In this episode of FinPod, Asim Khan & Glenn Hopper talk about AI-enhanced financial analysis. They discuss the power of ChatGPT in democratizing data and how this technology can help improve workflows and analysis.

Learn more about AI’s impact on roles in investment banking and accounting and how to use generative AI to work more effectively.

Transcript

Asim (00:12)
Welcome to the What’s New at CFI podcast. I’m Asim Khan with CFI and I’m joined today by Glenn Hopper.

You’re here to talk to us about your new course in AI Enhanced Financial Analysis. It would be lovely if you could take a minute and tell us about yourself.

Glenn Hopper (00:27)
Sure. I guess I got pulled into the AI game because I wrote a book a couple of years ago on leveraging AI and finance. Of course, back then, it was a very niche book that was just really targeted at finance leaders and technologists who were using Python and SQL and everything it took back then. Since the onset of generative AI, suddenly…

this thing that just as nerds were talking about before, now everybody’s talking about it. So just because I had a little bit of a background in it before, I’ve really been focused, I’d say in the last year and a half, primarily on working with companies, kind of moving away from my traditional role as a CFO and consultant to now really helping companies implement

this new technology in their companies. And also, I’ve been working a lot on training individual contributors and finance leaders how to use AI in their regular workflows.

Asim (01:24)
Excellent. Now, we don’t mind plugging other people’s work. The title of your book, once again.

Glenn Hopper (01:29)
It’s deep finance, corporate finance and the information age.

Asim (01:32)
Excellent. Available at fine bookstores everywhere. Okay. Okay, good. Good. And then you’ve been working with CFI on developing AI related courses. I recently took your course on AI related financial analysis. And I thought it was really great. The kind of, well here’s what impressed me most. And I had a very limited background in this. I’ve played around some with ChatGPT, the free version.

Glenn Hopper (01:35)
That’s right.

Asim (01:57)
It’s that you can prompt chat GPT using everyday colloquial English. You don’t need to know jargon. Is that fair or?

Glenn Hopper (02:09)
Absolutely. And that’s really, I mean, that’s the thing to me, even we’re still in the nascent phase of this technology, but what you can already see as the technology gets better, we’re going to be able to trust it more and do more things with it. But what we’re already seeing, and one of the greatest strengths that this will give us is the ability, it sort of democratizes not just

data, but data science and the deeper analysis that we can do. Because it used to be, if you wanted to do anything beyond Excel, you had to enter the world of SQL and Python or R in programming languages. So, you know, the, the barrier to entry was very high, but with these new chatbots, instead of writing code, you can actually interact with them in human language, and that’s going to open up. Capabilities that, you know, it’s going to superpower, uh, employees and leaders who are using it.

Asim (03:01)
There’s a lot to talk about there unpack as they say. So, hallucination is a problem in these early days of AI. You ask it a question and it gives you some answer that has no bearing on it. Or maybe it’s a bit of an unintentional misdirection. But I noticed in your course you submitted some Excel files, recast so a machine could read them, and then asked

chat GPT to perform some analysis, but you could see in the background that it was writing Python code, right? So its work is verifiable in that sense.

Glenn Hopper (03:34)
Yeah, and that’s really, I mean, you know, so large language models are just, you know, they’re like the auto predict on your email or your text. It’s just natural language processing, trying to give within its context window, what it thinks is the statistically most likely response. So they were designed to do one very specific thing. Now there have been emergent capabilities that have come out of.

Of these large language models where they can do some math because they’ve been exposed to it, but they’re not doing math the way humans do math. So if you’re going to be doing financial statements, I mean, I think especially about a public company where you, you know, CFO, you’re signing off on your financial statements. Do you want to trust a black box that isn’t skilled at math and doing those? And the answer on that is no. So the

the way around that is to build in other capabilities. So instead of the large language model trying to do math just within its normal context, it actually under the hood will write a Python script that will do the math for it. So, you know, it’s like asking a cobbler to bake you a pie, you know, the LLM to do math, but then…

If the cobbler lives next to a baker and can use the services of the baker and integrate them together, that’s what we get out of the Python. So, you know, we’re still in trust, but verify with the responses that we get from LLMs, but this is a huge hurdle overcome in being able to do kind of complex data analysis.

Asim (05:12)
after I went through your course, I had a look at some of the online job boards and they’re replete with bids for Python programmers and SQL people. Is that gonna go away at some point?

Glenn Hopper (05:24)
That’s that’s one of the hundreds of million dollar questions that I’m hearing around AI right now. You know, people are trying to figure out should I still learn to code if AI is going to be able to write code for me? And you know, I would say, yes, it’s good to still learn to code. And the reason for that is, even if

AI is able to generate code for you, I think by understanding what’s going on under the hood and sort of understanding the nature of the program that you become better at identifying, you know, at engineering these software applications. And again, trust but verify it. You may ask the large language model to write a code for you. It may have misunderstood something because you didn’t phrase it correctly. So it goes down the wrong road. So if you don’t know,

the basics of programming, you wouldn’t know how to begin checking the code that the large language model wrote. So I think it’s still valuable, but the great thing is in reality, how many FP&A people are going to take the time and energy and effort when they are already domain experts in finance or accounting, how many are going to actually take the time to then get a whole new skill set where they’re programming and writing SQL queries and all that. So I think

AI has the ability to bridge that gap and open up some functionality, but to be really a superstar in it, I think having both skill sets to me, that would make you a rock star in the future, you know?

Asim (06:54)
Excellent. And so I guess those computer science departments at universities aren’t going away anytime soon. And

Glenn Hopper (06:59)
Yeah, I, you know, one that I follow, because it’s, it was one of the most popular early online courses is CS50 from Harvard and David Mallon is the professor in that and he’s, I don’t know how many hundreds of thousands, millions of people have whatever have taken this course and he’s a pretty forward looking guy and he early on adopted in so he’s teaching

and they built their own chat bot that could debug code and answer student questions and everything. So instead of running from it, they’re embracing it and incorporating it into the classroom. And I think that’s what you’re gonna see in CS50 where you have these co-pilots who can QC your code and can help write code and make suggestions and help you bug fix and stuff like that. So I think there’s, never say never, but I…

I do believe that the need for programmers is going to be around, but they’re going to be much more efficient and quicker now with tools like this.

Asim (07:55)
That’s great. And because we’re the corporate finance institute, let me get your comments on a recent New York Times article. You may have seen it, but the gist of it was that AI would eliminate part of the investment banking track the first two years, where the worst part of it, I mean, I think things have softened up since my day, but you’re really just kind of like living in the office and building models,

doing presentations and things like that around the clock. What’s your view on that?

Glenn Hopper (08:23)
I mean, I think it will, and it’s not just investment banking. Think about accountants and think about those entry roles where you’re fresh out of school, you’ve got all this new education, but no practical application. So our way of entering the workforce and moving up and understanding the business better was well, investment banking is certainly more than anyone. They work you to death, but you’re…

the amount of exposure you’re just drinking from the firehose in those first couple of years. But the nature of the work that was being done there is going to be some of the first that is really successfully automated. Anything that is just taking data, turning it into information that you can work with. So kind of the lowest level of that. And then I think as at the bottom level, is stuff is automated there, it’s going to push, it’s going to ask us on the

human level to provide a greater value, to get out of this in the foxhole tactical work and be expected to sooner than later be able to provide that strategic value. So that’s gonna present an interesting challenge to people. How can you strategically think if you bypass that entry level? So there’s an educational component too of how are you gonna educate people to be able to provide that strategy without having the practical experience of

doing what people do in their early stage of their career.

Asim (09:40)
Exactly right, and that was my critique of the article, and I think CFI generally would agree that to be a really good senior banker, you’re a really good senior banker because you spent a lot of time when he has a junior banker in the weeds analyzing companies, modeling their financials, and the slightest delta in the numbers gives you a lot of information. You know what your…

clients up to and what they could use in terms of products and services from your firm. So if you cut out that first two, three years of the development process, you’re not going to get a fully formed senior banker.

Glenn Hopper (10:18)
Yeah, absolutely. So I mean, there are so many paradigms that are gonna have to shift with this technology because it’s, you know, some jobs will just go away, some will be changed. People who’ve been, you know, become deep level domain experts are gonna have to think, okay, well, I did all that, now I have to figure out how I’m also gonna be able to use this new technology and the

trying to stay relevant and trying to stay ahead of the technology curve. It’s, you know, it’s always the case. Technology moves fast, but I’m an old guy now and in my lifetime, I’ve never seen anything move as quickly as sort of the AI technology wave. So there’s a lot changing. And I think a lot of people are kind of standing around now reading news articles, wondering what they need to do, but it’s kind of past time for people to be planning because those…

early adopters and the people out in front of the curve are about to get a massive advantage for those who are left behind. So there’s a bunch of questions to be asked, but we’re a, we need to adjust on the fly right in this wave, you know?

Asim (11:21)
I think the course that you develop is an excellent start for people in the financial field. And I encourage everyone to have a look at it. And Glenn Hopper, thank you so much for your time today. It’s a pleasure chatting with you and getting your insights. We could have gone on for like an hour or two. Maybe for another time.

Glenn Hopper (11:34)
Yeah.

Yeah, well this was great. Always loved talking to you and happy to join you today.

Asim (11:41)
Thank you, glad to have you. Take care, Glenn.