ATLANTA — (Propmodo) A developer I work with had a $40M multifamily deal that didn’t pencil. The team spent four months value-engineering it. Thinner finishes. Simpler amenities. Cheaper fixtures. They still couldn’t close the gap, so the deal sat.
The site wasn’t wrong. The market hadn’t turned. Rents could carry the project. But once the spec cuts were exhausted, nobody on the team could tell him whether a different unit mix, two fewer floors of parking, or a shifted footprint would have changed the answer. The numbers ran. The design held. Neither was talking to the other.
That deal didn’t stall because of bad judgment. It stalled because the workflow ran out of levers.
Development has always followed a relay race. Early underwriting sets a target. Design tries to hit it. The model gets re-run to see if it worked. Each leg runs with limited visibility into the others, and by the time someone realizes a 5% shift in unit count would have changed the outcome, the team has moved on or the option has expired.
AI is starting to collapse that relay into something closer to a single conversation.
What parallel evaluation actually looks like
Instead of modeling one configuration and stress-testing its pro forma, teams can now define the variables that matter—unit mix, program, height, code envelope, parking ratios—and generate dozens of feasible configurations simultaneously, each carrying its own going-in yield. You stop asking “does this design work?” and start asking which version works best, where it breaks, and what it depends on.
That distinction matters more than it used to. The inputs underneath CRE deals have become less reliable. AGC’s analysis of Bureau of Labor Statistics data shows aluminum mill shapes surged 28% from November 2024 to November 2025, driven by a 50% tariff on imports. Permit timelines range from under two weeks in Houston to over a year for discretionary multifamily projects in Los Angeles. NAHB-NMHC research found that regulations now consume 40.6% of total multifamily development costs. A pro forma built on one set of fixed inputs has always been a simplification. In this environment, it’s an expensive one.
AI doesn’t make that uncertainty go away. It makes it visible before a team has spent six months and seven figures in predevelopment. You can see that a deal holds up across cost scenarios, or that it only pencils if lumber stays below a threshold and approvals clear in under 14 months. You can see what your reversion cap rate has to be for the IRR and equity multiple to land where you told the LP they would. The downside case stops being a 10% rent haircut someone added at the end. It becomes the actual range of outcomes the deal has to survive.
Why this is different from previous tech
If you’ve sat through enough proptech demos, you know the pattern. The dashboard is faster, the report is prettier, the workflow is the same. Most technology the industry has adopted over the past decade sits on top of existing processes—better tools for asset management, faster lease-up systems, sharper energy monitoring. Useful work, but not work that changes how decisions get made at the front end of a deal, where the most consequential commitments get locked in.
AI operating at the intersection of design and underwriting is a different category. It doesn’t optimize a single step; it changes how the steps connect. In my work helping teams integrate building performance analysis with early-stage financial modeling, I’ve seen what shifts when those disciplines stop handing files back and forth and start working in the same environment. Speed of evaluation matters less than the quality of the questions a team can now afford to ask. Firms that treat AI as another software layer on an unchanged process tend to see incremental results. The value is in restructuring what the process produces.
What operators should consider
Scenario planning needs to become standard practice in early-stage evaluation, not something added when problems surface. And not the version where a team builds a base case and grudgingly creates a “downside” by trimming rents 10%. Real scenario planning means asking what happens if construction costs spike mid-build, if the municipality changes setback requirements after entitlement, if the target unit mix doesn’t match where demand actually lands, or if the next buyer’s property tax reset prices you out of your own exit. Running those scenarios used to take weeks of analyst time. That constraint is disappearing.
Information flow between disciplines has to change as well. If architects, development teams, and capital markets are still working from separate models reconciled at milestones, the benefits of any AI tool will be limited. The value comes from integration—design and financial data living together and updating in real time.
The firms that consistently outperform over the next cycle won’t be the ones with the most aggressive capital stack or the deepest relationships. They’ll be the ones who evaluated 40 viable versions of a project before their competitors evaluated one. The advantage in development is moving from execution to decision quality—from how well you build the plan to how rigorously you tested it before you committed to it.
That developer I mentioned? The deal is moving again. Not because the market changed. Because someone finally tested what would happen if the design did.