How I'm Using AI to Value Commercial Property

They're coming for my job, may as well make friends now

I now use ChatGPT to (help) value investment properties, since AI is coming for my job anyway.

The results are shocking – not what AI can do now, but what’s coming.

This is the future.

Here’s my 5-step AI underwriting process on an active deal, with a screen record at the end of this piece.

But first, some notes and caveats:

1) Unfortunately, this takes work.

You can’t just upload an OM and say “how much is this property really worth” and think the system is going to give you a valuable result.

Because …

2) ChatGPT and other LLMs are tools

I use them as an analyst and brainstorming partner. They are not replacements for Excel, Argus, etc.

Not yet anyway.

3) Pick your system, they're all pretty good

I use ChatGPT because I’m comfortable with it, but hear that Claude is also quite good. So try a few and stick with the one that works best for you.

4) The more information you feed in the better.

The idea here is not to “test” the system to see how good it is without any background data. You want it to help you make better decisions, so give it the tools to do its job.

5) Chat GPT version 4o vs o1

I use a combination of Chat GPT version 4o (because it can read PDFs and other files) and version o1 (because it can reason).

In my limited experience with it, version o1 has been the unlock – big upgrade from previous versions.

It can take a little while to “think,” but the reasoning ability allows it to solve more complex problems from a basic prompt. The potential here is tremendous and we are only scratching the surface.

I haven’t explored creating my own GPT yet, but once I get my process a bit more dialed in, I look forward to giving that a whirl.

6) Still in BETA

Chat GPT version o1 is still not fully rolled out so there's a limit to how much you can use it.

I bumped up against that limit just creating this post, so not quite ready to replace Excel just yet.

7) My 5-step process for underwriting a multifamily deal is below:

I run through how I use Chat GPT on a deal we are actively underwriting.

I’m constantly refining this process, so would welcome (constructive) critiques and other ideas.

--

Step 1) Define your objectives.

Here’s how I start:

I would like this thread to be the place where you learn how to analyze real estate investments along with me. What information can I provide you to be most helpful?

The response is a pretty good template for any real estate investment strategy)

Step 2) Answer the prompts

Provide as much detail as possible. Again – you’re not testing this thing, you’re using it.

Here’s how I responded (modified slightly – can’t give out all my secrets!)

Remember most systems have not trained on 2024 data and probably have not seen all the super localized data that you have access to.

I am interested in buying multifamily properties in the San Francisco Bay Area, with a medium-term hold focused on cash-on-cash yields of 6-8% achieved within a 3-year stabilization period.

After an initial pop during the pandemic, the rental market in the Bay Area has stagnated and prices have begun to fall, creating opportunities to acquire properties at an attractive basis and going in yield.

Employment remains strong region-wide, with pockets of weakness in certain technology sectors. Return to work is picking up steam, which is making workers want to live closer to where they work.

I am looking at properties that are 10 units and up, and there is no upper bound for price or building size.

In San Francisco and Oakland, we need to consider city-specific rent control regulations which create both risks and potential to add value. California also has statewide rent control for all properties outside San Francisco or Oakland.

For each property I will deliver a set of assumptions to use in your financial modeling.

I have been making investments like these for the past 15 years and am knowledgeable about these topics, and am looking for help screening deals, identifying opportunities risks, and knowing which deals to focus my time on.

Step 3) Define your investment parameters.

For each deal, you’ll need to tell the system how you want it to run its calculations. Here’s what I feed in for every deal:

(Note: you’ll see that this does require some underwriting. In time, I envision having a custom GPT for different property types and cities with global assumptions and parameters not unlike standard pro forma.)

I am going to upload a rent roll for 123 Main Street, please use the following parameters to determine the maximum purchase price

Annual Rent growth
Annual Vacancy
Purchase debt terms
Annual property tax rate
Insurance costs
Repairs and Maintenance
Utilities costs
Property Management cost
Other operating costs
Sale Costs
Exit cap rate
Hold period
Target gross levered IRR
Rent roll: (pasted in)

=> For the rent roll, version 4o is pretty good at extracting the rent roll from an OM into a spreadsheet, but you’ll want to double check the accuracy. You can then paste the data into version o1.

Step 4) Wait while the system works and spits out the results.

If you’re used to the regular versions of these LLMs, be prepared to wait longer than expected for version o1 to “think.” It can take a minute or more.

The great part is that the system “shows its work” so you can edit check the data as it comes in. I have not found any major mistakes, the errors are usually in me providing incomplete or hard to interpret information.

In this case, Chat GPT spit out a price within $250,000 of my underwritten pro forma, which is not bad for a first pass.

I can then layer in further assumptions like capex, unit renovation costs, etc and tighten up the analysis.

Step 5) Iterate and refine, explore new scenarios, stress test

After the initial results, you may notice inconsistencies or errors, or just want the system to redo the analysis in a different way.

And here's where it gets fun.

Get the system to explore new scenarios, stress test business plans and assumptions, and optimize your business plan.

For example:

The deal I’m evaluating has a bunch of studios which I think could be converted to 1-bedrooms.

I could ask it:

What is the rent level that I’d have to achieve for converting the studios to 1-bedrooms to be accretive to my IRR, assuming I spent $20,000 extra per unit renovation?

One thing I don’t love about this particular deal is that the basis is a bit higher than I’d like, both on the way in and the way out.

So I could I also ask:

Please provide the exit basis using the following scenarios?

Scenario A: 4.5% exit cap rate
Scenario B: 5.0% exit cap rate
Scenario C: 5.5% exit cap rate

Keep all other parameters the same.

Could I do that in Excel as well? Sure, but imagine what other scenarios I can run without having to fiddle with my assumptions in a complex and error-prone spreadsheet.

For a 15-unit building changing assumptions isn’t that time consuming, but what about an 80-unit deal underwritten at the unit-level? This can be a tremendous time saver.

Keep iterating and pushing the system, you’ll be surprised.

Conclusion:

If none of this seems impressive, that’s probably because you’re sophisticated enough to do all this in a system you’re more familiar with.

But ask yourself this:

What value am I adding if all this can be done with a few simple text prompts?

And this was the point of my original post that went viral – why would someone need to hire me if they could just ask an LLM these questions in plain English?

This was the ah-hah moment that made me realize that if I don’t learn to use these systems as tools, they’ll replace me.