AI data market
Real estate operations data investment: 7 segments, 2 legal walls
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Real estate operations data investment works only segment by segment. Maintenance work orders, inspections and building-system records carry checkable outcomes and removable personal data; tenant screening files are consumer reports under federal law, and pooled rent data drew a Justice Department suit and a 2025 proposed settlement limiting model training. The open RealEstateCore ontology reflects the split: 7 of its 1,365 classes describe people or leases.
Value drivers
Real estate operations data
Key takeaways
- Judge a property-records supplier on segment mix, not units managed: most records that grow with each unit are leases, screening files and rent histories.
- RealEstateCore's 1,365 open classes are 88.2% equipment and sensing, with only seven for people, organizations or leases and none for rent amounts or screening.
- Maintenance and inspection records have an outside grading key in HUD's NSPIRE severity scale, but no public test of their value for model training.
- Screening files are FCRA consumer reports with documented errors, and pooled rent data drew a 2024 antitrust suit whose proposed settlement limits training data.
- Ask for maintenance chains with verified outcomes, the denominator behind acceptance figures and who holds licensing authority for each property.
Which real estate operations records could an AI buyer actually use?
The usable records are the ones tied to buildings and repairs, not the ones tied to renters' applications and money. A property manager's systems hold both: work orders, inspections and equipment logs sit beside screening reports, rent ledgers and pricing histories, often in the same platform. Asking whether real estate operations data is an AI opportunity really means asking which of those segments can be licensed, prepared and graded, and the answer changes sharply from one segment to the next.
O*NET's task list for property, real estate and community association managers (occupation 11-9141.00) shows the spread inside a single job: planning repairs and inspecting grounds and equipment, collecting rent and deposits, preparing budgets, showing units and determining tenant eligibility under government rules, investigating complaints, and keeping to fair-housing rules. Each task leaves records, and each family of records carries its own mix of buyer use, legal exposure and preparation work.
The thesis tested here is narrow. Defensible supply in this sector sits in the physical-operations layer, while the decision layer that looks most valuable, screening and pricing, is walled off by consumer-reporting and competition law. If that holds, a supplier should be judged on its segment mix, not on how many units it manages.
What 1,365 RealEstateCore classes say about where the open standard stops
The industry's own open ontology is built almost entirely around buildings and equipment. RealEstateCore, an ontology published on GitHub by a consortium of property owners, defines 1,365 classes in its DTDL source at the commit we read (c3984ed, October 2026, with version 4.1 release notes), and 1,094 of them sit in a Brick folder of equipment, point and system classes; the README credits its equipment and point hierarchies to a collaboration with Brick Schema. Only seven describe people, organizations or leases.
That detail matters for preparation cost. REC's work-order class inherits a service-object skeleton with created, acknowledged and closed times, a severity of Major, Minor or Severe and a status, plus links to the agents who created, acknowledged and closed it. Its Lease class links a leasee, a leasor, the leased object and a governing contract, with start and end dates. Across all 1,365 classes the only property name that touches rent vocabulary is rentableArea, a floor-area measure; nothing models rent amounts, payments, deposits, screening results or arrears.
An open schema proves neither supply nor demand. It does lower the cost of normalizing records from different owners into one dataset, and the work-order links show exactly where personal data enters an otherwise physical record: through the people fields.
Counting rules, as the script applies them: one class is one JSON file under Source/DTDLv2; segments are assigned by folder in the order the table lists them, with Lease and LeaseContract matched by name; property and relationship names are split into camelCase words and flagged when a word starts with rent, price, pay, deposit, invoice, income, credit, screen, evict, delinquen or arrear.
RealEstateCore classes by record segment:
| Segment | Classes | Share of 1,365 | What the classes cover |
|---|---|---|---|
| Equipment and sensing | 1,204 | 88.2% | Brick-folder equipment, points and systems, REC assets and observation events |
| Spaces and structure | 133 | 9.7% | Sites, buildings, rooms, building elements, collections such as apartments |
| Other information and events | 16 | 1.2% | Documents, geometry, addresses, warranties, elevator trips |
| People and organizations | 5 | 0.4% | Agent, person, organization, company, department |
| Work orders and alarms | 5 | 0.4% | Service object, work order, error report, alarm, notification |
| Leases and contracts | 2 | 0.1% | Lease event and lease contract document |
Repairs and inspections: the segment with a public severity scale
Maintenance and inspection records sit closest to a gradable AI task, because their outcomes can be checked against something outside the record. HUD's NSPIRE inspection standard assigns each deficiency a severity level, and its checklist sets a 24-hour correction timeframe for life-threatening deficiencies, with longer windows for lower levels that differ by program. A triage model's call on a reported leak or a missing smoke detector can be graded against that scale and against the work order's own timestamps.
The records still carry personal data. A work order names who reported the problem, often records entry permission and whether anyone was home, and may hold photos of a home's interior; REC's own skeleton links every work order to the people who opened, acknowledged and closed it. Those fields come out or get generalized before licensing, and requests involving disability accommodations or safety complaints need their own review. What remains, report, triage, dispatch, repair and recheck, is mostly about the building.
For an investor the open question is whether a supplier's maintenance history keeps the outcome link: a closed work order matched to an invoice and no repeat request, across enough properties and years to grade a model. That is measurable in a sample, and no public source has measured it.
Tenant screening files are consumer reports, which changes the deal
Screening records fall under the Fair Credit Reporting Act, so they are a regulated decision trail, not ordinary operating data. The FTC's guidance for landlords says an adverse action notice is required when a landlord denies an application, or demands a co-signer or a larger deposit, based even partly on a consumer report; the notice must name the screening company, say that company did not make the decision, and explain the applicant's right to dispute errors and obtain a free copy of the report.
The labels are weak as well. The CFPB's November 2022 report on the tenant background check market found documented errors in the public court records that screening reports include, said no definitive analysis exists of how often matching errors occur, and found that prior rental payment history is overwhelmingly not reflected in screening reports or risk scores. A model trained to reproduce past screening decisions would learn from outcomes that are legally sensitive and, by the regulator's account, often built on inaccurate inputs.
None of this makes the segment worthless to every buyer, but it moves screening out of a data-licensing thesis and into a compliance-product thesis, with different buyers, approvals and risks. When screening files are counted as supply, ask how they could be licensed at all.
Pooled rent data has an antitrust record, not only a privacy one
Rent-setting data is the segment where the main risk is competition law. In August 2024 the Justice Department sued RealPage under Sections 1 and 2 of the Sherman Act, alleging that landlords share nonpublic, competitively sensitive information about rental rates and lease terms with the company, which uses that data to train and run its pricing software.
The proposed settlement filed on November 24, 2025 reaches directly into model training. The department's competitive impact statement says RealPage may train only on backward-looking data at least 12 months old and not tied to active leases; the press release adds that its software may not use competitors' nonpublic data in runtime operation and its models may not determine geographic effects narrower than the state level. We did not confirm whether the court has entered the judgment.
States are acting too. California's AB 325, chaptered on October 6, 2025, makes it unlawful to use or distribute a common pricing algorithm as part of a contract, combination or conspiracy in restraint of trade, and defines that algorithm broadly: a method, including software, used by two or more persons that uses competitor data to recommend, align, stabilize, set or otherwise influence a price or commercial term. Pooled lease and pricing records need competition counsel before they need a data buyer.
A seven-segment scorecard, with the evidence each one lacks
Rated on stated criteria, three segments look licensable with carve-outs, one needs contract work and three should stay out of a data-licensing thesis. The ratings rest only on the sources cited here; none uses SourceX data, and measured model utility is missing in every row.
Rating rules: buyer task is strong when O*NET documents the task and a public standard can grade it, partial when only the task is documented; exposure is high when a cited federal or state regime governs the records or they are individuals' financial records, medium when personal or confidential fields can be removed, low otherwise; the preparation signal comes from the RealEstateCore count.
Segment scorecard, editorial ratings as of October 27, 2026:
| Segment | Buyer task and grading key | Legal and personal-data exposure | Preparation signal | Verdict and missing evidence |
|---|---|---|---|---|
| Maintenance work orders | Strong: repair planning task; severity scale and timestamps | Medium: reporter, entry and interior photo fields | Open work-order skeleton in REC | License with carve-outs; no public test of model utility |
| Building systems and equipment | Partial: equipment inspection task, no public key for decisions | Low, except occupancy and access-control points | 1,204 REC equipment and sensing classes | License with carve-outs; demand for operations-only data unproven |
| Inspections and compliance | Strong: NSPIRE severity levels and correction clocks | Low to medium: unit addresses and household details | Public inspection standard | License with carve-outs; share linked to repairs unknown |
| Commercial lease administration | Partial: budgets and lease terms | Medium: confidentiality clauses, counterparty names | Two REC lease classes, no rent fields | Contract review first; counterparty consent terms unknown |
| Leasing and tenant screening | Partial: eligibility task | High: FCRA consumer reports with documented errors | No open schema | Hold out; lawful basis and label accuracy unproven |
| Rent setting and revenue management | Partial: no public grading key | High: Justice Department action and California AB 325 | No open schema | Hold out; competition review needed in each market |
| Rent payments and collections | Partial: rent collection task | High: individuals' financial records | No open schema | Hold out; purpose limits and de-identification untested |
Why units under management overstate licensable supply
A large portfolio is not a large dataset, because most of what grows with each unit sits in the segments this scorecard holds out. Every added unit brings a lease, a screening file, a ledger and a rent history; only some bring work orders and inspections, and fewer still keep the link from report to verified repair.
Ownership and system control split the supply again. A manager may operate records for many owners under management agreements, and the platform vendor's terms shape what can be exported; the guide to property management software vendors on this site sets out that split. SourceX's general guide is that buyers look for years of consistent history, typically two or more, and tens of thousands of records in one system, with smaller sets sometimes qualifying when the work is rare; in this sector that test belongs to each segment, not to the portfolio.
The plainest explanation for a sector that looks data-rich is that rent and screening records are generated for every unit, every month. Remove them, count only maintenance chains with a verified outcome, and the supply estimate falls by an amount nobody has published. That count is the first figure to request.
Risks: rulings, rules and substitutes that could undo the case
The segmentation case could fail in at least five ways, and each deserves a hearing before any upside does. Where a dated source exists, it is named so the risk can be tracked.
- Regulation could widen. The RealPage judgment is described as proposed in the filings read here, and AB 325 defines pricing algorithms broadly enough that pooled maintenance cost data used to set rents could draw the same questions.
- Personal data could reach further into maintenance than expected. Entry notes, accommodation requests and safety complaints can turn a repair record into a sensitive one, and removing them may strip the context that makes the record useful.
- Free substitutes could cover the easy part. NSPIRE's severity scale and open ontologies such as RealEstateCore cost nothing, so a buyer may build synthetic maintenance tasks rather than license real histories.
- Demand could be thin or concentrated. No source cited here shows AI developers paying for property maintenance histories; demand for this segment is a hypothesis until a buyer tests it.
- Control could sit elsewhere. Owners, managers and software vendors can each hold part of the rights, and the party holding the records may not be the party able to license them.
Eight questions for a property-records supplier
Ask for evidence segment by segment, and put the same questions to SourceX. SourceX arranges licenses for chosen company records used in AI work and runs the steps from qualification and rights checks through contracting, preparation, delivery and payment, and the supplying company approves scope and terms and signs before anything is delivered. Its company and commercial metrics are not public; it shares them with investors directly. SourceX is raising from accredited investors under Rule 506(c), and investor materials are available on request through the investors page.
Questions, each answerable with a sample or a document:
- What share of the offered records falls in each of the seven segments, counted in records and in years of history?
- For maintenance, how many work orders link report, dispatch, invoice and a verified close, and how many are followed by a repeat request inside a fixed window?
- Which fields that identify residents, entry permissions or accommodations are removed, and what remains for grading?
- Who holds licensing authority for each property, owner, manager or software vendor, and where is that written down?
- Are any screening, rent-pricing or payment records in scope, and if so, on what legal basis and with whose sign-off?
- Could any offered field reveal a competitor's nonpublic rents or lease terms, alone or once pooled with other records?
- Has any buyer run a held-out test on these records, and what exactly did it measure?
- When the supplier reports acceptance or rejection of property records, what is the denominator?
Ratings method, sources and what stays unmeasured
The class count was run on 2026-10-27 against RealEstateCore at commit c3984ed with a script kept for reviewers; its counting rules appear in the ontology section. The scorecard ratings were assigned the same day from the cited sources: O*NET 11-9141.00 for tasks, HUD's NSPIRE checklist, the FTC's landlord guidance, the CFPB's 2022 tenant screening report, Justice Department filings in the RealPage case and California's AB 325. The rating rules sit above the scorecard table.
Not measured: model utility for any segment, the size of any supplier's archive, the share of maintenance records with verified outcomes, buyer demand, prices and the current status of the RealPage judgment. No SourceX data was used. The ratings are an editorial diligence framework, not a valuation, and the legal points are general information for counsel to check.
Frequently asked questions
Can tenant screening data be licensed for AI training?
It is the hardest segment to license. Screening reports are consumer reports under the Fair Credit Reporting Act, landlords owe applicants adverse action notices when a report affects a decision, and the CFPB documented errors in the court records these reports include. Treat it as outside a data-licensing thesis unless counsel finds a lawful basis.
Why does the RealPage case matter for real estate AI data?
It shows that pooled, nonpublic rent and lease data used to train pricing software can draw antitrust action. The Justice Department's proposed settlement from November 2025 limits RealPage's model training to data at least 12 months old and not tied to active leases. Anyone pooling pricing records for AI needs competition counsel first.
Which real estate operations records have the clearest licensing path?
Maintenance work orders, inspections and building-system records, once resident identities, entry notes and interior photos are removed or generalized. They map to documented manager tasks, and HUD's NSPIRE severity scale gives some of them an outside grading key. Their value for model training is still unmeasured in public sources.
Does a bigger property portfolio mean more licensable data?
Not in proportion. Most records that grow with each unit are leases, screening files, ledgers and rent histories, which carry the highest legal exposure. Count maintenance chains with verified outcomes instead, segment by segment.
Does SourceX publish real estate deal or supplier figures?
No. SourceX's company and commercial metrics are not public, and it provides them to investors directly. Investor materials can be requested at sourcex.si/investors, and any offering is limited to verified accredited investors through SourceX's offering documents.
Sources
- RealEstateCore's DTDL source at commit c3984ed defines 1,365 classes, 1,094 of them in its Brick folder, with five agent classes, two lease classes and a work-order skeleton of times, severity, status and people links. Source
- O*NET's profile for property, real estate and community association managers lists tasks spanning repairs and inspections, rent collection, budgets, tenant eligibility, complaints and fair-housing compliance. Source
- HUD's NSPIRE checklist sets a 24-hour correction timeframe for life-threatening deficiencies. Source
- The FTC tells landlords that adverse action based even partly on a consumer report requires a notice naming the screening company and explaining dispute rights. Source
- The CFPB's November 2022 tenant background check market report documents court-record errors in screening reports and finds rental payment history largely missing from them. Source
- The Justice Department's August 2024 suit alleges RealPage uses landlords' nonpublic rental rate and lease term data to train and run its pricing software. Source
- The Justice Department's settlement announcement bars RealPage from using competitors' nonpublic data in runtime operation and from modeling geographic effects narrower than a state. Source
- The competitive impact statement limits RealPage's model training to backward-looking data at least 12 months old that is not tied to active leases. Source
- California's AB 325, chaptered October 6, 2025, makes certain uses of a common pricing algorithm unlawful and defines it by its use of competitor data. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.