The Question Behind One Wrong Cell by Jake

The Question Behind One Wrong Cell by Jake - October 2026 Scholarship Essay

The project that taught me the most didn't come from a class. It came from a question I couldn't stop asking: how much of financial analysis is actually analysis, and how much is just moving numbers from one place to another?

The question started with a memo. In the summer of 2025, as a Financial Analyst Intern at AlwaysOn Digital, I wrote a six-page memo on The Trade Desk built on a five-year DCF, and I ended with a Hold rating. I was proud of it, but I noticed something while I was building it. Most of my hours didn't go into thinking about the business. They went into pulling revenue, costs, and cash flow out of SEC filings and typing them into a spreadsheet. And I already knew from experience what one bad input could do. In a stock pitch competition, a single formula that pulled from the wrong cell had gotten me disqualified, even though the stock later hit my price target. So I started wondering if the slow, error-prone part of the work could be separated from the part that actually needed judgment.

In July 2026 I decided to find out. I started building an AI research agent in Python on my own time. The first piece was a retrieval pipeline that indexes 10-K and 10-Q filings, so the agent answers questions from the filings themselves instead of guessing. Then I used it to automate full DCF builds, covering revenue, free cash flow, and terminal value. The part I cared about most was the last step. I added checks that flag inconsistent data before it ever reaches the valuation, because that's exactly where my stock pitch fell apart.

The project changed how I understand finance. I used to think a good analyst was someone who could build a model fast. Now I think a good analyst is someone who knows which parts of the model can be trusted and which parts need a second look. Building the checks forced me to write down rules I had only ever followed by feel, like making sure every input traces back to a source and comparing the output to a rough estimate done by hand. Once those rules were written down, I could see where my own process had gaps.

It also made me more curious, not less. Every answer created a new question. If the agent can pull the numbers, what should I spend the saved time on? Which assumptions in a DCF move the price target the most? How do you know when a company's story and its numbers have stopped matching?

I'm a finance major and an equity analyst in Quinnipiac's Student Managed Fund, and I bring those questions to every model I build there. The project didn't replace the thinking. It gave me more room for it, and it showed me that the most useful thing I can build is a process I can trust.