Analyze an annual report with generative AI: Financial statements

- [Instructor] In this interactive case study, we're diving deep into Apple's financial statements. We'll be focusing on three main financial statements: the income statement; balance sheet; and statement of cash flows. These documents provide a comprehensive view of Apple's financial health. We can certainly get the financial statements from the 10K report, but we get much better results by uploading CSV versions of the financials. With all the formatting stripped out, these files are easiest to use for the models and let us analyze data much more efficiently. Here, we see the cleaned financial statements from Apple. The first row shows which financial statement we're reporting from, income statement, balance sheet, and cashflow statement. The second column shows the account revenue, revenue growth, et cetera, and the subsequent columns show quarterly data, going from 2020 through 2024. When I upload data for analysis, I like to tell ChatGPT what the data is and how we're going to use it. So we're uploading our CSV, and here's a sample prompt I normally use. "Please review these financial statements from Apple, and ensure you can read them and the data is ready for processing." Then I identify what the data is so that ChatGPT doesn't have to figure that out. Okay, so the data has been uploaded and ChatGPT has identified it. It's ready for time series analysis and visualization. Let's start with a broad question. I'm going to ask ChatGPT to review the income statement and produce key charts for use in a board report. We can obviously use much more specific prompts to get more specific output, but I like to start broad and see what the model picks up on before I send it in any particular direction. Notice that in just a few moments, ChatGPT created a chart that shows revenue, gross profit, operating income, and net income. In subsequent charts, it shows quarterly revenue growth, and now we have their analysis. So that's a good start. Let's move on to the balance sheet. I want to point out again that instead of giving very specific prompts or trying to be a prompt engineer, I just gave it a very basic, "Now let's look at the balance sheet." Contextually, ChatGPT understands what I mean and continues the conversation. Okay, let's check out our results. Again, we have a single chart that shows total assets, liabilities, and shareholders' equity. Let's do one more query and look at the statement of cash flows. And we see here, one chart that shows our operating, investing, and financing cash flow, as well as our free cash flow and capital expenditures, and we see the insights it added. Now, let's try a specific prompt. Instead of just asking for broad overview, let's see if ChatGPT will create a DuPont analysis. So first, we're going to make sure that ChatGPT can find all the information it needs for a DuPont analysis. It's identified net profit margin, asset turnover, making sure it has everything it needs there. So let's go ahead and ask it to perform. Perform the analysis and provide visualizations to show trend over time. And there you have it, in a matter of moments, ChatGPT has completed our DuPont analysis. We see the ROE over time and all the DuPont components over time, and here are the insights that ChatGPT provides on each. What I hope you notice there and what really sets generative AI apart from traditional software is that you can speak to it as you would an intern or a coworker, maybe a junior financial analyst, and that it understands contextually what you're looking for. In a matter of minutes, we just went through and analyzed the three financial statements. Imagine how much more productive this would make your FP&A on a daily basis. To really see the power of this tool, though, I encourage you to try it for yourself.