After the AI ‘Dopamine Rush,’ Are Deal Teams Any Faster?

Author:
Co-Founder and CEO
Stress, computer and business man in office with mistake for coding error, software and ai project. Programmer, worry and person on tech for website developer, glitch and programming fail at desk

Four ways to keep AI focused on busting bottlenecks.

Imagine an acquisitions director in Phoenix or Atlanta who needs a rent roll for an open-air shopping center anchored by Sprouts or Whole Foods. His analysts won’t be free for another week, leaving him with two possibilities:

  1. Go through the slog of pulling valuation and underwriting data out of leases for the likes of Sephora, PetSmart, Old Navy, Athleta and about 15 others, all by himself.
  2. Dump it into Claude, go grab an ice-cold La Croix from the breakroom, and return to his desk to see a freshly minted rent roll, courtesy of Anthropic.

Number 2 it is.

“Wow,” he says to himself, feeling a thrilling rush of empowerment. “Now I can get this done without depending on anyone else, and I can do it incredibly fast!”

Dopamine hits like this are coming fast and furious in commercial real estate as more investors and executives experience the newfound sense of autonomy created by AI. It’s the feeling of no longer needing to wait for others to handle highly specialized subprocesses related to legal review, market research, financial modeling or investor reports.

But in the case of that acquisitions director in Atlanta, two questions loom:

  • Does the underwriting team actually arrive at accurate numbers any sooner as a result of this new workflow?
  • And can a colleague reproduce the same results in the same way, even when the acquisitions director or the analysts are out of the loop?

Breaking through bottlenecks vs. AI tinkering for tiny gains

Analysts can lose an afternoon tinkering with AI for the same reason they often get lost in Excel tweaking formulas or alternate-row shading: There’s always one more thing to try. On a live deal, though, stakeholders need to zero-in on what has really caused progress to stall.

Eli Goldratt’s The Goal raises a practical question that should be front and center when the analyst finishes the abstract: Whose desk does it sit on next, and for how long?

Imagine a situation in which the draft abstracts for a shopping center are ready on Monday instead of on Wednesday. If the senior analyst can’t check the rents against the leases and amendments until Thursday, underwriting still has to wait.

If the underlying bottleneck remains unchanged, faster output simply means more work piling up in front of it.

Real-world example of using AI and human review to accelerate a retail deal

To be sure, it is possible for a faster first pass to get lease abstracts reviewed and to the landlord sooner, as CREModels proved on a project covering about 200 retail tenants. It was a great example of AI-acceleration actually breaking a bottleneck. Madeline Miller, who leads CREModels’ lease abstraction practice, loaded a single ZIP file into a proprietary internal tool called Mercury, which sorted the documents by tenant and kicked out the first-pass abstracts. 

That left Maddie and her team more time to check the terms against the leases, including which amendments changed the rent and which exclusions applied to expense recoveries. The team finished in under two weeks. An abstraction process for a portfolio of that size could have taken up to three times as long if handled manually.

The division of labor matters here: An analyst can use AI to find the rent terms, check them against the lease and pass them to the model. The approved formulas then calculate the cash flows. (Read a companion article for more on how to keep the math consistent and guarantee accurate, reproducible results, even if the original analyst is OOO.)

Four ways to keep AI-assisted workflows focused on busting bottlenecks

  • Take a team-based approach to AI platform testing and standardization. If there is a six-week wait for a key step in a larger process, those who need it done will be more likely to try to bypass internal “middlemen” by quietly leveraging an LLM on their own (“I did it. Yes!”). Map the workflow and identify where work waits. Before choosing what to automate, measure the whole process: how long work takes from request to delivery, how much review it needs and how often it comes back for fixes. Then point AI at the step that holds everything up, not the one that is easiest to speed up. A supportive and encouraging organization-wide pilot program can focus the right people on the right problems.
  • Build lightweight review into the workflow. A team piloting a new AI workflow can keep review simple. One person who knows the work can often spot the omission that everyone else would miss. Sit down with the relevant subject matter experts to come up with a reasonable approach that preserves accuracy without entirely sacrificing speed. That might be, for example, a lawyer who says, “We need an entry in the template to make sure our limitation-of-liability language is in there!” 
  • Measure the gain across the whole workflow. Run a pilot on the next batch of work. Compare request-to-delivery time, review hours and corrections with the baseline. Then hand the next file to a colleague and see whether they can deliver it without calling the person who built the tool.
  • Build a library of approved tools for internal use. Once a second analyst can run the tool successfully, put it somewhere the rest of the team can find it. Give it an owner, keep a version history, and save a few real test cases with answers the reviewers have checked. Run them again whenever the underlying model changes.

(Pro Tip: These steps can be built into an agentic-optimization process that computer scientists refer to as ‘hillclimbing.’ More on this in a forthcoming piece …)

Churning out that first rent roll in minutes with AI was a thrill. Lasting satisfaction comes from knowing that a colleague can pick up the stress-tested tool you have created, check the rents and get them over to underwriting in record time, accelerating a game-changing deal.