Real Estate
Last reviewed
May 13, 2026
Sources
84 citations
Review status
Source-backed
Revision
v2 ยท 5,415 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 13, 2026
Sources
84 citations
Review status
Source-backed
Revision
v2 ยท 5,415 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Real Estate ChatGPT Plugins
Real estate has become one of the most data heavy industries to adopt artificial intelligence, with applications spanning home valuation, mortgage underwriting, property management, brokerage workflows, generative interior design, construction monitoring, and smart building controls. The sector also produced two of the most consequential AI failures of the 2020s: Zillow's loss-making iBuying program, shut down in November 2021 after a $304 million inventory write-down, and the United States Department of Justice antitrust suit filed against RealPage in August 2024 over its YieldStar revenue management software, which the government alleged enabled algorithmic price coordination among competing landlords. The DOJ and RealPage reached a proposed settlement on November 24, 2025.
Real estate technology, often called proptech, currently sits between two pressures. Brokerages, lenders, and landlords have spent billions deploying large language models and computer vision systems to write listings, value homes, optimize rents, and approve mortgages. Regulators, state legislatures, and courts have begun limiting how those same systems can be used, especially when the inputs include competitively sensitive landlord data or when outputs can produce disparate impacts on protected classes.
AI in real estate covers eight broad use cases:
| Use case | Representative products |
|---|---|
| Automated valuation models (AVMs) | Zillow Zestimate, Redfin Estimate, Realtor.com Estimate, HouseCanary, Quantarium |
| iBuying and home flipping | Opendoor, Offerpad, the discontinued Zillow Offers |
| Property management and revenue management | RealPage YieldStar, Yardi Virtuoso, SmartRent, Latch |
| Mortgage underwriting | Rocket Logic, Better.com ChatGPT app, Maxwell AskMax, Blend |
| Brokerage productivity | Compass AI, eXp Mira, Side App, Coldwell Banker AI in Action |
| Generative interior design and virtual staging | InteriorAI, REimagineHome, Decor8, Collov, Foyr Neo, BoxBrownie |
| Construction site monitoring | OpenSpace, DroneDeploy, Buildots, Doxel, Procore Assist |
| Smart building automation | JLL Hank, BrainBox AI, Honeywell Forge, 75F |
Residential applications dominate consumer attention because the Zestimate has been on millions of homepages since 2006. Commercial real estate (CRE) systems are larger in dollar terms but less visible, with platforms like CoStar, Cherre, VTS, and CompStak running on proprietary lease and sale data that never reaches the public web.
The statistical foundations of automated home valuation predate the modern computer industry. Hedonic regression, the technique that decomposes a property's price into the value of its individual features, was first published by G. C. Haas in 1922 as a method for valuing farm land in Minnesota. Andrew Court applied the same idea to automobile prices in 1939, and the method spread to housing markets in the 1960s and 1970s.
In the 1980s, as computing power became cheaper and county assessor records were digitized, the first generation of automated valuation models appeared. Karl Case and Robert Shiller developed the repeat sales index that bears their names while studying Boston home prices, and in 1991 they co-founded Case Shiller Weiss with Allan Weiss to commercialize the data. Fiserv acquired Case Shiller Weiss in 2002, and S&P later licensed the methodology for the tradable S&P CoreLogic Case-Shiller index.
Through the late 1990s, AVMs were used mostly by mortgage investors evaluating risk on collateralized mortgage loan pools. The technology was institutional, expensive, and invisible to consumers.
That changed on February 8, 2006, when Zillow launched its website with a free Zestimate for roughly 40 million homes. Founders Rich Barton and Lloyd Frink, both veterans of Expedia, built the Zestimate around tax assessment data, comparable sales, and a public-facing map. For the first time, anyone could look up an estimated value for almost any house in the country without paying for an appraisal or hiring an agent.
The Zestimate's median error rate at launch was 13.6 percent for off-market homes. Zillow reduced the error progressively through the 2010s, partly by ingesting more on-market data and partly by re-architecting the model. In June 2021, the company replaced its gradient boosted machine pipeline with a neural network that the team internally called the Neural Zestimate. By 2023, the company reported a median error of roughly 6.9 percent for off-market homes and under 2 percent for actively listed properties.
Redfin launched the Redfin Estimate in December 2015, drawing on more than 500 data points per property and emphasizing accuracy on on-market homes. The current Redfin Estimate reports a national median error rate of about 1.98 percent for listed homes and 7.72 percent for off-market homes.
From roughly 2017 onward, deep learning began to appear in production real estate systems. Restb.ai trained convolutional image classifiers on listing photos to detect features such as granite countertops or hardwood floors. Zillow added a computer vision module that incorporated photo data into the Zestimate. Construction technology firms like OpenSpace and Buildots built 360-degree video pipelines that compared site progress against building information modeling plans.
The arrival of generative AI in 2022 and 2023 widened the use cases. Pieter Levels released InteriorAI in October 2022, using Stable Diffusion to redecorate room photos. Compass began integrating OpenAI models into its agent platform in August 2023. Restb.ai shipped automated property descriptions later that year. By 2024 most of the major brokerages, lenders, and CRE data vendors had at least one generative AI feature in production.
Automated valuation models estimate what a property is worth based on its physical attributes, location, and recent comparable sales. The models are owned by portals, banks, appraisal vendors, and government sponsored enterprises, and they are used for everything from web traffic to mortgage loss reserves.
| AVM | Operator | Launch | Stated accuracy |
|---|---|---|---|
| Zestimate | Zillow | February 2006 | About 6.9 percent median error off-market, under 2 percent on-market |
| Redfin Estimate | Redfin | December 2015 | About 1.98 percent on-market, 7.72 percent off-market |
| Realtor.com Estimate | Move (News Corp) | Mid-2010s | Powered by Quantarium, Collateral Analytics, CoreLogic |
| HouseCanary AVM | HouseCanary | 2014 | About 114 million properties covered |
| Quantarium QVM | Quantarium | 2010s | Hit rate above 96 percent in third party tests |
The modern Zestimate is the most studied AVM in part because its accuracy stats are public. Zillow's June 2021 architecture switch to a neural network folded in raw price history, photographic features extracted by computer vision, and tax data into a single deep learning pipeline. The company also runs separate models for unique features such as views or remodels.
HouseCanary, founded in 2014, sells AVMs to mortgage investors and capital markets firms. Its Comprehensive Valuation Report covers about 114 million homes. Quantarium uses an ensemble of machine learning models and an image recognition layer to assess condition from photographs; its valuations power Realtor.com's headline estimate alongside Collateral Analytics and CoreLogic.
IBuying, short for instant buying, is the practice of using an AVM to make algorithmic cash offers on homes, renovate them, and resell them within a few months. Opendoor opened the category in 2014. Zillow joined in April 2018 with Zillow Offers, hiring crews to refurbish purchases in Phoenix, Las Vegas, and other Sunbelt markets.
The program scaled aggressively through 2021. To attract more sellers, Zillow tuned the Zestimate to make more competitive offers and committed billions of dollars in inventory. By the third quarter of 2021, the company was paying above its own forecasted resale prices, holding too many homes, and stretching its renovation capacity. On November 2, 2021, Zillow announced it would shut Zillow Offers down. The third quarter earnings included a $304 million write-down on inventory purchased above current resale value. The company cited two issues: the Zestimate had overpredicted home values in a turning market, and operational throughput on renovations could not keep up with purchase volume.
Zillow eliminated about 25 percent of its workforce as part of the wind down. Total losses on the Zillow Offers business reached roughly $1 billion. The episode is one of the most cited cautionary tales for machine learning deployed against thin tails, because the model performed reasonably well in stable markets but degraded sharply when housing turnover and price velocity changed.
Opendoor remained in iBuying after Zillow's exit, although its volume contracted substantially through the 2022 to 2024 interest rate cycle. Offerpad continued operating at smaller scale.
Revenue management software lets multifamily landlords set rent for thousands of units using daily price recommendations from an algorithm. The dominant product, RealPage's YieldStar, has been at the center of a multi year antitrust dispute.
YieldStar originated in 2000 as a project led by Jennifer Nevitt at Bravo Strategic Marketing. Camden Property Trust acquired the technology, then sold it to RealPage on July 19, 2002. RealPage merged the original platform with a pricing engine developed by Jeffrey Roper, a former Talus and Manugistics director, and built the modern revenue management product. Through later acquisitions, including Lease Rent Options (LRO) from The Rainmaker Group, RealPage consolidated the multifamily revenue management market.
The basic idea: instead of letting on-site staff guess at rents, the software analyzes vacancy rates, lease expirations, market comps, and competitor pricing data to recommend a daily price for every unit. Landlords typically accept the recommendations through an auto-accept feature.
ProPublica published a long investigation on October 15, 2022 titled "Rent Going Up? One Company's Algorithm Could Be Why." Reporter Heather Vogell documented how YieldStar drew on non-public landlord data shared by participating clients, including actual rents collected, not just advertised rates. In Seattle, the investigation found, 70 percent of apartments in some submarkets were managed by ten property managers that all used RealPage. Tenants filed class action suits in late 2022 against RealPage and large landlords including Equity Residential, Greystar, and others.
The Department of Justice Antitrust Division opened an investigation in late 2022. On August 23, 2024, the DOJ filed a civil suit against RealPage in the Middle District of North Carolina alleging that the company's YieldStar and AI Revenue Management (AIRM) software violated the Sherman Act by facilitating coordinated pricing among competing landlords. The complaint argued that landlords agreed to share competitively sensitive information, that RealPage used the pooled data to generate recommendations, and that the design steered users toward accepting higher rents. Eight states joined: North Carolina, California, Colorado, Connecticut, Minnesota, Oregon, Tennessee, and Washington.
On November 24, 2025, the DOJ filed a proposed settlement with RealPage. The consent decree, scheduled to run for seven years with possible early termination after four, restricts RealPage to using lease data at least 12 months old, bans real-time lease data, prohibits geographic modeling below the state level, and requires auto-accept features to use user-set parameters with symmetric bounds. A court-approved monitor selected by the DOJ has access to RealPage's source code, model training documentation, and runtime logs. RealPage admitted no liability.
A related consent decree with LivCor, one of the largest multifamily landlords in the country, was filed alongside. Equity Residential agreed to pay $56 million to settle a related class action.
While the federal case proceeded, dozens of state and city governments moved to ban or restrict algorithmic rent setting outright:
| Jurisdiction | Action |
|---|---|
| San Francisco | First U.S. city to ban algorithmic rent setting tools that use non-public competitor data, 2024 |
| Philadelphia | City ordinance restricting rent fixing algorithms |
| Berkeley, California | Local ordinance; RealPage sued the city in 2024 alleging free speech violations |
| Minneapolis | Fourth U.S. city to pass a ban |
| Jersey City, Hoboken, Providence, San Diego, Seattle | Local ordinances passed in 2024 and 2025 |
| California | State amendments addressing algorithmic price coordination |
| New York | Donnelly Act amendment signed October 16, 2025, effective December 15, 2025; RealPage sued the state in late 2025 |
| Colorado | Governor Jared Polis vetoed a similar bill in May 2025 |
Yardi, RealPage's main competitor in property management software, launched Yardi Virtuoso in 2024 to add AI agents across leasing, accounting, and operations. SmartRent and Latch sell smart access, thermostat, and leak detection hardware paired with software dashboards. SmartRent introduced a conversational AI assistant in 2025 that lets property managers query device data in natural language. Latch focused on smart locks and intercoms before its financial troubles in 2023 forced a corporate restructuring.
Residential brokerages have shipped a stream of generative AI tools since ChatGPT launched in November 2022. The clear leader in feature count is Compass.
| Brokerage | AI product | Notable features |
|---|---|---|
| Compass | Compass AI | Listing description writer, Likely to Sell lead scoring, voice-activated assistant, OpenAI integration since 2023 |
| eXp Realty | Mira | AI assistant trained on eXp data, AI Accelerator training series with 11 trainers |
| Side | Side App AI | Document validation, offer extraction, signature auto-tagging, reporting insights |
| Coldwell Banker | AI in Action | University training program, AI Innovator of the Year award |
| Rocket Homes | Rocket platform | Integrated with Rocket Logic mortgage AI |
Compass began experimenting with OpenAI models in 2019, while GPT-2 was the state of the art. The company shipped Compass AI in August 2023 as a beta integration of GPT-4 into the Compass agent platform. Subsequent releases added a voice activated assistant, automated CRM workflows, and the Likely to Sell prospect scoring tool.
eXp Realty announced Mira and an AI Accelerator training program in October 2025. Side rolled out four AI features in its Side App in early 2026, including document validation and offer extraction. Coldwell Banker uses its Coldwell Banker University to run the AI in Action program and named Kanani Ching, a broker in Hawaii, its first AI Innovator of the Year in 2025 for building a custom GPT system serving roughly 95,000 agents.
MLS-level tools complement brokerage products. Restb.ai sells listing description generation and computer vision classification to multiple listing services across the United States, claiming up to 50 supported languages and a five times reduction in listing prep time.
Mortgage origination has long been one of the most paperwork heavy financial workflows, which makes it attractive for AI. Lenders use machine learning to verify income documents, classify uploaded paperwork, flag fraud, and predict default risk.
Rocket Companies introduced Rocket Logic in April 2024 as a proprietary AI platform combining more than 10 petabytes of data and 50 million annual call transcripts. The system uses deep learning to identify and classify documents, with computer vision models extracting fields directly from uploaded files. Rocket reported that by February 2024 the platform was automatically processing 70 percent of the 1.5 million documents received monthly, saving more than 5,000 hours of underwriter work in a single month. Rocket said the platform reduced team member touches per loan by about 25 percent year over year and helped the company close loans approximately 2.5 times faster than the industry average.
Better Home & Finance partnered with OpenAI to launch an app inside ChatGPT in March 2026. Marketed as a 47 second mortgage demo, the integration combines Better's underwriting engine with OpenAI's models to let loan officers at competing banks, brokers, and fintechs run their own underwriting on the Better stack. Better said partner lenders save an average of 21 days on underwriting cycles.
| Lender or vendor | Product | Focus |
|---|---|---|
| Maxwell | AskMax | Analytics concierge for small and midsize lenders |
| Blend | Blend AI | Document classification and verification |
| LoanDepot | Mello | Digital origination workflow with AI components |
| Pennymac | Servicing AI | Default prediction and call routing |
Maxwell, founded as a digital point of sale platform for community lenders, shipped AskMax to surface analytics across the loan pipeline. Blend reports that lenders using its platform reduce origination time by up to 30 percent.
Virtual staging adds digital furniture to empty listing photos. Until 2022 it was almost entirely manual, with editing shops like BoxBrownie charging roughly 20 to 30 dollars per image and turning work around in 12 to 48 hours. Generative AI cut both cost and time dramatically.
| Product | Launched by | Approach | Notes |
|---|---|---|---|
| InteriorAI | Pieter Levels, 2022 | Stable Diffusion based room restyling | Pricing from $249 per year, virtual staging mode for empty rooms |
| REimagineHome | ReimagineHome.ai | AI staging plus shoppable furniture links | Connects renders to real purchasable products |
| Decor8 | Decor8 AI | 56+ design styles, 30 second renders | $14.99 per month, marketed to real estate agents |
| Foyr Neo | Foyr | 2D to 3D to 4K render workflow | 60,000 model furniture library; plans from $22 per month |
| Collov AI | Collov | Virtual staging for empty rooms | Standard plan $16 per month |
| HomeDesignsAI | HomeDesigns.ai | 80+ styles, indoors and outdoors | Comprehensive house and garden coverage |
| Spacely AI | Spacely | Multi-tool AI rendering suite | Credit-based pricing as of September 2025 |
| BoxBrownie | BoxBrownie.com | Manual editor service with AI augmentation | Reputation for photorealistic results |
| Apply Design | Apply Design | Browser based AI virtual staging | Drag and drop simplicity |
| Virtual Staging AI | virtualstagingai.app | AI furniture insertion | Marketed to agents at scale |
Pieter Levels, the Dutch indie developer behind Nomad List and Photo AI, released InteriorAI in October 2022. The site went viral on Twitter/X and helped popularize generative interior design.
Decor8 emphasizes speed and is widely used by real estate agents on small budgets. Foyr Neo sits at the other end of the market with full computer aided design workflows for professional interior designers. REimagineHome adds a commerce layer by linking AI-generated rooms to actual purchasable furniture from partner retailers.
Construction is a large, fragmented industry that has been slow to digitize. Companies have used AI mostly to monitor jobsite progress, detect safety issues, and reduce rework.
| Product | Function | Capture method |
|---|---|---|
| OpenSpace | Visual jobsite intelligence | 360 video clipped to a hardhat |
| DroneDeploy | Site mapping, earthworks measurement | Drone imagery, orthomosaics, point clouds |
| Buildots | BIM comparison and predictive delay analysis | 360 cameras and machine learning |
| Doxel | Progress tracking and labor productivity | LiDAR plus computer vision |
| Procore Assist | Document retrieval and RFI drafting | Generative AI on Procore project data |
| Autodesk Construction Cloud | Design and document AI | Autodesk product suite |
OpenSpace clips a 360 camera to a worker's hardhat and captures continuous video as they walk the site. Computer vision stitches the footage into a navigable visual timeline tied to floor plans. Buildots cameras compare actual progress against the BIM model and flag schedule slippage. Doxel uses ground based LiDAR and image analysis to measure installed quantities such as drywall area or installed pipe, then compares those volumes to the planned schedule.
DroneDeploy operates a complementary outdoor product. Its drone imagery pipeline produces orthomosaics, 3D meshes, and point clouds for earthworks, civil sites, and exterior progress. DroneDeploy acquired StructionSite in 2022 to add an indoor 360 camera capture product.
Procore introduced Procore Copilot in 2024 and rebranded it as Procore Assist. The product retrieves data across Procore project records and drafts text for RFIs, daily reports, and safety documentation. Procore announced an expanded Procore AI platform with new agents at Groundbreak 2025.
Smart building AI applies machine learning to building automation systems, mostly to optimize heating, ventilation, and air conditioning. The dominant cost in a commercial building is energy, and HVAC accounts for most of it.
| Product | Operator | Approach |
|---|---|---|
| Hank | JLL Technologies | Digital twin and HVAC optimization, acquired by JLL in 2022 |
| Falcon | JLL Technologies | Generative AI platform integrating proprietary CRE data with LLMs |
| BrainBox AI | Trane Technologies | Deep learning HVAC autopilot, acquired by Trane in 2025 |
| Honeywell Forge | Honeywell | Energy analytics and fault detection across BMS hardware |
| 75F | 75F | IoT controllers and zone level optimization |
Hank, founded in Sacramento, sells an AI HVAC platform that plugs into an existing building management system, builds a digital twin in the cloud, and tunes setpoints automatically. JLL acquired Hank in January 2022 through its JLL Technologies subsidiary. Royal London Asset Management reported a 708 percent return on investment, 59 percent energy savings, and 500 metric tons of avoided carbon emissions per year at its 11,600 square meter Birmingham office building after deploying Hank.
JLL Falcon, announced in 2024, is a separate generative AI platform that combines JLL's proprietary CRE data with large language models for client-facing applications. Falcon powers the next iteration of JLL GPT.
BrainBox AI takes a similar approach to Hank. The company is headquartered in Montreal and reports HVAC energy savings of up to 25 percent and carbon footprint reductions of up to 40 percent in production deployments. Trane Technologies acquired BrainBox AI in 2025 and announced the BrainBox AI Lab.
Honeywell Forge runs on top of a much larger installed base of Honeywell building management hardware. It provides energy analytics, fault detection, and portfolio benchmarking. 75F sells IoT controllers and zone level cloud software, mostly in the U.S. mid-market commercial segment.
Commercial real estate operations run on proprietary lease, sale, and property data. The biggest vendors are CoStar, Cherre, VTS, and CompStak.
| Vendor | Product | Focus |
|---|---|---|
| CoStar Group | CoStar Suite, LoopNet, Apartments.com, STR, Ten-X | Dominant CRE data and listings, estimated 70 percent share among institutional users |
| Cherre | Property knowledge graph | Connects CoStar, Yardi, Trepp, MSCI/RCA, Green Street; AI entity resolution with about 97 percent accuracy |
| VTS | Deal management and leasing | AI deal sourcing and market insights |
| CompStak | Crowdsourced lease and sale comps | CompStak AI launched December 2025 with agentic semantic search |
Cherre's main offering is an AI entity resolution engine that matches the same property across CoStar, Yardi, Trepp, MSCI RCA, Green Street, and roughly 50 other data sources into a single property knowledge graph. The company reports about 97 percent accuracy on standard U.S. commercial property records. Cherre launched Agent.STUDIO in 2025 to expose its data through an agentic AI interface.
CompStak draws on tens of thousands of broker contributors who share lease and sale comps in exchange for access to peer data. CompStak AI, launched in December 2025, adds semantic search and a hybrid econometric and machine learning rent estimate at scale.
VTS sells deal management software to leasing teams and added AI-driven market insight features through 2025. CoStar's portfolio includes CoStar Suite, LoopNet, Apartments.com, STR, and Ten-X; the company has folded AI-powered search and listing tools into each product line.
Real estate AI has produced more public controversy than most other industries, partly because it directly affects rents and home prices.
The November 2021 shutdown of Zillow Offers, described above, is the canonical AVM failure. Zillow tuned its algorithm to win more offers, then took a $304 million inventory write-down in the third quarter alone. Total losses approached $1 billion before the wind down was complete. The episode is taught in business schools and quoted by machine learning researchers when discussing distribution shift.
The DOJ suit against RealPage, filed August 23, 2024 in the Middle District of North Carolina, is the leading U.S. antitrust case involving algorithmic pricing. The government's theory is that an algorithm trained on competitor data can serve as a mechanism for tacit price coordination even without explicit agreements among competitors. The proposed settlement of November 24, 2025 introduces a new set of operating constraints for the industry, including a 12 month data age minimum, a state level minimum geography, symmetric pricing guardrails, and a court-appointed monitor with access to source code. The case has been cited as the most important algorithmic pricing precedent since the digital cartel cases of the early 2010s.
In April 2024, the Department of Housing and Urban Development issued two guidance documents on the application of the Fair Housing Act to AI tools. The first addresses tenant screening systems that use machine learning on credit, eviction, and criminal history data, where bad or incomplete inputs can disproportionately exclude Black and Hispanic applicants and people with disabilities. The second addresses advertising platforms that target housing ads, where algorithms can exclude households with children, service animal users, or specific language groups in ways that violate the Fair Housing Act. HUD recommends best practices including disparate impact testing and transparency on inputs.
Mortgage lenders deploying AI underwriting face fair lending scrutiny from the Consumer Financial Protection Bureau and federal banking regulators. The Equal Credit Opportunity Act and the Fair Housing Act both prohibit discrimination in credit decisions, and adverse action notices must include specific reasons. Generative AI tools used for income verification, document review, or pricing have to satisfy these statutory rules even when models are opaque.
Researchers cite the Zillow Offers shutdown as a textbook case of an out-of-distribution problem in deployed machine learning. The Zestimate was trained on prior sales and worked well in stable price environments, but it underestimated the cost of operational throughput and overpredicted in a turning market. Several academic papers in 2022 and 2023 used the Zillow case to motivate research on monitoring concept drift.
Generative virtual staging has prompted debate inside the National Association of Realtors and several MLS organizations about disclosure. Some MLSs now require an "AI staged" tag on photos that include digitally added furniture. The concern is that buyers may make offers based on idealized images without understanding which features are real and which were rendered.