A robotic arm holding a house, symbolizing how AI is influencing real estate.
getty
Moody’s bought Cape Analytics in 2025 for a reason nobody at a traditional property closing would expect: the startup reads rooftops, parking lots and yard debris from aerial images, then spits out a property risk score without anyone setting foot on the ground. When a company worth $80 billion buys that capability, it tells you where real estate lending is heading. Finance wants data it can scale. Realtors sell local knowledge and handshakes. Those two worlds are colliding. The money side is winning.
What Lenders Actually Use
Property lenders have three new toolkits, and each one chips away at what used to require a human on-site.
Start with geospatial imagery. Cape Analytics, now inside Moody’s, uses computer vision to scan aerial photos and flag roof conditions, lot changes, solar panels, overhanging trees and signs of neglect on an address-by-address basis across the U.S., Canada and parts of Australia. SpaceKnow does something similar for construction monitoring, tracking building progress with satellite time-series data so lenders know whether a project is on schedule or stalling. Orbital Insight, which was founded in 2013, built a foot-traffic analytics platform that processes billions of anonymized mobile location data points daily to give CRE investors real-time visitation estimates, dwell times and trade-area maps.
Next: lease intelligence. Vendors like Leverton and MRI Software use natural language processing to extract rent schedules, renewal options, co-tenancy clauses, early-termination triggers and compliance deadlines from commercial leases automatically. One global property firm used Leverton’s tool to process 40,000 legacy leases across 18 markets, cutting manual review time by 85% and uncovering $2.4 million in missed escalation revenue. V7 Labs reports that AI lease abstraction tools now routinely hit accuracy rates above 99%, with processing times dropping from 4-8 hours per document to minutes.
Last is alternative data fusion. JLL acquired Skyline AI in August 2021 to bring machine learning into CRE investment analysis. Skyline’s platform processes data from over 300 sources and tracks 10,000 attributes—owner information, demographics, historical transactions—across 400,000 U.S. multifamily properties. Zillow’s Zestimate, the most visible consumer-facing example, now uses a neural network trained on data from over 124 million U.S. homes. A 2025 NYU Stern study published in Marketing Science found that Zestimate increased buyer surplus by 5.94% and seller profits by 4.36%, with the biggest gains going to lower-income neighborhoods where price uncertainty was highest.
MORE FOR YOU
Standout fact: Skyline AI’s platform tracked 10,000 attributes across 400,000 multifamily properties before JLL folded the technology into its advisory tools, giving its brokers a quantitative edge their competitors didn’t have.
How This Changes The Lending Workflow
Speed and scale are the practical upshot. A lender who can scan a satellite image, pull lease data and overlay foot-traffic patterns doesn’t need to wait for a site visit to triage a deal. Automated pre-screening filters the portfolio down to the files that actually need human attention.
Ongoing monitoring is where AI saves the most money. Instead of sending inspectors to properties on a quarterly rotation, lenders can get continuous updates. SpaceKnow’s radar imagery works through cloud cover, so even weather doesn’t create blind spots. For loan servicers managing thousands of properties, replacing periodic inspections with near-continuous monitoring cuts operational costs and catches problems earlier, which means fewer surprise workouts.
Freddie Mac clearly sees this direction. In March 2025, the agency published a formal AI/ML governance framework in its Seller/Servicer Guide, followed by Bulletins 2025-16 and 2025-17 in December 2025. Every seller and servicer must comply by March 3, 2026. Requirements cover AI governance, risk management, monitoring and accountability. Shadow-IT AI projects and unvetted vendor tools are no longer acceptable.
The Gaps Lenders Need To Watch
None of this runs perfectly. Satellite imagery can misclassify objects. There’s a lag between when an image is captured and when it’s analyzed. World Bank-funded research led by Jean et al. showed that satellite-based economic inference has real limits, particularly at granular levels where ground-truth data is thin.
Data quality in commercial real estate remains a persistent obstacle. As ULI’s Urban Land Institute reported in 2024, CRE data lacks standardization. AI models trained on inconsistent records can produce inconsistent outputs. Matias Recchia, CEO of AI investment firm Keyway, told ULI that AI adoption in real estate investing is still “in diapers.”
Then there’s the bias problem. A 2025 Shelterforce report found that the homeownership gap between Black and white households is larger today than in 1960, and AI models trained on historical data risk baking those disparities in further. There is currently no federal guidance requiring AI in mortgage lending to be fair by design.
Where Realtors Still Matter
Realtors aren’t disappearing. An algorithm can flag a vacancy trend, but it can’t renegotiate a lease or talk a tenant through a renewal. Local market knowledge, on-the-ground verification when the satellite flags something unexpected and the relationship work that gets deals closed still require a human. The role shifts, though: fewer cold inspections, more targeted responses to what the AI surfaces.
Consolidation tells the story. JLL didn’t buy Skyline AI to replace brokers. It bought Skyline AI to arm them. Richard Bloxam, JLL’s CEO of Global Capital Markets, said the acquisition gives clients insights that are “beyond human” when paired with on-the-ground expertise.
The Bottom Line
If you invest in, lend against or manage commercial property, these tools are already shaping how your deals get priced and monitored. Ask your lender what models they use. Ask your servicer whether they’ve met Freddie Mac’s March 2026 compliance deadline. The more you understand about how these systems score properties, the better positioned you are to push back when the data gets it wrong.