Verified intelligence · Expert-reviewed from public records, platform data, primary sources, and private research beyond AI search.

MODELSReal estateGLOBAL

AI models for real estate

Routes AI models for residential and commercial real estate across AVMs, source data, search, tenant access, brokerage, lease intelligence, CRE investment, AECO, climate risk, property computer vision, controls, vendor economics and cost per accepted property output.

The colophon

Landscape reviewed: 2026-05-31
Latest revision: 2026-05-31
Kinetics: Real-estate AI has moved from public estimates and broker helpers into routes that can affect credit, tenancy, leasing, insurance, capex and investment decisions. AVM controls, fair-housing guidance, property-data standards, source-provider releases and model-provider terms are changing faster than annual procurement reviews.
To watch: FHFA automated-valuation-model quality-control implementation update; valuation routes could require stronger testing and nondiscrimination evidence ²⁵¹.
HUD tenant-screening guidance page revision; search, screening and lead-routing routes could require new fair-housing evidence ²⁵.
Fannie Mae value acceptance or Freddie Mac ACE policy release; mortgage collateral routes may change eligibility or review thresholds ⁶⁷.
RESO Data Dictionary release note changing listing-field coverage; listing routes could lose approved-field coverage ¹³.
First Street Risk Factor methodology update; property-risk routes could require stricter provenance evidence ²³.
Known unknowns: Comparable public evidence remains thin for accepted-output economics, protected-class proxy tests, AVM sample-validation results, lease-extraction error rates, climate-score validation, image-condition provenance, tenant-screening appeals and CRE workflow audit exports.

Contents

Executive summary
Board capital allocation
Model economics
Decision frame
Route approval map
Valuation and credit
Search and consumer routes
Implementation routes
Property intelligence
Data standards and source rights
Controls and failure modes
Private eval design
Throughput and rate limits
Known unknowns
Recommended next steps
Vendor and source directory
Source ledger
Glossary

▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒Executive summary

Buy real-estate AI by scarce evidence. The executive decision is which route has source rights, review capacity and a measurable accepted property output. The scarce asset is the right to use property records, listing fields, mortgage data, lease documents, comps, geospatial context, building photos, climate models and workflow logs. Treat Cotality/CoreLogic and ATTOM as property-data inputs ³³³⁴, MISMO and RESO as standards inputs ³⁵¹³, and CoStar/Homes AI, Dealpath/MSCI and Esri/Autodesk Forma as workflow-source inputs ³⁸³⁹ ⁴⁰.

The board should split spend by consequence. Mortgage AVMs need quality-control and nondiscrimination evidence. Search, lead routing, leasing and tenant routes need fair-housing, settlement-aware and pricing-boundary controls ¹²⁵. NAR settlement scripts and compensation explanations need broker-approved workflow evidence ³⁶³⁶. CRE underwriting needs licensed source data, lease extraction and reviewer logs. Climate, ESG, spatial and computer-vision routes need method, time horizon, asset condition and provenance checks.

Use foundation models for low-consequence drafting and retrieval when retention, connector, OWASP and audit controls pass ⁵⁰. Buy vertical or proprietary-source routes when data rights, market coverage, workflow logs or regulated evidence are the edge. Block routes that set rents, deny housing, move credit, clear collateral or change insurance without route-level proof.

Real-estate model routing tree

Start with the decision consequence, then choose the model lane.

Will the output affect credit, collateral, tenancy, insurance or pricing?
Yes
Use a regulated route with testing, review, adverse-action or fair-housing evidence.
No
Continue to source and workflow checks.
Next: Q2 source
Does the output depend on property records, listings, leases, market data, photos or hazard data?
Yes
Approve the source model and data licence before the generator.
No
A controlled general model may be enough for drafting or Q&A.
Can the reviewer trace the answer to fields, pages, images, geocodes or model methodology?
Yes
Pilot with accepted-output logging.
No
Block production use until provenance is visible.

The route tree keeps valuation, search, CRE workflow and property-intelligence uses from being approved as one model.

Approval verdict

Decision	Route	Control
Fund	Source-provider and vertical routes where data rights are scarce	Property coverage, source licence, update cadence, audit export and accepted-output economics.
Approve	Low-consequence drafting and internal property Q&A	Enterprise model terms, OWASP application tests, source citations and reviewer approval.
Pilot	Lease, rent-roll, OM, comps and CRE memo automation	Extraction replay, source pages, licensed comps, reviewer edits and cost per accepted memo.
Restrict	Public estimates used for investor or lender decisions	Private backtests, asset-type slices, agency policy and appraisal fallback.
Block until proven	Rent setting, tenant screening, steering, credit, collateral and insurance outputs	Fairness, nondiscrimination, adverse-action, appeal, antitrust, pricing-data and sample-testing evidence.

Related report

AI models for financial services

Financial model routes for source authority, customer impact, model risk, retrieval quality, benchmark evidence and accepted financial outputs.

Related report

AI agents for real estate

Agent and workflow approval across search, brokerage, leasing, property management, mortgage, valuation and commercial real-estate operations.

Related report

Deployment options for open-source LLM models

Use the deployment report for licence, hosting, managed endpoints, cloud catalogues, serving engines, RAG, gateways, GPU economics, controls and recovery.

Where to buy, build, gate or block real-estate AI spend. Capital should go to routes that turn source rights into accepted property decisions. Split spend by operating lane. The collateral team needs AVM validation. Brokerage needs source and conduct controls. Property operations needs PMS-integrated workflows. CRE investment needs licensed comps and document intelligence. Development needs geospatial and planning context. Corporate functions need secure retrieval and drafting.

That split also clarifies build-versus-buy. Buy where the vendor owns hard-to-recreate source coverage or workflow instrumentation. Build where proprietary underwriting, asset strategy, local policy or portfolio data is the edge. Gate any route that changes housing access, pricing, credit, collateral, insurance or public disclosure ³³³⁹.

The budget owner should approve a route only when the denominator is named: accepted collateral decision, accepted lease term, qualified inquiry, resident case, investment memo, risk flag or development assumption.

Capital allocation by lane

Lane	Executive owner	Buy / build posture	Spend gate	Decision
Collateral and AVM	Chief credit officer / valuation lead	Buy agency-compatible and source-provider routes; build only validation, monitoring and portfolio replay.	Private error slices, sample review, nondiscrimination evidence and fallback appraisal path.	Approve only with AVM controls.
Brokerage search and portals	Chief digital officer / brokerage president	Buy portal or MLS-aligned routes; configure ranking, disclosure and human handoff.	Fair-housing query tests, listing-source rights, NAR settlement scripts and handoff logs ³⁶³⁶.	Pilot for assistive search; gate lead routing.
Leasing and property operations	COO / head of property management	Buy vertical PMS-integrated routes where source rights and workflow logs are inspectable.	Tenant-access parity, appeal path, maintenance escalation, pricing boundary and resident-service record.	Approve service routes; block rent coordination.
CRE investment and asset management	CIO / head of acquisitions	Buy source-provider and deal-workflow routes; build proprietary underwriting overlays.	Licensed comps, lease abstraction replay, analyst edit rate and investment-committee acceptance.	Scale by accepted memo economics.
Development and AECO	Head of development / construction	Buy geospatial and BIM-adjacent source routes; build site-specific feasibility workflow.	Parcel, zoning, land-use, environmental, infrastructure and design-context lineage.	Pilot as planning evidence, not valuation fact.
Foundation-model knowledge work	CIO / CISO / legal	Buy enterprise model access when retention, region, connector and audit controls are provable.	Source citations, connector least privilege, OWASP tests, reviewer approval and deletion terms.	Approve for low-consequence drafting and retrieval.

Model-market map

Source moat

Workflow moat

Control moat

Blocked action

AVM and property dataCotality/CoreLogic, ATTOM and agency routes.

StrongDeep records and refresh

PartialOriginations or portfolio route

GateSample review and fairness

GateUnreviewed collateral decision

CRE investmentCoStar, MSCI, Dealpath, JLL, CBRE and internal asset data.

StrongLicensed comps and market data

StrongDeal screening and IC workflow

PartialAnalyst edit and source logs

GateUnlicensed comps in memo

Leasing and operationsPMS, prospect, resident and maintenance routes.

PartialResident and unit records

StrongPMS-native tasks

GateParity, appeal and audit

GateRent coordination or opaque denial

Development and AECOGeospatial, parcel, zoning, BIM and infrastructure context.

StrongParcel and GIS context

PartialSite planning and feasibility

StrongZoning and method lineage

GateDesign output treated as legal entitlement

Spend advantagePilot with proofApproval gateBlocked decision

Price the accepted property decision after review. Use the accepted property output as the economic unit: verified valuation, accepted lease term, qualified handoff, resident case, investment memo or risk flag. Real-estate work has a verification tail. A model output often needs source inspection, field reconciliation, fair-housing review, appraiser or analyst judgment, policy fit, appeal handling, approval and record retention.

That tail can dominate model price. A cheap model that sends reviewers back to the lease, comp set, listing record, inspection photo or hazard source five extra times loses to a more expensive route that returns the right evidence with a usable source trail.

Artificial Analysis-style comparisons are useful because they put price beside intelligence, output speed, latency and context ⁶³⁶⁴. Real-estate buyers need one more column: accepted property output, measured as a reviewed valuation, reconciled rent-roll row, approved lead handoff, resident case, asset memo or risk flag.

Procurement should run a cost-per-accepted-output eval. Count source licences, false answers, refusals, missing-source reports, reviewer minutes, reruns, latency, throughput limits, token spend, tool calls, appeals and audit-export quality.

Token price misses property work

Two ways to read model cost. Left: the sticker price, what the provider charges per million tokens blended at the standard 3:1 input-to-output ratio. Right: the bill for running the full Artificial Analysis Intelligence Index on that model. Data checked on 31 May 2026. Across the six bands shown, the median cheapest token price has fallen about 120x since March 2024; the median benchmark bill has fallen about 25x. The two prices come apart because models use different amounts of input, reasoning and answer tokens. Cheap tokens do not always mean a cheap finished job, and pricey tokens do not always mean a pricey one. Each line tracks the cheapest model available in a given intelligence band; lower is cheaper. Data from Artificial Analysis ⁶³; intelligence bands after a16z ⁶⁴.

Sticker price is only the first cost signal. Lane choice determines whether the same workload is interactive, batched, cached, flexed, provisioned or delayed, which changes both the bill and the review burden.

The best discount for lease abstraction, rent-roll imports, listing refreshes, tenant-file batches and private eval reruns is often patience. Long extraction packets, eval reruns, classification sweeps and archive enrichment do not need a user waiting in the browser. Move that work to async or lower-priority lanes, keep deadline work on the fast path, and compare cost per accepted property output.

The discount lane is part of the route

Lane	How it works	Good for	Watch
OpenAI Batch API	Async jobs use a separate batch lane with a lower price and a 24-hour completion target. ⁵³	Large document extraction, eval reruns, classification, enrichment and non-urgent archive work.	Not an interactive path; queue time, file packaging, partial failures and output reconciliation become part of the route.
OpenAI Flex processing	Eligible calls use lower-priority capacity and are charged at Batch API rates for supported models. ⁶⁵	Long-document async work where cost matters more than immediate availability.	Treat it as a cheaper lane with lower priority, not as the path for a signing call, filing deadline or live customer answer.
Anthropic Message Batches	Bulk async requests can be priced below the standard synchronous lane and can be combined with prompt caching. ⁶⁷⁶⁸	Large non-interactive packs: extraction, summarization, evals, source tagging and overnight review queues.	Batch queue behaviour and cache checkpoint design decide whether the savings survive production.
Gemini / Vertex AI batch	Vertex batch prediction is the non-real-time lane; Google pricing separates standard, batch and cached-input treatment. ⁶⁹⁷¹	Bulk document analysis, classification and back-office model calls that do not need immediate response.	Batch mode, provisioned throughput and cached context are separate mechanisms; do not assume one discount buys the others.
Amazon Bedrock batch inference	Bedrock batch inference runs async jobs over S3 inputs, with discounted batch pricing for supported models. ⁷²⁷⁴	Cloud-controlled bulk inference where the input/output artefacts can live in the approved AWS path.	Prompt caching and batch inference are not universally combinable; model, region and feature support decide the real lane.
Groq Batch and Flex	Groq's Batch API is a discounted async lane; Flex processing trades immediacy for access to available capacity rather than a lower price. ⁷⁶⁷⁷	Throughput-sensitive work where the buyer can separate background processing from live assistant use.	Low published developer limits can still make a clever route unusable until the account, plan or enterprise path changes.
Usage tiers and quota requests	Spend tiers, usage tiers, quota requests and account-specific limits can change the ceiling before any model changes. ⁵²⁵⁴⁵⁵⁵⁶⁵⁸⁶⁰	New projects that need to absorb a sudden file drop, launch backlog or review queue.	Tier language is not capacity evidence. Check the live project, account, subscription, region and deployment type.
Provisioned or reserved capacity	Azure PTUs and reservations, Vertex Provisioned Throughput and Bedrock Provisioned Throughput buy predictability rather than a smarter model. ⁵⁷⁵⁹⁶²	Recurring volume, deadline-sensitive work and routes where shared-capacity 429s would be business failures.	Commitments, idle capacity, regional availability, spillover rules and reservation terms belong in the price model.

Prompt caching is easy to misunderstand. It is not model memory and it does not cache the answer. It reuses the provider's work from the repeated beginning of a prompt: instructions, tool schemas, output schema, playbook text, matter taxonomy or source bundle. If the next request starts the same way, the provider can bill and process that repeated prefix differently. If the route prepends a timestamp, request id or one-off user sentence before the stable material, the cache may miss.

Design the prompt like a production artefact. Put the durable, reusable context first; put the unique question last; record cache hits in the real-estate model route register. Cache savings are real only when the route keeps a stable prefix and still produces accepted property output.

How to make caching work

Design rule	Why it matters	Implementation
Put stable material first	Most cache mechanisms reward an identical or explicitly marked prefix.	Start with system instructions, tool schemas, output schema, jurisdiction rules, playbook text and reusable source packs.
Put variable material last	A unique timestamp, request id, user aside or matter-specific fact at the top can break a prefix match.	Append the specific question, run id, new facts, reviewer note and requested output after the cacheable prefix.
Keep tools and schemas stable	Tool definitions, JSON schemas and message ordering can be part of the cache key.	Version schemas deliberately; avoid rebuilding tool lists, whitespace, examples or column order on every call.
Warm the cache before fan-out	Parallel extraction can waste money if every worker pays to prefill the same long context.	Send one cacheable request for the source pack or playbook, then fan out clause, entity, citation or issue tasks.
Cache source packs, not shortcuts	The cache should preserve approved context, not hide missing retrieval or stale policy.	Attach source version, route version and expiry; rebuild the prefix when authority, playbook or source set changes.
Measure cached-token economics	A cache discount that never hits is procurement fiction.	Log cache reads, writes, cached tokens, hit rate, latency and accepted-output rate beside the normal token bill.
Respect the data boundary	Caching is still provider processing under the approved route.	Cache only context that is allowed inside the approved data boundary. Caching is a billing and latency control, not permission to reuse one client's facts in another client's work.

Provider caching is not one thing

Provider	Mechanic	Buyer read
OpenAI	Automatic prompt caching applies to repeated prompt prefixes above the documented threshold; usage surfaces cached tokens, and prompt-cache keys can help route related work. ⁶⁶	Keep the first part of the prompt identical: instructions, tools, schema and reusable source context before matter-specific text.
Anthropic	Claude uses explicit cache-control breakpoints over tools, system and message content, with separate cache write and read economics and documented cache durations. ⁶⁸	Design cache checkpoints deliberately. The best target is the long shared context, not the short one-off question.
Gemini / Vertex AI	Vertex supports context caching, including explicit cached content and implicit prefix caching; pricing separates cached input from ordinary input and batch treatment. ⁶⁹⁷⁰	Check whether the route uses explicit cache objects, implicit prefix reuse or batch prediction; each has different cost and expiry behaviour.
Amazon Bedrock	Bedrock prompt caching uses cache points with time-to-live controls and separate cache pricing, while batch inference has its own feature constraints. ⁷²⁷³⁷⁴	Do not assume a cached interactive prompt can simply move into batch. Confirm feature support by model and region.
Groq	Groq prompt caching is automatic for exact repeated prefixes above the documented threshold, with discounted cached tokens and separate treatment in rate-limit accounting. ⁷⁵	The route must keep a stable prefix. This matters most when rate limits are low and large token packets would otherwise queue.

The practical question is whether the buyer can deliberately choose the cheap lane without breaking the work. A cheaper async run is useful for overnight extraction. It is dangerous for an answer needed during a signing call. A cache hit is useful when the prefix is stable. It is imaginary when every prompt is rebuilt from scratch.

What to prove before calling the route cheap

Test	Proof	Pass condition
Interactive versus async split	Tag each workflow step as user-blocking, same-day, overnight or archive work.	Only non-blocking work moves to batch, flex or lower-priority processing.
Batch economics	Run the same packet through interactive and batch lanes, including failed files, retries, queue time and output handling.	The cheaper lane still produces accepted property output inside the work window.
Cache hit rate	Log cache reads, writes, cached tokens or provider-equivalent fields by route version.	The actual hit rate is high enough to change the finished-task bill, not just the theoretical token price.
Tier and capacity path	Capture account tier, quota, regional capacity, provisioned-capacity option and support escalation.	The buyer knows which spend, quota or reserved-capacity lever unlocks the next workload level.
Accepted-output denominator	Combine model spend, queue time, cache hit rate, reviewer minutes, failed runs and rework.	The route is cheaper per accepted property output, not merely cheaper per million tokens.

Cost per accepted output

Cost item	Token view	Accepted-output view	Measurement
Source licence	Not counted in model price.	Can dominate AVM, comps, listing, climate, image and parcel routes.	Cost per accepted valuation, comp set, risk flag or asset memo.
Retrieval or source miss	Cheap answer.	Expensive wrong property, stale listing, bad comp, missing clause or unsupported climate flag.	Correct-source recall, no-answer rate and reviewer rejection reason.
Reviewer minutes	Outside the prompt bill.	Appraiser, analyst, broker, property manager or compliance review becomes the real denominator.	Reviewer minutes per accepted output and override rate.
Appeals and disputes	No direct token signal.	Tenant, borrower, buyer, owner or investor challenge creates rework and legal exposure.	Appeal volume, correction rate, adverse-action support and audit-reconstruction time.
Throughput and latency	Average price per call.	Listing refreshes, leasing queues, portfolio batches and deal-screening deadlines need enough capacity.	Accepted output per hour at p95 latency and approved quota route.
Model and source change	New price or model version.	Reapproval, holdout replay and source-methodology review can erase cheap-token savings.	Holdout rerun cost and reapproval burden after material route change.

How to price an accepted output

Workflow	Why token price misleads	Buyer test
AVM and collateral support	The model call is small compared with source coverage, sample review, appraisal fallback and fair-lending evidence.	Cost per accepted collateral decision after overrides, defects and reviewer minutes.
Lease and rent-roll extraction	Bad scans, amendments and reconciliation can move the bill from tokens to analysts.	Accepted clauses or rows per analyst hour with post-close correction rate.
Consumer search and leasing	A cheap assistant can still create fair-housing, disclosure, appeal or pricing-boundary work.	Qualified inquiry or resident case resolved after parity, escalation and audit gates.
CRE investment memo	Licensed comps, source export and committee-ready assumptions matter more than answer cost.	Accepted memos per analyst week with source-defect and committee-rework rate.
Climate, ESG and property intelligence	Peril methods, inspection evidence and official-data comparison sit outside the token bill.	Verified asset action or risk flag tied to method, source and owner approval.

Separate the model by source and consequence. Real-estate model selection starts below the logo: what data is allowed in, what output comes out and who is harmed if it is wrong. Match the demo to the decision. A direct API can summarise a lease while the data licence lives elsewhere. A source provider can know the market while exposing too little method detail for a regulated decision. A brokerage assistant can speed a workflow while lead-routing or screening still needs conduct evidence.

Use six registers. One covers AVMs and mortgage collateral. One covers consumer search, tenant screening and lead routing. One covers leasing and property operations. One covers commercial-real-estate workflow: leases, OMs, comps, diligence and asset management. One covers property intelligence: photos, spatial data, ESG, hazards and climate risk. One covers corporate knowledge work on approved model platforms.

Each register should name source class, source licence, model lane, control plane, user role, output type, reviewer, log fields, fallback and reapproval trigger.

How property-model approval gets made

Layer	Controls	Advantaged route	Buyer read
Source	Listing fields, mortgage standards, public records, agency rules, leases, comps, photos, hazard data, geospatial context and licences.	RESO and MISMO-aligned data, agency collateral systems and specialist source providers.	Bad field semantics make fluent answers incomparable.
Model	Foundation model, AVM, extractor, computer-vision model, climate score or vertical workflow.	The model class that matches the output.	A generator may be only the last step.
Control plane	Retention, region, connectors, identity, audit export, source lineage and appeal records.	Cloud and enterprise routes with evidence the buyer can inspect.	Cloud procurement still needs source-rights proof.
Review	Appraiser, underwriter, broker, analyst, property manager or compliance review.	Routes with rejection logs and accepted-output metrics.	Reviewer minutes are part of the cost model.

Source and model comparison

Route	Best for	Approval test	Caution
General frontier model	Drafting summaries, broker questions, market-note first passes	Retention, connector scope, source citations, reviewer approval	Can sound authoritative while using stale, unlicensed or incomplete property data
Cloud model platform	Enterprise document workflows and controlled internal assistants	Region, logging, retention, identity, DLP, audit export	Cloud control plane does not solve data-rights or fair-housing risk
Vertical real-estate SaaS	Valuation, leasing, deal flow, portfolio and brokerage workflows	Source licence, workflow logs, customer-impact controls, exit path	Strong workflow evidence may hide weak model transparency
Proprietary source model	Property-data, climate, hazard, computer-vision and CRE analytics	Methodology, update cadence, geography, sample validation, explainability	The source model can become the decision without being treated as a model
Property-data and AVM provider	Residential valuation, risk analysis, customer acquisition, portfolio monitoring and source enrichment	Coverage, data age, confidence bands, use-case tuning, sample testing and redistribution rights	The same AVM family can require different controls across consumer estimate, originations and risk routes
PMS-integrated leasing route	Prospect response, tour scheduling, resident service, payments, maintenance and operations	Fair-housing parity, appeal records, pricing boundary, resident-service logs and human escalation	A service assistant can become an access, screening or rent-pricing system
Geospatial and AECO route	Site planning, parcel context, zoning checks, environmental context and development feasibility	GIS source lineage, parcel and land-use mapping, method version and planner review	Planning context is not a legal entitlement, appraisal or construction approval

The same model family can be safe in one real-estate lane and blocked in another. The signature proof is a control map, not a leaderboard. Real estate mixes regulated finance, consumer conduct, physical assets and local market information. A route that can summarise an OM should not automatically score tenants. A route that can estimate home value for a consumer should not automatically clear mortgage collateral. A route that can describe flood exposure should not automatically set an insurance or capex decision.

That split turns model selection into evidence selection. The practical shortlist is the proof a buyer needs before a model moves from assistive use into decision support.

Approval evidence by route

Assist

Pilot

Approve

Block without evidence

Mortgage AVMCollateral and credit consequence.

LowExplain output to reviewer

PartialBacktest on portfolio

GateAVM controls and sample review

GateNo fairness or adverse-action support

Consumer searchFair housing and steering risk.

PartialNatural-language query help

PartialLead routing with tests

StrongFair-housing evidence and disclosure

GateOpaque screening or steering

CRE underwritingSource, lease and market-data accuracy.

PartialMemo drafting

StrongExtraction plus analyst review

StrongAccepted-output log

GateUnlicensed comps or copied OMs

Climate and property intelligenceLong-horizon physical and capital risk.

PartialPortfolio screening

StrongAsset triage with official-data comparison

StrongMethod, peril and inspection evidence

GateBlack-box score used as fact

Low-friction assistNeeds controlled pilotApproval evidence existsDo not use for decisions

Terms used in this report

Term	Plain meaning
Route	The approved path from work to accepted output: source set, data boundary, model or vendor channel, tools, reviewer, evidence record and reapproval trigger.
API	Application programming interface: a governed software connection used by an app or workflow to call a model or service.
LLM	Large language model: a model trained to read and generate text-like tokens.
Token	A chunk of text used for billing, context windows and rate limits. It is usually smaller than a word, not a page count.
RPM	Requests per minute: how many separate API calls the route can accept each minute.
RPD	Requests per day: the daily cap on separate API calls.
TPM	Tokens per minute: how much token traffic a route can process each minute, unless the provider splits input and output separately.
TPD	Tokens per day: the daily cap on token traffic.
ITPM	Input tokens per minute: how many prompt and source-document tokens may enter the model each minute.
OTPM	Output tokens per minute: how many generated tokens may leave the model each minute.
PTU	Provisioned throughput unit: reserved Azure OpenAI capacity bought for steadier throughput.
PayGo	Pay as you go: shared cloud capacity billed by usage rather than pre-reserved throughput.
RAG	Retrieval-augmented generation: search approved sources first, then pass selected evidence into the model.
OCR	Optical character recognition: converting scanned or image text into machine-readable text.
JSON	JavaScript Object Notation: a structured text format commonly used for machine-readable API output.
429	HTTP 429 means too many requests: the route hit a rate limit or shared-capacity throttle.

Definitions are reader aids, not separate evidence claims.

MCP PREVIEW INVENTORY

What the locked report contains

Schema endpoints

Sample object: /api/v1/reports/RPT_01KSP0REAL3STATEAIMODELS/sample
Inventory object: /api/v1/reports/RPT_01KSP0REAL3STATEAIMODELS/inventory
Schema contract: /api/v1/reports/RPT_01KSP0REAL3STATEAIMODELS/schema

Coverage fields

Category: MODELS
Sector: Real estate
Region: GLOBAL
Topics: real estate AI, automated valuation models, property search, tenant screening, commercial real estate, lease intelligence, property computer vision, climate risk, ESG data, mortgage collateral, fair housing, RESO, MISMO, property data, source rights, algorithmic rent pricing, AECO, geospatial data, property management AI, CRE investment, AI governance

Schema contract

Contract: colophon.report_document v2026-05-28
Payload schemas: document, summary, sample, inventory, full_report
Required top-level fields: title, subtitle, abstract, category, sector, region, topics, sources, evidence, claims, colophon, body
Integrity rules: evidence.sourceId must reference sources[].id claim.evidenceIds[] must reference evidence[].id evidence.claimIds[] must reference claims[].id section and block evidence references must resolve colophon.kinetics.speed must match the momentum band for the business impact x momentum model

Data-point inventory

vendor_or_actor_ledger: entity names; product categories; source references; confidence notes
buyer_decision_fields: pricing signals; deployment constraints; risk flags; monitoring fields
evidence_index: source class; access date; claim linkage; verification status

Section inventory

1. Executive summaryPreview
decision-ready conclusions; major findings; confidence caveats
2. Board capital allocationPreview
structured findings; supporting evidence; buyer-use fields
3. Model economicsPreview
structured findings; supporting evidence; buyer-use fields
4. Decision framePreview
selection criteria; tradeoff map; recommended buyer actions
5. Route approval mapPreview
structured findings; supporting evidence; buyer-use fields
6. Valuation and creditLocked
structured findings; supporting evidence; buyer-use fields
7. Search and consumer routesLocked
structured findings; supporting evidence; buyer-use fields
8. Implementation routesLocked
structured findings; supporting evidence; buyer-use fields
9. Property intelligenceLocked
structured findings; supporting evidence; buyer-use fields
10. Data standards and source rightsLocked
normalized entity rows; source ledger; evidence-to-claim mapping
11. Controls and failure modesLocked
structured findings; supporting evidence; buyer-use fields
12. Private eval designLocked
structured findings; supporting evidence; buyer-use fields
13. Throughput and rate limitsLocked
structured findings; supporting evidence; buyer-use fields
14. Known unknownsLocked
structured findings; supporting evidence; buyer-use fields
15. Recommended next stepsLocked
structured findings; supporting evidence; buyer-use fields
16. Vendor and source directoryLocked
normalized entity rows; source ledger; evidence-to-claim mapping
17. Source ledgerLocked
normalized entity rows; source ledger; evidence-to-claim mapping
18. GlossaryPreview
structured findings; supporting evidence; buyer-use fields

Relevance signals

Confidence score: 0.86
Source classes: regulator, regulation, mortgage_agency, property_data, company_research, product_docs, industry_standard, vendor_docs, industry_report, industry_survey, climate_risk_model, government_data, standard, privacy_terms, cloud_docs, industry_policy, vendor_press, security_framework, official_model_docs, model_benchmark, market_analysis, pricing_page, agent_benchmark
Known unknowns: Comparable public evidence remains thin for accepted-output economics, protected-class proxy tests, AVM sample-validation results, lease-extraction error rates, climate-score validation, image-condition provenance, tenant-screening appeals and CRE workflow audit exports.
Watch signals: FHFA automated-valuation-model quality-control implementation update; valuation routes could require stronger testing and nondiscrimination evidence [[E41, E61]]. HUD tenant-screening guidance page revision; search, screening and lead-routing routes could require new fair-housing evidence [[E25]]. Fannie Mae value acceptance or Freddie Mac ACE policy release; mortgage collateral routes may change eligibility or review thresholds [[E6, E7]]. RESO Data Dictionary release note changing listing-field coverage; listing routes could lose approved-field coverage [[E34]]. First Street Risk Factor methodology update; property-risk routes could require stricter provenance evidence [[E37]].

Locked continuation

Purchase or subscribe to unlock the full findings, evidence ledger, and buyer-ready recommendations.