AI can price the deal. Your IC still won't sign it.
The bottleneck was never speed, it's trust. Here's how to build an offer that survives the room.
There's a stat going around that an AI system can now underwrite a full mixed-use deal in ninety minutes, down from a week. It's true, and it's a real unlock. It also isn't the bottleneck, and mistaking it for one is about to cost people deals.
- Sasha Deneux
The Take: AI can price the deal in ninety minutes. Your IC still won't sign it.
The pitch this year is speed. Leni's head of industry says an agentic system runs a full mixed-use underwrite in about ninety minutes now, versus a week five years ago. Twenty more minutes for an analyst to check it, and you're done. That's the difference, he says, between bidding on three deals a quarter and fifteen.
I believe the number. I just don't think speed is the wall.
Here's the wall. You can generate a price in ninety minutes, but you still have to walk it into an investment committee that puts real capital behind it. Committees don't trust a number they can't take apart. Ask most acquisitions teams and the honest answer is that AI is welcome for the grunt work and quietly unwelcome the moment it touches the actual go/no-go. One audit partner put it flatly: he wouldn't hand AI twenty million dollars to invest. The machine can produce the answer. It can't own the decision.
So the constraint that matters isn't how fast you can price a deal. It's whether the price survives scrutiny. For AI to move upstream into valuation and the offer, every number has to be explainable, auditable, and traceable back to a source a human can check. A fast answer nobody can defend doesn't win the deal. It gets you a confident LOI your own committee kills.
Which is why the offer, not the underwrite, is where this gets won. So let's build one that survives the room.
The Teardown: the comp-backed LOI, end to end
You screened the deal, it came back HOT, now you have to make an offer that's aggressive enough to win and defensible enough to clear your own committee. Here's how to build the LOI without guessing.
Start with the comps, and make it show its work. Feed it your actual comp set: recent sales in the submarket for the price, in-place and asking rents for the underwrite. Tell it to pull each one with the source attached, the address, the date, the price per unit or per foot. Not "market's around a six cap." The five specific trades it's reading a six cap from.
Then the price, off your return, not its opinion. Give it your target: the going-in yield or price per unit you need. Have it back into the number that hits it, and show the two or three assumptions the whole thing rests on. Now you have a price with a spine, not a guess dressed up as precision.
Then the terms that signal you're real. Deposit, diligence window, closing timeline, financing contingency. A broker reads terms as fast as price. A clean structure separates a serious bid from a tire-kicker at the same number.
Then the note, three lines the broker will actually read. Not a basis-free lowball. "Here's our number, here's the comp basis, here's our timeline, we close." Specific enough they take you to the seller instead of the next email.
And the hard rule: every figure is either sourced to a comp or flagged CONFIRM. An offer built on a made-up rent doesn't just lose you money, it loses you the broker the first time diligence blows it up.
What comes back is an LOI your committee can take apart and still approve, because every number points somewhere. The machine assembled it. You signed it. That's the version that closes.
Signal
CRETI: AI-native proptech pulled about $4.5B in venture funding in 2025, and roughly $1.7B in January 2026 alone, a 176% jump over January 2025. Why it matters: the capital is betting on AI that touches underwriting and valuation directly. More funded tools plus faster underwriting means more buyers bidding, so your offer wins on basis and credibility, not speed.
Uniti raised a $12M Series A (led by Pathlight, with MetaProp) for AI agents that run landlord leasing and ops; it says one self-storage operator saw a 214% ROI and 306% net revenue retention. Why it matters: the ROI gets real when AI runs one loop end to end, not sprinkled across the org. Pick the loop, prove it, then expand.
Land-use and zoning intelligence is moving earlier into deal pricing, so buyers can price redevelopment potential and use restrictions at the offer stage instead of in diligence (Commercial Observer, Jul 2026). Why it matters: a HOT verdict that ignores what can actually be built there is a mispriced LOI. Entitlement belongs in the number, not a surprise after you're under contract.
McKinsey estimates generative AI could unlock $110B to $180B in value for real estate. Why it matters: the prize is real, but it's gated by exactly what the Take names, trust and traceability. The value shows up for the firms whose numbers a committee can actually audit.
From NextAutomation
An LOI that pulls your comps, prices off your return targets, and flags every number it can't source is exactly the kind of loop we build and run with acquisitions teams. If you'd rather find where AI pays back before you commit to a build, that's what the AI Opportunity Audit is for: two weeks, your opportunities ranked by dollar impact, and two working proofs on your own deals. Bring a deal you're about to bid on.
Book a callThe listing is the last step of a sale, not the first. Many of the signals that precede a sale sit in public records long before a listing exists, and anyone can read them.
Next week: the inbound side. Classifying every OM that hits your inbox HOT, WARM, or PASS against your buy box, with the reasoning attached and the broker reply drafted.
- NextAutomation Team