In this episode of What’s New at CFI, we are talking about a course that is going to help make everyone’s job a little bit easier. The course itself may be in the context of financial statement analysis, but in today’s world, everything is data-driven, and in turn, so is almost every job. Spend some time with Glenn as he talks about what AI can do, what it is best at, and how we can interact with it now that generative AI gives the layperson a friendly user interface.
If you want to learn and practice how to make your working day a little easier, this is a course you don’t want to miss.
Transcript
Meeyeon (00:13)
Hi everyone and welcome to another episode of What’s New at CFI. Today I’m joined by none other than Glenn Hopper and we are going to be talking all about advanced prompting for financial analysis. And now Glenn’s been in the AI space long before, certainly before I…
and I feel like kind of before anyone has and I just want to talk about what this course, like why did you make it, who’s it for and what can people expect to get out of it. I know you’re really passionate about like the AI space quite generally. So yeah, this is extremely topical for everyone at CFI. Financial analysis is such a broad topic, but yeah, building AI into it, what could be more ideal?
Glenn Hopper (00:59)
Well, thanks so much for having me on.
So it’s weird AI right now has sort of two components to it. There’s one fear of it or not quite understanding it. And then on the other side, there’s this fear of missing out. And I think for some more straightforward areas, like maybe writing marketing copy or summarizing meetings, things like that.
It made a lot of sense of, okay, this is how you can use an LLM powered chat bot, but for financials and math and numbers, it’s not as intuitive. And a lot of people didn’t know how to do this or don’t know when you can use an LLM to do math and not. So the whole idea between this course and the other courses I’ve done is understand the sort of the strengths and the weaknesses of large language models and see that this is where
the future is going. And it’s even right now, if it’s, you know, it’s a little bit clunky because it’s not built into your normal workflows. It’s not built right into Excel, which is where we mostly live. And it’s not built into our ERPs yet. All this is coming. And I think right now though, being able to experiment with it and use it and see the functionality of it is going to put you ahead when kind of when the other, the software companies start catching up. So for this course,
I think there was, there’s a gap between sort of the knowledge base and the finance industry right now and the capabilities are out there. So the thinking of this is let’s not talk about using generative AI and the abstract. Let’s dive in, let’s get deep and let’s do some actual financial analysis. So in this course, the idea is we’re going to take actual financial statements from public companies and go through and do an analysis and show,
really show in this course how generative AI is very different than any kind of financial software we’ve used before, because it’s not just binary where you flip a switch and make these determinations of this is exactly what I want. It’s more of a collaborator and a coworker, the way that you work with this generative AI. So I’m hoping to show that in this course.
Meeyeon (03:14)
And that’s super interesting because I think so with AI, it’s always kind of been around, I guess, in a less in-your-face way in that. So Excel always had predictive tags and it would do certain things that were AI-esque, but it wasn’t spelled out and put in front of your face. And with things like ChatGPT and AI generally, like you said, it is…
people feel much, and certainly I feel, much more comfortable using it for writing copy text because it’s extremely easy to edit. You don’t really need to second guess it. It’s your verbal language, right? It’s super, super easy. But then when you want to go and use that to do numbers, there’s all of a sudden this fear. Like, I have the same thing. just think of the idea of, so back in the day when I used to work in debt capital markets and investment banking,
and I did a lot of bond calculations. Now the person in that seat today could probably use a great deal of benefit by taking your course and thinking about, right now the AI that we use, as you said, is a little bit clunky. You have to go to a different like webpage, you have to download something, you have to put something in, you have to edit it there, you have to transfer it over. But that’s all going to be integrated soon. And to be able to leverage large LLMs to try and do
your work a whole lot faster, especially when it’s at the more, the less intensive part, like data collection, data, not necessarily collection and gathering, but making sure it’s, you know, like blessed data, like it’s good to work with. AI, think is excellent for that. And so are the learners going to get a bit of practice kind of doing that type of work in this course as well?
Glenn Hopper (04:57)
Yeah.
Absolutely. And you know what you just mentioned there. the, so the first off the trade off of it being a little bit clunky because you’re coming out of Excel or coming out of your ERP or whichever system you were in and in moving it to this web interface. The trade off there is the outputs, if you do it right, are so incredible. And you’ve, you really, it’s, always describe working with an LLM as working with a very, very smart, but very green intern.
So the idea is, you you, you prompt it and that’s what we talk about, how to interact with this and how to, how to prompt, the LLMs to get the best results, but you prompt it and you work with it and you get increased value out of it. So the other thing that you noted was, that the AI was happening under the hood in some areas in ways that we didn’t really think about or understand yet. Yeah.
Meeyeon (05:59)
That’s the best phrase under the hood.
Glenn Hopper (06:01)
And it’s so the thing with generative AI is going to that natural language now. So, so when it was happening under the hood, you had to be in the secret club that knew how to write Python and you were a coder or a machine learning engineer. But now since you can talk to it and interact with it in natural language, that barrier is removed. So we’ve been talking about democratization of data forever for 20 years now or whatever, but now we’re talking about democratization of data science. So this.
deep analysis that we’re doing that you used to have to be able to write Python to do. Now we can just do it in natural language. And that’s what I really like hope to show in this course.
Meeyeon (06:38)
With LLMs, are they built off like computers crawling large neural networks and continuously kind of expanding that neural network?
Glenn Hopper (06:48)
So the simplest way to describe the LLMs is that they first off the data that they’ve been trained on. It’s such a huge amount of data that the only way to describe it is just think of them. They’ve read the entire internet. That’s the which is yeah.
Meeyeon (07:04)
Yeah, the whole internet. So my husband’s a software engineer. He does firmware. So he tells me about this from time to time. But then, you know, like, I want to say like maybe 80% goes in and then like 20% goes out. So it’s like, I’m always kind of taking one step forward, two steps back.
Glenn Hopper (07:13)
Yeah.
Yeah, I mean, it’s incredible to think about, and if you say it’s read the entirety of the internet, well, that means that’s the sum of all human knowledge, which is kind of crazy to think, but as, we’ve spent the past two and a half decades or going on three decades now sort of digitizing everything. and just ignore all the copyright and concerns around that and assume these companies aren’t telling us what all they’ve trained on, but we know it’s…
Meeyeon (07:30)
Which is wild.
Glenn Hopper (07:47)
billions and billions of documents and you know, just everything so all the good and bad and and everything that’s in it so But then so there’s that training where it’s just absorbing all the internet. It’s not Here’s something, the hardest part to understand with that is LLMs are not memorizing facts. So if I tell an LLM who I am, it may you know, it’ll retain that in memory
for my instance, when I interact with it, just like Google knows what I search for and the algorithms and everything. So that’s all there. But the LLM isn’t going to learn from me uploading one piece of data about myself, because it’s not learning facts, it’s learning probabilities. So an example I always use is take someone like a famous person, Michael Jordan. If I type, what is Michael Jordan known for? Well, it’s read, I don’t know how many tens or hundreds of thousands of articles about
Michael Jordan. So it knows, you know, probability-wise that if I’m talking about Michael Jordan, that I’m talking about the basketball player, or it may say, Hey, there’s this other guy named Michael B. Jordan, who’s an actor, that who you probably have not as many entries. Yeah. But it’s not actually memorizing the fact. That’s, you know, it’s, I always tell people that because
Meeyeon (08:53)
Yeah!
I’m sure he’s sick and tired of that.
Glenn Hopper (09:07)
Again, I’m not saying go upload all your proprietary information into these chatbots, but that’s not really how works. It’s not going to just take, if you put your social security number in, it’s like, yeah.
Meeyeon (09:16)
It’s not taking facts, it’s making inferences based on like so much data that’s collected, which is like the whole probability aspect of just making these inferences and saying like, this is commonly searched, this is what’s out there, this is probably what the person’s looking for, which is kind of eerie.
But then, that gets me to the question of not, so we’ve covered, we’ve talked about why you’ve made this course and what it’s all about, but who is it for? So we have, you, we are, at CFI, Corporate Finance Institute,
Glenn Hopper (09:21)
Exactly.
Meeyeon (09:46)
and the title of the course is Advanced Prompting for Financial Analysis. But this sounds like even if you’re not necessarily working in finance, you could learn it in the context of numbers, and it could be incredibly helpful.
Glenn Hopper (09:59)
Exactly. And so, know, so the course is obviously targeted for the CFI core audience. It’s FP&A folks, it’s corporate finance and accounting roles too, because there is a lot of, you know, think about reconciliation. are, you know, month in closed process things that these can all be maybe not fully automated, but we’re getting there much closer using AI to sort of assist and increase efficiency there. you know, I’d say analysts, students,
anyone really anyone interested in understanding and applying AI and in financial statement analysis, but to your point, so I wrote a book earlier this year on AI for finance professionals, but the first, you know, the first four chapters of the book are explaining how AI works. So I think it is important that everybody understands sort of the basics of this is what we mean when we say AI, this is what we mean when we say,
say machine learning, is what we mean when we say artificial neural network, deep learning, all that. But anyone who’s going to be doing something involving math with LLMs could benefit from this. So even someone in sales and marketing that’s doing customer segmentation and in clustering and maybe making sales forecasts and all. So this would apply to folks outside of finance. Anyone who’s wanting LLMs to work with,
customer data, financial statements, etc.
Meeyeon (11:27)
Yeah, and one of the things that before we recorded or before we did this podcast, a headline that made me think of like this course, would be super helpful for something like this is Macy’s. You see that headline in the Wall Street Journal? I think it’s been a few weeks now. They had a, so Macy’s had a large, large accounting error. We’re talking about like a $154 million worth of kind of like,
Glenn Hopper (11:46)
So, no. What was the Macy’s story?
Yeah.
Meeyeon (11:56)
missed P&L due to shipping expense and whatnot. Apart from the whole, I mean, what is NetMacy’s whole P&L ? What is their net profit margins? What is their net income? Like, 154, that sounds like a pretty significant line item, but it just seems like, even with the most basic kind of AI, for a company that’s
Glenn Hopper (12:16)
Yeah.
Meeyeon (12:25)
is as big as Macy’s. There’s lots of companies that are larger out there. This seems like it can be incredibly helpful for reducing even those kind of glaring errors.
Glenn Hopper (12:36)
Yeah, and that’s a great point. mean, computers are much better than humans at anomaly detection, at pattern recognition. And so if there’s, you know, if there’s historical information that is being recorded and suddenly it doesn’t show up, maybe in a massive P&L like in a company the size of Macy’s, a human may not catch that, but an algorithm going through looking for these anomalies is going to say, hey, we had this shipping
Meeyeon (12:44)
Yeah.
Glenn Hopper (13:04)
expense, gone now, You know, might want to look into that or noticing where the, you know, you’re noticing that there was a trend on your net income or on your profit margin or whatever. And that’s changed, you know, and there’s a new pattern in your profit margin. You know, that’s the kind of thing that a computer system would pick up on before people. So I think that’s the way we’re going to start seeing this. There’s going to be sort of the interaction that we do in real time with AI.
But also there’s going to be AI running in the background looking for these anomalies and pattern detection and correlations and all that.
Meeyeon (13:39)
Yeah, because as one individual person, there’s only so much time in a day and there’s only so much you can do. And typically, when you’re doing year-over-year analysis, I think, for a large public company, you’re probably going to look at the last five years. Even with FP&A, you’ll probably look forward to the next one, three, five, seven years. But if you ask a person to say, let’s
check all of our kind of data and all of our seasonality and just like really double check all our numbers by looking at the past 15 years of data. That sounds like a large task. But with generative AI and everything that we have today, it seems like, okay, it’s much more manageable now because like you said, it’s the whole internet and gives you everything that you could possibly think of. So if you feed all the information that you have into it, it’s…
Glenn Hopper (14:20)
Yeah.
Meeyeon (14:32)
it’s certainly going to be, I think, more reliable than myself at reviewing it initially.
Glenn Hopper (14:42)
And that’s so that here’s the interesting point of clarification on that is the generative AI piece will be the front end of how you interact with it. But the sort of analysis that’s happening, that’s going to be machine learning. That’s going to be what I guess we’ll now call classical AI that we’ve been using for 15 plus years that is doing this pattern recognition. And we could have been doing this all along, but the thing is now we can, like I said, we don’t have to write code to do this. We can just, you
through generative AI say, hey, we need to check this, you know, and maybe there’ll be something that’s set up in sort of a real-time monitoring, but most likely it’s going to be, here’s our GL for the month. You know, add this to add this to the information you have, do a correlation analysis, look for anomalies, variance analysis, all that. mean, so the generative AI is how we’re talking to it, but what’s happening behind the scenes is just classic
machine learning algorithms that are studying the data. But we’re going to be able to do it now, a much broader group of people than we’re able to before.
Meeyeon (15:44)
Yeah.
Because I know that I don’t like, so if I were in school, if I were in, you know, university 10 years ago and all this was happening, the language that I would choose to learn apart from Chinese is a computer programming language. I would learn C sharp, C plus, whatever it was. And so now I didn’t learn those things 10 years ago. And with the advent of generative AI, I can actually participate.
Glenn Hopper (16:18)
And I will say, so we’ve got some other courses coming that are, that talk about how to use generative AI to write code, to build your own applications. It’ll help. And I will say having not that you have to be a real coder, but at least having an understanding of sort of the basics of a language. And I keep referencing Python just because that’s the one I’m Python and R are the two I’m most familiar with, but, and how much you can do with Python. So even if you’re not the best coder, it’s kind of like,
I can write SQL queries. just takes me a longer than it would for someone who does that all day, every day. Yeah. So, but if you know what you’re, if you know the basics of coding and you know what you’re asking and you know what to look for, it may, it does make it easier to do that. So I still, I think there’s reason to, it’s not, that someone has to be, you know, you mentioned Chinese, not that you have to be deeply fluent in
Meeyeon (16:49)
My God, select star from, ugh.
Glenn Hopper (17:14)
encoding, but you could be conversational encoding to where you can sort of navigate.
Meeyeon (17:18)
Exactly. No. like, I remember, so this brings me back to my DCM days. There would be a lot of VBA, but there would also be just like general, like C language stuff in some data analysis programs that we have. And I knew enough of the language to, to walk through. Forget what the correct term is, but to basically like walk through it and then play it and then edit a couple of things there.
And then I would have to hit compile and then like errors would come up and I would panic. But I knew like just enough to be able to read it and edit it ever so slightly. But nowadays, it’s fairly easy to like go on the internet and at least like get that base level of proficiency. Whereas back then, I was actually, had a, it looked like a Bible. It was in those like thin Bible pages where I’d kind of have to go through the appendix and see what my topic was and where it is.
Glenn Hopper (17:49)
Thank you.
Meeyeon (18:12)
But it seems like what everyone’s gonna get out of this course is a way to make your daily life, I think almost everyone at this point is in an analytical role, no matter what you do. Whether you work in finance, whether you work in marketing, everything is quote-, data-driven. I think we hear that phrase so much now. Everything is data-driven. So regardless of what role you’re in, it is gonna be an analytical
role, no matter what, whether people realize it or not. think it is sales, marketing, everything is very data driven. And so this course is going to show and let people practice how they can get comfortable with using AI to make their daily tasks more, not necessarily streamlined, but I think just easier of a lift, especially with more heavy tasks that they have.
Glenn Hopper (19:07)
Yeah. And so, mean, the course really, so it goes through, it talks about sort of kind of the foundational, the foundations of this is generative AI, this is LOMs, this is how we use them in finance. And then we go through data prep, which is different, you know, what we are used to doing in Excel and creating these Excel spreadsheets that are human readable. So we do all this formatting and this make them look pretty and all that, all that formatting and everything just makes it more complex for a,
for an algorithm. we talk about stripping it down, value only CSV files. Let’s get them into a format that the computer understands because it makes it more efficient when we start loading them in. But we talk about that. And then we talk about chain of thought prompting techniques so that we can say, here’s how you can ask a broad question of the LLM. But if we’re driving for specific results, here’s the order that we need to go through and ask the questions and interact with it.
And then we go through, do some practical exercises, calculating, you know, financial ratios, doing forecasting and creating narratives around the financial statement analysis. And the creating narratives part is the most interesting to me because we’re all used to, we can, we have our models that we build in Excel and we do our analysis and all that. it’s, we’re doing the interpretation and the, you know, in providing that analysis, but with generative AI, you can get it to do the first round. You can say,
analyze these financial statements and give me an overall assessment of the health of this company or what trends do you see or what, you know, where do you see variances and asking these open-ended questions. And I think that’s the real eye opening piece to people. And it’s also the one that’s going to change the way we work so much because this is having an assistant right there with you that you’re asking these questions to. It’s, it’s seeing things and actually interacting with you in a way that maybe, you know, a lot of the stuff is the same that you would have seen, but
sometimes, it’s gonna come up and amaze you and find some correlation you didn’t even realize and find some recommendation that you didn’t realize. And I hope that that comes out of the course as well.
Meeyeon (21:11)
My gosh, it’s amazing. Sounds like it’s going to be the most, it’s going to be an analyst in your pocket, analyst in your smartphone, in your laptop. And that analyst is available 24 hours a day, seven days a week, 365 days a year.
Glenn Hopper (21:15)
Yes.
Yeah, well said. That’s exactly it. Yeah.
Meeyeon (21:25)
Amazing. So I am so excited for everyone to take this course. I myself actually have started the course. So I just, yeah, I can’t wait for everyone to take it. It’s super exciting. It’s going to change the way we work.
And that’s been a wrap to our episode of What’s New at CFI. Thanks everyone.
Glenn Hopper (21:43)
Well, thank you.
link
