What AI won't find

A case study in tax analysis.

The deal died on taxes. A phone call saved it.

The short version: a luxury multifamily recap penciled on every line except taxes. My tax comp was a near-identical property in the same submarket and the analysis held up to scrutiny, but the deal still did not work. What I was missing was a tax reduction the sponsor had gone and asked the municipality for, tied to a revenue adjustment. It appeared in no public record and belonged to no formal program. It surfaced in a phone call. No amount of research, and no AI tool, would have found it, because it had never been written down anywhere.

I was on the multifamily acquisitions team at a public REIT, underwriting a luxury apartment. It was a recap, the sponsor was experienced, and most of the expense underwriting was straightforward. Then came the tax analysis.

A comp you could not argue with

Tax analysis is where multifamily underwriting gets tricky. Every county runs assessment differently, and a clean comp is hard to find. I got lucky. There was a similar luxury property a few blocks down the street: same vintage, same quality, same submarket. Hard to find a better basis.

I ran the analysis. I finished the underwriting. The deal did not work.

Deals are rarely that simple, and this one was no exception. I got pressed on the underwriting, and everything came down to taxes. I walked through the analysis line by line. It held. One of the directors on my team could not find a hole in it. He finished with "You're missing something."

Finding it was my job.

Then came the call with the sponsor. We talked through the deal, and eventually we got to taxes. He mentioned how he had secured a tax reduction tied to a specific revenue adjustment.

That was the missing piece.

I did not find a single public record of it. But this was a seasoned operator who had done a lot to revitalize that downtown. He knew how the municipality actually handled assessments, outside of any formal program I could look up, and he knew how to get a reduction in taxes.

With that piece in place, the deal worked. We were ready to negotiate terms.

What the file could not tell me

Here is what stays with me. The answer was not in the county records and not in the comps. It existed in one person's head, and it only came out in conversation.

AI would not have found it. AI can do everything I did before that call: pull the comps, run the analysis, check the math. It would have reached the same wrong answer faster. What it could not do is know a wrong answer was there, or pick up the phone.

That is not a temporary limitation of the tools. Some information has never been written down. It lives with the operator who negotiated it, and it comes out when someone who knows the deal asks the right question.

So use AI for the analysis. It is faster than you at the mechanical work, and that gap will keep widening. But the mechanical work is not where this deal was won. Two things won it: a director who kept on pushing because he knew a deal was there to be had, and a sponsor willing to tell me why. Neither of those is getting automated. As everything around them gets cheaper, they get more valuable.