already updated, at least partially. So it says 20%, depending on your budget, whatever budget you have, some companies give employees specific budget, like you have this many dollars.

So you know that 100% is, let's say, $30, $40 of token usage. So then you know if you move 5%, then you can get the cost.

Okay, great.

Beautiful. All right. So Claude is still working.

While Claude is still working, I want to just do something very quickly. Feel free to throw questions in there, but I want to talk about this right here.

What Claude is doing right now is it's already probably more than halfway through. It built the income statement.

It's going to build the balance sheet and link it to the income statement, and then it's going to build the cash flow statement, and it's going to close the loop or close the entire thing. Everything is going to come from what I call the anchor prompt, which is a prompt that I've obviously developed.

And I'm going to still iterate. Why do I want to iterate? Because I want to look at the model.

I'm going to find out if there's anything weird.

Anything I don't like. It doesn't have to be a critical error.

It could be a formatting thing. Because by the end, when I'm 100% comfortable with the way the model looks, I can turn that into a skill. So next time, no prompt. Next time, I have my template, which I already have, and I say, "Hey, run this skill." So before I go back and look at the status, let me just say to save some time, what are the takeaways from this first section? Number one, if you have a good first prompt, that's going to save you a lot of time.

So what I would avoid doing is doing something like build a three-statement model for Nike.

Understand and nothing else. Now, by the way, the template is doing a lot of the heavy lifting because Claude is going to read the template, and it's going to go, oh, you want this layout? Oh, you want this font? Oh, you want this format? So the template is already doing a lot of the work, but you want to give it some of your thoughts, even in the first prompt.

Make it a bit sophisticated. And if you want to use meta prompting, that means have AI build you the prompt, perfect.

The next thing is what I just asked, checks proof the model. So make sure that the checks check out and go in there and check it yourself.

And the last thing, which we're going to see in a minute, is you can take that entire chat history and turn it into a skill.

So let me see if Claude is being nice to me. It is being nice to me now.

It finished everything.

We have only 25 minutes to go, so I won't read everything here, but really important to say, this is Claude's debrief of what it did for you. And in this debrief, it's going to give you a lot of critical information you must be aware of.

It's easier to read it here than to actually try to figure it out from the model itself. So it will tell you things that, oh, look, I've assumed this, this wasn't there. For example, minimum cash, $5 billion.

It says my judgment.

And I'm like, "Well, I don't agree with your judgment, Claude.

I'm going to change that." So I'm aware now that Claude has made some assumptions.

Great.

I'll talk about the circularity next. Okay, so you'll read this.

It will tell you about the checks and some of the assumptions it made and so on.

Let's go to the folder, and when you go to the folder, you will see now version one.

Perfect. Let me open version one. I'm just going to make, just to be conscious of time, I'm going to make one big change, create a version two, and then create the skill. Okay? Great. So a couple of things. Go to the 3SM file and you will see, look, it looks great. Okay, it's pretty close.

Now, the first thing you might wonder is, hey, look at all these errors.

What is that? Right? Well, what happens is, we have a circular switch. Right? So if you go to the info tab, you will see circular switch is one.

That means that we are allowing for the model to be circular. Right? Well, that's what's creating the error. So be very careful. Don't panic.

When you see that, notice that it's happening in these two lines, which is the interest expense and interest income.

Go to your info tab, turn this into a zero, turn it off, and then turn it back on.

And now this should be fine and working. Okay? Look, I'm very happy with this model. I've checked everything.

That's not really the truth, because in reality, you are going to want to change every single detail until you make it 100% perfect for you. Right? But let's just say I'm very happy with this.

My balance sheet is fine. Right? It's looking amazing.

Okay, no issue there.

This looks great, but there's one thing. What is that thing? Well, if I look at this, this is a bit odd.

Right? So I have three lines for interest.

I have interest income, interest expense, and then I have another line that says, "Net interest income expense as reported historically." So basically what Claude did is for the historical years, it created one line, but for the forward projected years, it created two lines, separate lines. And that's not very normal. Right? Now why did it do that? Because of my instructions.

I told Claude, make sure the interest expense and the interest income lines are separate lines.

And on the financial model, there is only one line called net interest expense.

There's no way of knowing what's income and what's expense because they are netted.

So Claude is really, really good at following instructions.

In fact, that's what it's known for. It followed instructions.

So what am I going to do? Well, my prompt is too restrictive. So I'm going to go back here.

Actually, let me go back and say, okay, what rows is that? 48, 49 and 50. So let me go back here and say, "Combine rows 48, 49 and 50 into one row that contains the net interest expense, which is calculated as interest expense minus interest income." Okay. And I'm hoping that's enough, okay? But I just want to show you how you iterate through this, okay? Just in case, I'm going to close this just in case, okay? And I'm going to just run this, and this should take a few seconds.

Okay, great.

Time flies, doesn't it? We're already 20 minutes from the end of the session.

But what's coming later is a little faster and I should be able to showcase the full agenda here.

All right, while I'm doing that, go to your PDF, go all the way down, and there will be one more prompt.

The prompt says, "Turn the chat into skill." Remember how we run the anchor prompt? We did iterations. In real life, you will do multiple iterations until you get to the final product.

Well, now you want to take that history of work and turn it into a skill. So in the future, you run the skill and it gives you a very close to final version of your model. Okay? So I have this prompt which says, "Turn everything we just did into a reusable skill, call it 3Statement Build. Generalize it so it doesn't only work for Nike, but it works for Under Armour as well.

And do not have hard-coded file names, tickers, anything.

Make it general so it applies to any company I want to do this for." Okay? So I'm going to go ahead and copy that. Just have it ready.

Okay, copy the whole thing. Go to Claude.

Claude should be finishing in any minute now. There you go.

So Claude said, I did this, I did that. If I go back to my file, there is version two.

Open it up.

And I'm going to quickly show you this.

3SM.

Go down to the income statement and there you...

Oh, I have to fix my switch every time.

And there you go. Look at this line 48. There you go.

Historicals and projected, only one line. I'm happy with this.

Let's close this.

Let's now go back here and I'm going to turn this into a skill.

Okay, so this is the prompt that says, hey, turn everything I just did into a skill. All right. I'm going to run this.

For this one, we don't need Opus. It's not as complex.

It's just a skill-making task. So I'm going to change this to Sonnet When you switch models in the same chat, it's going to ask you, are you sure? Okay.

It'll be fine. It will not lose any context.

So switch and then run. Okay, I'm going to let this run in the background.

While I do that, I'm going to start a new chat.

So this one is what I'm going to call 3SM. That's done, right? I'm going to start a new chat, but this time we're going to build the WACC, okay? The cost of capital.

Now, I'm going to do things differently now. I'm way faster. Okay? We're going to use a skill.

So imagine you've already built the skill.

Right? I send you the skills. So what I'm going to do here, and I'm going to show you quickly how to install skills.

Go to customize, go to skills, and then go to add, and then say upload skills.

Okay? So customize skills, add, upload skills.

Click on this.

Go find your skills.

Take all of the skills, add them, upload, and there you... Okay.

Oh.

Yes, I had already done that. Okay. Yes. Okay.

Anyway, this is the process, right? You would do it on your own.

It would say, okay, add it.

You will see a green check. In my case, I'm not going to do any of this, okay? So I'm just going to go back. They're already here.

So how do I know? Go to skills, go to yours, and these are the skills already in there. Okay? So once the skills are in there...

By the way, notice that Claude already built the skill for the 3SM.

The great thing about Claude now is once it builds a skill, it gives you a preview of the skill. So you can open up the content, and you can see what's in the skill. And then if you save, it will automatically install the skill. Okay, I'll do that right now. Save.

I can go to customize, skills, yours, and you will see somewhere here.

It hasn't refreshed yet. It will refresh.

It should show here the new skill. Okay? It takes a second to refresh.

All right? Now, let's go back here and let's run our cost of capital. So how do you run a skill? You say forward slash, the name of the skill, WACC, click on the skill, and that should be enough. But I'm going to be more careful and I'm going to say, workbook.

Latest version of the Nike valuation model, and tab WACC. Just to be careful, make sure it takes that workbook, the latest version, version two in this case, and make sure it goes to the WACC tab and runs this skill. So this skill is something we've already built.

Okay? For this one, I honestly could use Sonnet, because it's not as complex as to build a three-statement model, that everything needs to tie out. Here, it's just a bunch of calculations.

So let me run Sonnet. It's a little faster.

Run, and there you go. Okay? It's running.

You can see here I have a new chat, rename, call it WACC, save. And while this is happening, because we only have 13 minutes...

By the way, this is going to create a new version, and it's going to give me the estimate of WACC. Okay? While this is happening, I want to open up, the template really quick.

All right? And the reason I'm opening up a template is because for the last 10 minutes, we're going to be working on the DCF model. Okay? Now, the way we're going to do it is slightly different.

We're not going to use a big prompt.

We are going to use skills again, but we're going to show how to use chain linking, or skill chaining, whichever app you want to use. It's basically using, a set of skills to build an entire model in pieces. Okay? That's one thing. The next thing is we're going to do it directly in Claude in Excel, right? As opposed to in the app.

So many people like this type of workflow.

When you actually see things, you can work with skills, you can work with prompts, but you are building the model directly in Excel, and you are seeing what's happening.

It allows you to edit things as opposed to wait for the file and then open the file, then close the file, right? So it's a different workflow. Okay? So this is the empty template, right? So I'm not going to work on this one just yet because I need to finish for the WACC calculation to end. But I just want to show you Claude in Excel. Okay? So if you click on home, and you click on Claude here.

Now, this is an add-in. Those of you who don't have it, if you don't have it here, you want to click on add-ins here and then search.

It's loading.

A bit slow.

Maybe I'll just tell you. Oh, there you go.

You will just search Claude, and you will click on it, and it will just install really quickly.

It will take seconds. Then you need to log in, right? Of course.

Okay, in my case, I'm going to click on Claude.

It's already pre-installed, and you can see I'm already logged in because it took me straight into the Claude interface, which you can see here I have my prompt field. Okay? Just a couple of things.

You can change the model here, just like we did before. Okay? And one very important thing is you can actually change the mode of how Claude approves Any edits. So here you have ask before edits.

In this case, for any edits it makes to the Excel file, it would ask you, do you approve it or not? I think that's too much.

If you accept all edits, Claude will just do it, will just write stuff on Excel.

But it will still ask you for approval if it ever needs to delete data, which is great.

So it won't just delete randomly stuff.

It will say, "Hey, I need to delete these 10 rows.

Do you approve?" Okay? I saw a message on the skills zip.

Are you saying that the skills didn't get shared with you? Let me do one skill at a time.

So I just pushed the WACC build skill. Okay? It's just a zip file. You should be able to download it.

Okay, great. So I'm going to go with accept all edits, and that is Claude.

So we work on the side. Okay? I can also share the template.

That's the template now. Okay, so you're getting slowly everything, right? We're only 10 minutes away from the end, so I'll do as much as I can.

But if I go to Claude, you can see that Claude has said, "Hey, I finished the WACC calculation." Yep, and it tells me here what it is.

Look, WACC comes out to 10.6%. Perfect.

When I open the new version, now we're down to version three.

Now we're down to version three.

I'm sorry, I'm just sending the skills.

Maybe I'll do this one by one.

Not much time left, so it's okay if we might not finish everything.

But I'd rather you be able to watch the recording and actually do the work yourselves. Okay? I shared six skills with you. Okay, great.

So watch. My WACC is ready.

My WACC is ready. Let me change the info, the switch to one.

My cost of capital is ready.

It looks fine to me.

Normally, I would check it.

This is the WACC that I need to use, 10.6.

So when I go to my DCF tab, there is a field there that I need to link to my WACC. So I'm going to link that to the WACC here.

Now I have all of the inputs I need for my DCF.

So in the last five to 10 minutes, all I can do is really show you how this process would work. But I do want to say a few things.

Claude in Excel, extremely powerful.

In my opinion, the workflow is a little faster for two reasons.

You get to run smaller skills, do smaller steps, watch the output, and fix the output in the same flow. Right? It's really, okay, that came out great, let's go next, or that needs to be fixed, fix it. Now let's move on. Right? So it's actually a closer experience to building the model yourselves. Okay? So what I'm going to do is, these are some key inputs on the DCF valuation. Yep. You can see the valuation date, the WACC that we just computed, the long-term growth rate of 2.5, and an assumption for the exit EBITDA multiple. That's for the terminal value.

Okay? So I'm going to open up Claude now that everything is working.

I am going to come here and I'm going to do this.

I'm going to say /dcf-fcf.

Now, that skill is going to do only the free cash flow build, not the full DCF, just the free cash flow build. Okay? So now, in this case, because I'm already in the DCF tab, I don't really need to say much. So what's great about Claude in Excel is that the workbook, it's the context. So when you're doing financial models in Excel, you really want Claude to be aware and work closely with every single cell in the workbook. So every cell in this workbook is the context.

Okay? So it knows the layout very well. It's going to do a great job. Right? Now, just in case, I'm going to run this skill here.

You can see on the bottom, that's the free cash flow, FCF, in the DCF tab. This is just my way of making sure it doesn't go to another tab, but it's highly unlikely it would go to the WACC tab and do it. Okay? So let me run this, and I'm going to use Sonnet. That's perfectly fine.

And what I want you to see is the experience here, right? It's more like modeling.

Okay? While this is running, can you explain again the purpose of what you did for the circular switch? Yeah, look, it's not something I have to do, right? Basically, every time you open the workbook, the interest expense and interest income lines are showing errors. Okay? And so what I need to do is I need to reset the circular switch. Okay? Remember, three-statement models are circular.

Okay? The income statement flows into the cash flow statement.

The cash flow statement determines the cash, which in turn determines the interest income on that cash.

So that's a loop, right? That circular behavior requires that we implement a circular switch.

But that switch can create errors, right? So all you got to do is reset the switch.

That's what I'm doing.

It's one, I make it zero, then back to one, everything goes back to perfect. Okay? All right. I don't know if you saw what happened, but we have our free cash flow. Now, if you see the layout, if you see the notes down here, if you see the notes to the right, why? Is Claude doing this natively? No.

Claude is following instructions.

Those instructions come from my skills, the ones I built.

I told it to have notes on everything it's doing for me to read, for me to audit what it's doing. It's part of my goal to be able to have everything. I don't want just a bunch of numbers.

I want you to tell me what you did. Okay? So that's the free cash flow. That's our unlevered free cash flow.

But as you know, when you look at a DCF, let me show you this quickly.

The DCF has multiple steps, right? What are they? Project free cash flow, compute the terminal value, discount the values, bridge to equity value, and then always do a sensitivity analysis.

Now, all we've done is the first step with one skill that I've built before this session. But we need four more skills, which you already have.

So what I can do to run the rest of the skills is go back and do this. I can say, run, and then the skill, which in this case, next will be the terminal value, so DCF or TV.

Okay. Then run the skill, which will be the DCF discounting piece.

Then run the bridge skill, the DCF bridge, and then run the sensitivity skill.

There you go. So this is what's called chaining the skills.

Okay. In a way, this is like prompt chaining, but instead of them being prompt, they're more sophisticated tools, which are skills.

Now, what this is going to do, it's going to run every skill in one go, and it's going to give me my finished product. Okay? What's great is that each of these skills, and this is a key point, right? Each of these skills will do a very, very specialized task. The more specialized a skill is, the better it will be at giving you consistent, structured, reliable output. So instead of having this huge skill that has the entire all of the steps in one place, there is more room for variation in your output every time you run the skill.

It's much better to have it in micro skills or specialized skills and then chain them into one prompt, which is really, really simple when you look at it, right? Now, for this one, just for the sake of, because the discounting piece is a bit more complex, so I'm going to use Opus, and I'm going to run it.

And that's going to take us to the end.

It won't take a minute, okay? It will take a few minutes, but at least you're going to see this thing work live.

By the end of the few minutes, we are going to have a full-blown DCF built on Excel that we can then work from, validate, check, change assumptions, make it ours, right? So I like to say this all the time.

AI is not meant to do your work for you. It's meant to help you be more productive in terms of speed and in terms of quality. But you need to own this work, make sure that you are 100% comfortable with it. Okay? We have just about a minute or so left.

I'm happy to answer a couple of questions in this minute or so if you have any.

I'll be quiet for a moment as Claude is doing amazing work.

It takes Claude some time to get started, right? You can see how it's really pushing.

Now, by the way, I'm using Opus now, which means that it's a little slower because it's doing more reasoning, deeper reasoning. Okay? And that's why it's taking a little longer, but at some point, it's going to start taking action, and you can see it's already doing that.

It's already creating space for the next block, which should be the terminal value calculation, and it's going to start writing.

Great. And the last thing I'm going to say, because we're really at the time now, is if you go up here, you see the terminal value computed.

So it finished that skill, and it's probably moving on to the next skill.

You can click here. You can see the history of your chats in the past.

It will save them here, okay? You can also click here and go to settings, and in settings, you have a few options that are really worth looking at.

We don't have time for it here, but really look at this. Do you want session login? You can actually put instructions here that will impact your Excel Claude only.

You can do other things like let Claude work across different Microsoft apps like Word and PowerPoint, sort of communicate. Okay? That is it. Claude is going to continue working past the 10:00 AM, in my case, time. It's actually almost finished. Great. We're all done here.

Thank you so much for your time. I hope this was useful.

It was a quick session, so not a lot of time for debriefing, but the idea is to show you some best practices on financial modeling with Claude.

Hope you enjoyed it.

I will be seeing you around. Okay. Take care.