AI for Real Estate and Construction
The one sector where a specific AI use has been made outright illegal in a growing list of jurisdictions. Nine verticals, written for the person who has to sign off the deployment rather than the person writing the market outlook.
Property is where AI regulation stopped being theoretical.
In most sectors the AI compliance conversation is about frameworks arriving in 2027 and what documentation you will need. In real estate it is about rules already in force and litigation already decided. The interagency quality control standards for automated valuation models took effect on 1 October 2025. A tenant screening algorithm produced a $2.28m settlement and a five-year restriction on how the score can be used. Algorithmic rent-setting is now prohibited outright by ordinance in more than a dozen US jurisdictions, with a state statute under constitutional challenge. None of that is a forecast.
The practical consequence is that in this sector the legal question comes before the technical one, and not by much of a margin. Whether you can deploy a pricing model depends on where the asset is. Whether a screening model is defensible depends on whether anyone has tested it for disparate impact. We would rather establish that in week one than build something that has to be withdrawn, and on more than one occasion the honest answer has been that the system the client wanted cannot lawfully be built the way they described it.
Where AI actually earns its place in property and construction
Ranked by evidence rather than by how often it appears in a market outlook. The maturity column is our own read; the catch column is what the vendor deck leaves out.
| Where | What it does | Maturity | The catch |
|---|---|---|---|
| Document extraction from leases and contracts | Pulls dates, options, rents, obligations and clauses into structured data. | Proven | The strongest case in the sector. Long documents, structured facts, immediately checkable, and most portfolios have never had them in a database. |
| Valuation and price modelling | Estimates asset value from comparable transactions and attributes. | Proven | Statistically mature. In US mortgage use it is now a regulated activity with five mandatory quality control factors, one of which is nondiscrimination. |
| Building energy and plant optimisation | Tunes HVAC and plant against occupancy, weather and tariffs. | Strong | Real savings on real buildings. Requires working meters and controls, which is where most of these projects actually stall. |
| Project controls and schedule risk | Predicts cost and schedule outcomes from historical project data. | Strong | Works where projects are comparable and history is honest. Most cost overrun data is recorded after commercial negotiation, not before. |
| Site progress and safety monitoring | Reads site imagery for progress, plant and protective equipment compliance. | Strong | Effective, and it observes workers. Design it as a workforce intervention before a technology one. |
| Demand, leasing and churn modelling | Predicts renewal, vacancy and time to let. | Good | Useful. Transaction counts in commercial property are small, so honest uncertainty matters more than model choice. |
| Tenant and applicant screening | Scores rental applicants for approval. | Contested | Legally exposed. Fair housing disparate impact applies to the outcome regardless of what the inputs were, and this has been litigated. |
| Algorithmic rent and price setting | Recommends rents from market and competitor data. | Restricted | Prohibited by ordinance in a growing list of US jurisdictions. Whether you may deploy it depends on where the building is. |
| Generative design and automated planning | Produces layout and massing options against constraints. | Early | Genuinely interesting and rarely construction-ready. Treat outputs as options for a designer, not as documents. |
Two of these are legal questions wearing technical clothes
Screening and pricing are the two uses in this sector where the build decision is not yours alone to make. Rental screening sits under fair housing law, where a model that produces a disparate impact is exposed regardless of whether any protected characteristic appeared in its inputs — proxies are the whole problem, and it has already been litigated to a settlement. Algorithmic rent-setting sits under a patchwork of ordinances that did not exist three years ago and now covers jurisdictions from San Francisco to Seattle to Minneapolis. Both are buildable. Neither should be scoped without counsel in the room, and if a supplier tells you their product is fine everywhere, they have not read the ordinances.
The rules that already apply, and the ones arriving for buildings
Property sits under fair housing and lending law, a live and expanding body of algorithmic pricing regulation, building energy performance obligations, and construction product and machinery rules. This is our reading as at September 2026 and we work alongside your compliance and legal functions rather than in place of them.
In force since 1 October 2025
Six US federal agencies adopted quality control standards for automated valuation models used in mortgage credit decisions and covered securitisation determinations. Five factors: confidence in estimates, protection against data manipulation, avoidance of conflicts of interest, random sample testing and review, and compliance with applicable nondiscrimination laws. The fifth factor is the one most model teams have no evidence for.
The outcome is what is tested, not the inputs
A model that produces materially different outcomes across protected groups is exposed under the Fair Housing Act whether or not any protected characteristic was an input. Proxies — postcode, income source, credit thresholds that correlate with subsidy receipt — are the mechanism. Louis v. SafeRent settled at $2.28m in 2024 with a five-year restriction on using the score for housing voucher applicants, after the Justice Department filed a statement of interest.
A patchwork that decides deployment by address
Ordinances restricting rent-recommendation software using non-public competitor data are in force in San Francisco, Philadelphia, Providence, San Diego, Jersey City, Hoboken, Seattle, Berkeley, Minneapolis and Santa Ana, with Rockville following in 2027 and further bills in 2026 sessions. New York enacted a statutory ban that the principal vendor is challenging on First Amendment grounds. The DOJ settlement with that vendor in November 2025 restricted runtime use of competitors' non-public data and imposed a monitor — it did not make algorithmic pricing lawful everywhere.
Transposition landed 29 May 2026
The recast Energy Performance of Buildings Directive required transposition by 29 May 2026. New public buildings must be zero-emission from 2028 and all new buildings from 2030; the worst-performing 16% of non-residential stock must meet minimum standards by 2030 and 26% by 2033. Building automation obligations extend to systems above 70 kW by 2030, down from 290 kW. This creates measurement and reporting work before it creates modelling work.
Creditworthiness and critical infrastructure, not property generally
The AI Act does not name real estate. It reaches this sector through Annex III: creditworthiness evaluation for natural persons, which covers mortgage decisioning, and safety components in critical infrastructure, which can cover building and infrastructure control systems. High-risk obligations apply from 2 December 2027. Valuation, leasing analytics and project controls are ordinarily outside it — but the classification should be reasoned and recorded, not assumed.
Staged, and mostly not yet binding
The recast Construction Products Regulation became generally applicable on 8 January 2026, with digital product passport obligations staged through delegated acts under a working plan running to 2029 rather than fixed in the regulation. The Machinery Regulation applies from 20 January 2027 and reaches autonomous and remotely controlled site plant. Neither is an immediate AI compliance obligation; both change what data you will be expected to hold.
What this means for a build
Three things follow. Any model touching applicant selection or rent needs a legal position established before the first sprint, not before deployment — the answer changes what gets built. Any valuation model used in a US mortgage decision needs documented testing against the five quality control factors, including nondiscrimination, which usually means building the test harness before the model is worth having. And any system observing site workers should be designed as a workforce intervention first: what is captured, who sees it, what it is used for and what it will never be used for, agreed in advance. See AI policy development and AI bias audits.
Who we write for
Each page starts from that organisation's own problems, names the regulatory exposure it carries, and routes into the engineering. Depth varies and is stated on each page.
Residential real estate
Valuation under the AVM rule, lead qualification and listing operations — with fair housing testing built in, not bolted on.
Read more →Commercial real estate
Lease data extraction, portfolio analytics and leasing models — starting with the terms nobody currently holds in a database.
Read more →PropTech
Model engineering for property software — plus the legal constraints your customers are now obliged to ask about.
Read more →Construction & contracting
Estimating accuracy, schedule risk, site monitoring and document control — starting with the projects you already delivered.
Read more →Architecture & engineering
Practice knowledge retrieval, specification checking and fee analytics — with generative design placed where the evidence puts it.
Read more →Infrastructure
Inspection imagery, condition assessment and renewal prioritisation — with the advisory line drawn deliberately.
Read more →Facility management
Work order triage, maintenance planning and plant optimisation — instrumentation assessed before anything is modelled.
Read more →Property management
Maintenance triage, arrears and renewal modelling — with the screening and rent-setting law stated plainly.
Read more →Co-working
Utilisation, churn and capacity modelling — with pricing kept inside boundaries that now vary by city.
Read more →FAQ
Marked up with FAQPage schema so these answers can surface in search results and inside AI assistant responses.
Do you have real estate and construction experience?
Yes in lease and contract data extraction, valuation and price modelling, project controls analytics, building operations data and leasing forecasting. Less in structural engineering analysis, and none in building control system engineering — we would not claim otherwise. Each of the nine vertical pages states our depth in that area rather than implying uniform expertise across a sector that spans a residential agency and a tunnelling contractor.
Can we use AI to set rents?
It depends where the building is, and the answer is now genuinely different by city. Ordinances prohibiting rent-recommendation software that uses non-public competitor data are in force across a growing list of US jurisdictions, New York enacted a statutory ban currently under constitutional challenge, and the largest vendor settled with the Justice Department in November 2025 under terms restricting how competitor data may be used. Pricing from your own portfolio's data is a different proposition from pricing from a pooled competitor dataset, and that distinction is doing most of the legal work. Establish the position with counsel before scoping.
Is tenant screening AI worth building?
It is buildable and it is the most legally exposed system in this sector. Fair housing liability attaches to the outcome, so a model that never sees a protected characteristic can still produce a disparate impact through proxies — postcode, income source, a credit threshold that correlates with voucher receipt. Louis v. SafeRent settled at $2.28m with a five-year restriction on the score's use for voucher applicants. If you proceed, disparate impact testing is not a compliance add-on at the end; it is the thing that determines whether the model can exist.
What is the highest-value thing we could build this year?
For almost every property owner, structured lease data. Most portfolios hold their commercial terms as PDFs and a spreadsheet somebody maintains, which means nobody can answer a question about break options, indexation exposure or expiry concentration without a manual review. Extraction is mature, checkable, and it unlocks the analysis everything else depends on. It is unglamorous and it is the correct first project more often than anything on a vendor's roadmap.
Why do building optimisation projects stall?
Because the meters and controls are not there. A model that tunes plant against occupancy and weather needs sub-metering that works, a building management system that will accept a setpoint, and a maintenance regime that has not left half the sensors reading a constant. We assess that in the first two weeks, and it has ended engagements — the honest sequence is instrumentation first, optimisation second, and a supplier skipping to the second is selling you a model that will have nothing to act on.
Start with the problem, not the technology.
Thirty minutes, no charge, no deck. Tell us what is going wrong in your organisation and we will tell you whether AI is the right instrument — including when it plainly is not.