Year End Analysis and Optimizing My Portfolio

[Chris Brown]

Hi All,

At the end of the year, I try to look back and try to figure out what went well and poorly... any why. I'm just past a decade of real estate investing and am starting to set annual goals for myself. 

I've started a project in ChatGPT to try to assist me with my year end analysis. It's very much a work in progress, but I'd like to be able to compare year-over-year numbers moving forward to help me identify what went well and poorly. I'm a numbers guy and have a copy of Frank Gallinelli's book What Every Real Estate Investor Needs To Know About Cash Flow. I'm also a Stessa user for my accounting. I'm not a very experienced user of the generative AI models like ChatGPT, Gemini, Copilot, etc and was hoping for some recommendations, tips, tricks to improve my prompt that I'll be able to use year to year. I've included the link to the ChatGPT project with the current iteration of the prompt. I generate my income statement, net cash flow, rent roll, schedule of real estate owned, and maintenance logs from Stessa and RentRedi as inputs. Ultimately, I'm shooting for a fairly comprehensive 12 sections with various calculations from the book to benchmark my operation of my properties. I'm open to any and all suggestions. Thanks in advance.

1. Summarize the Reports

2. Property-Level Valuation Analysis

3. Portfolio-Level Metrics

4. Income Detail

5. Operating Expense Detail

6. Capital Expenses

7. Debt, Transfers, and Cash Flow

8. Equity & Investment Changes

9. Unit & Square Foot Metrics

10. Year-Over-Year Analysis

11. Market & Economic Context

12. Narrative Analysis

[Jade Miali Everett]

Hi Chris, from what I can see it looks like you're trying to build an AI CFO/Portfolio Analyst and I have successfully implemented AI myself into my goal planning/tracking and financial forecasting, so I get what you're trying to do and why. What can be overwhelming is that often AI will give you a portion of what it is that you need, but it tends to make suggestions that can take you down a rabbit hole and then next thing you know, you've wasted hours and still haven't crafted anything useful.

I am mentioning this because it sounds like you're asking for prompts but I think what you're actually trying to tackle is designing a repeatable analysis system that AI plugs into.

Instead of one giant prompt, try building these three prompts instead: a) a fixed analysis framework prompt (define how AI should think, calculate and structure outputs. This will rarely change, but you need to tell AI how to think like you by telling it what you would be looking for when reviewing your data. b) a data injection prompt (uploading your report exports from Stessa/RentRedi) c) comparison prompt (feed prior year outputs into the model for YOY analysis.

Each section that you listed should have its own prompt template so that results stay clear, errors don't contaminate the project, you can improve each module independently and rerun only the sections that you care about. Another tip that I have when using models like ChatGPT is to start new conversations once one gets long... I've noticed a difference in the speed of responses once it has to review so much data.

I can give you an example of a prompt module. Let's say you're crafting one around your income statement, and you're uploading the document as a .CSV. Your prompt could be: "Normalize income categories, flag anomalies, compute effective rent growth, vacancy drag, concessions impact and identify top 3 drivers of revenue change." When you're done crafting each section's prompt, you can create a synthesis prompt that combines all modules into narrative insights.

This approach is how I architect most dashboards or workflows for projects that I work on. It's important to remember that you don't need AI to do the math... Stessa already does this for you. You need AI to provide you with pattern detection, outlier detection, casual reasoning, scenario analysis, narrative synthesis and decision guidance.

Some things that you will need to be sure to do is making sure that your expense categories data doesn't vary from year to year and that you have everything categorized correctly.

Last, I would consider two things: adding benchmark intelligence with ranges from that Frank Gallinelli book into the prompt. Benchmarks like operating expense ratios, debt service coverage targets, cap rate expectations, cash on cash benchmarks and maintenance per sq. ft benchmarks. The second thing would be to further automate this flow by creating an automation stack. Your automation stack allows you to automate as many steps in this process as possible so that you're spending less time completing the task and in less steps.

If you'd like help or want to discuss any of this further, feel free to reach out to me. I'm incredibly passionate about AI and how it's being used as a tool to empower us to work faster, smarter and with more resources at our fingertips!