Decentralised News · 2027 planning series · Sources reviewed 22 November 2026
What Matters
A pricing system can process useful market information while leaving its customer unable to explain where a recommendation came from. The 2027 purchasing question is therefore about data, settings and independent decisions. DN’s evidence register helps buyers document those facts and identify unresolved questions. It does not determine whether a company broke competition law or calculate an allegedly excessive rent.
The consequential part of the story is in the software record
The RealPage case became a shorthand for a striking question: what happens when competing landlords use a common pricing supplier? The public record is more specific than the slogan. In its November 2025 settlement announcement, the US Department of Justice described allegations involving nonpublic landlord information and software features that could align prices or limit decreases.
The procedural status has since changed. The 19 May 2026 final judgment entered a consent resolution with RealPage in the Middle District of North Carolina. It states that there was no trial or adjudication of fact or law and no admission by a party. Its provisions address specified data uses, product features, monitoring and compliance. That is a binding resolution with a defined scope, not a verdict that every algorithmic rent is unlawful.
For a software buyer, the practical lesson is not to decide a court case from a dashboard. It is to request evidence that shows which information enters the system, what the system does with it and how the business retains responsibility for its own pricing decisions.
Evidence Block
Reviewed 22 November 2026 for 2027 planning. This analysis distinguishes US allegations, the RealPage consent judgment, the company’s response and UK CMA guidance. It does not combine them into one global legal rule. RealPage’s statement confirms the final approval date and says the agreement includes no findings or admissions of wrongdoing. DN has not inspected customer configurations, proprietary training data or a representative sample of rents. The procurement framework below is proposed research, not a legal opinion or a vendor compliance audit.
1. Define the pricing decision before reviewing the algorithm
A property business can use software for several different purposes: collecting published asking rents, forecasting its own demand, recommending renewal offers, setting concession limits or automatically sending prices into a leasing system. A buyer should state which of these jobs is being purchased. Calling all of them “AI revenue management” hides the action that needs review.
Define the output and its recipient. Is the system preparing an internal suggestion, generating a quote for a prospect or changing a price that will be presented automatically? Record the properties and business entities covered, the responsible decision owner and the production systems connected to the result.
Next identify the commercial objective. A model may optimise rent per occupied unit, expected total revenue, occupancy, renewal probability or another target. Those objectives can produce different recommendations even with identical data. The buyer needs a versioned statement of what the system is trying to achieve and which constraints are approved.
This is ordinary product diligence with a material consequence: it makes the pricing decision inspectable before the organisation asks whether its data and governance are appropriate.
2. Build a map of the data entering the decision
DN proposes classifying each declared input by origin, owner, date, granularity, permission and use. “Market data” is too broad a description. A public advertisement, the buyer’s own signed lease and another landlord’s confidential renewal plan are different records, even if a vendor converts them into the same numeric feature.
| Input class | Evidence to request | Question to resolve |
|---|---|---|
| Own-property records | Source system, ownership, date and permitted use | Does the dataset cover the intended properties and period? |
| Public observations | Collection method, original location and observation time | Was the information actually public when collected? |
| Vendor-derived features | Documented upstream sources and transformations | Which facts could influence the derived value? |
| Other businesses’ information | Source, confidentiality status, scope and specialist review | Could competitively sensitive rival information enter the recommendation? |
| Human adviser input | Author, evidence basis and retained communication | Does a manual adjustment introduce information absent from the software inventory? |
The table is a DN documentation template. It is not a permission list. A public source label does not establish that every subsequent use is lawful, and a signed data licence does not answer every competition question. The point is to replace a vague assurance with facts that the relevant specialists can evaluate.
3. Inspect training and live recommendation inputs separately
A vendor may say that current competitor data is excluded from its live feed while its model was trained on a different dataset. Conversely, a model with an appropriately documented training history can still consume an unsuitable operational feed. A complete inventory therefore needs both stages.
For training, request the categories of data, relevant time windows, ownership boundaries, update schedule and change controls. For live operation, request the feeds queried, refresh intervals, transformations and any manual adjustments. A statement about one stage should not be treated as a statement about the other.
The RealPage judgment’s detailed definitions and exceptions illustrate why this distinction matters. Its rules for the defendant include separate treatment of runtime use and model training. Readers should not extract one age threshold or exception from that order and use it as a general safe harbour for unrelated systems.
DN would also ask what happens when a source is removed. Does the change only stop future retrieval, or does it require a model or feature update? The answer can differ by architecture. Record the vendor’s explanation and the evidence supporting it rather than inferring deletion from an interface toggle.
4. A recommendation can still affect the decision
The existence of an override button is useful operational information. It is not a complete legal analysis. In a March 2024 statement of interest concerning hotel pricing, the FTC and DOJ argued that allegations of unlawful price fixing need not depend on direct competitor communications or fully binding recommendations. That filing states the agencies’ position in a separate US case; it is not a finding about every shared pricing product.
For procurement, the resulting evidence question is straightforward: how is discretion exercised in practice? Request the workflow for accepting, changing or rejecting a suggestion. Identify defaults, automatic actions, escalation requirements and incentives attached to acceptance. A policy that permits independent judgement should be examined alongside the interface and management process that staff actually use.
Keep the interpretation narrow. A high acceptance rate does not prove collusion; the recommendations might match the business’s independently established objectives. A high override rate does not prove independence either; staff might be overriding prices using the same unexamined external input. The retained decision record is more informative than either percentage on its own.
5. Record downward adjustments as carefully as upward ones
DN recommends reviewing the full range of permitted actions. Can the user lower a suggested price, change a concession, reject a recommendation or pause automatic publication? Does the evidence show which person configured those options and when? The objective is to understand the control surface, not to prescribe a rent.
In an isolated test environment, use invented properties and synthetic demand changes to inspect the workflow. Ask the vendor to demonstrate a recommendation, a rejection and a material configuration change. Keep the expected behaviour written before the demonstration. Do not use a joint exercise with competing landlords to discuss future prices or coordinate commercial plans.
Reviewers should also ask whether a change affects one property, a portfolio or all customers. A central software update can alter behaviour at a different scope from an individual manager’s setting. The organisation needs to know which changes it approves and which changes arrive automatically through the supplier.
6. Include advisers, meetings and integrations in the map
A data-flow review limited to API feeds can miss the person who adjusts a recommendation. Record whether advisers can view multiple customers, what information supports their advice and where customer-specific discussions are retained. A spreadsheet imported by an analyst is still an input even when it bypasses the main product interface.
The same reasoning applies to conferences, user groups and support sessions. Technical troubleshooting and discussion of sensitive commercial plans are different activities. DN would document the purpose, attendance and permitted subject matter of a shared session, with an established route for raising a concern.
The UK CMA’s March 2026 discussion of AI and collusion warns about direct or indirect exchange of competitively sensitive information and recommends understanding the data and algorithms used. It also recognises that algorithmic pricing can bring benefits and that risk depends on context. Those UK observations support careful diligence; they do not replace advice on a specific arrangement in another jurisdiction.
DN Pricing-Software Evidence Register
The local register asks for seven documented elements: the pricing objective, input-source inventory, training and live-use boundaries, production configuration, independent decision record, change and export process, and a recorded specialist review appropriate to the jurisdiction and facts.
Mark each element documented, missing or unknown. The output lists the outstanding records. A separate flag records whether potentially sensitive rival information is known to be involved, excluded on documented evidence or unresolved. A flagged or unresolved input requires specialist review; the tool does not decide that it is unlawful.
An optional calculation reports source-documentation coverage. If the declared inventory contains eight distinct source categories and six have a documented origin and use, coverage is 6 ÷ 8 = 75%. That percentage describes the inventory entered by the reader. It is not a compliance score, a fraud probability or proof that the vendor disclosed every source.
Define the counting unit before comparing periods. Splitting one well-documented feed into ten entries must not make the register appear more complete. Keep confidential pricing data, tenant details and private legal advice out of the public worksheet.
DN Pricing-Software Evidence Register
List documented, missing and unknown evidence about one pricing-software deployment. The optional coverage calculation describes your declared inventory only. Do not enter confidential rival pricing, tenant information or private legal advice.
This instrument runs locally in your browser. It does not connect to accounts, fetch evidence or submit instructions. Use aggregate values and non-sensitive references. Downloaded records contain your inputs.
Machine-readable illustrative record
An illustrative record of the example assumptions, not an observed transaction or provider test. A null calculation timestamp identifies this static example.
{
"schema_version": "1.0.0",
"article_id": "N004",
"evidence_basis": "illustrative",
"source_review_date": "2026-11-22",
"computed_at": null,
"inputs": {
"objective": "unknown",
"inventory": "unknown",
"boundaries": "unknown",
"configuration": "unknown",
"decisions": "unknown",
"changes": "unknown",
"specialist": "unknown",
"rival_information": "unknown",
"declared_sources": "8",
"documented_sources": "6"
},
"status": "Specialist review required: data question unresolved",
"outputs": {
"methodology": "DN pricing-software evidence register v1.0",
"self_reported": true,
"state": "specialist_review",
"records": {
"objective": "unknown",
"inventory": "unknown",
"boundaries": "unknown",
"configuration": "unknown",
"decisions": "unknown",
"changes": "unknown",
"specialist": "unknown"
},
"missing_records": [],
"unknown_records": [
"Pricing objective",
"Source inventory",
"Training/runtime boundaries",
"Production configuration",
"Independent decision records",
"Change and export process",
"Specialist review"
],
"rival_information": "unknown",
"declared_sources": 8,
"documented_sources": 6,
"coverage_percent": 75,
"compliance_certification": false
}
}Worked comparison: two buyers, one vendor name
Hypothetical example. Buyer A has a written objective, an inventory of eight declared input categories and evidence for six. Its team cannot establish how a derived market indicator is constructed, and it has not retained the configuration governing automatic publication. The software contract is signed, but the record remains incomplete.
Buyer B uses a separately configured product from the same supplier. It has documented its declared inputs, reviewed the relevant data boundaries and retained sample recommendation-to-decision records. It also has a process to reassess new feeds and model versions. Its documentation is more complete for those defined checks.
The comparison does not prove that Buyer B’s use is lawful or that Buyer A has harmed anyone. It demonstrates why a vendor-level badge is insufficient for a deployment-level decision. Different settings, integrations, data and practices can create different questions under the same logo.
The appropriate next step for Buyer A is to resolve the specific missing records before expanding the system’s authority. The appropriate next step for Buyer B is to preserve the reviewed configuration and monitor changes. Neither buyer gains a permanent exemption from further review.
Why a chart of rising rents cannot settle the case
A price trend is an outcome to explain. It is not, by itself, a record of the process that caused it. Rent can vary with property quality, location, lease terms, supply, demand and changes in the sample being measured. A chart that combines different buildings or different periods can obscure those distinctions.
A serious empirical study would need a defined question, a credible comparison and careful treatment of selection into software use. Properties adopting a product may differ from properties that do not. Asking rents and achieved rents are also different measurements; concessions can change the economic offer without changing the headline.
DN’s register is therefore deliberately limited to evidence about the software deployment. It does not estimate damages, identify a counterfactual rent or declare a market competitive. Those tasks need separate data and analysis. A reader can question a product’s documentation without claiming to have established a causal effect on everyone’s housing bill.
The 2027 extension: pricing agents with more ways to obtain information
An agent can potentially read documents, search websites, query APIs and exchange messages before proposing a price. That creates a larger source boundary to document. The organisation should identify which channels are permitted and what happens when the agent receives information outside them.
DN proposes a deployment record that includes model version, instructions, available tools, approved sources, output destination and the person responsible for review. Keep retrieved material separate from instructions that grant authority. A third-party page claiming to reveal competitors’ future plans should not silently become an approved pricing input.
Test source restrictions and logging with synthetic fixtures. For example, a fabricated document can contain an explicitly disallowed competitor-confidential label. The test asks whether the system isolates and escalates it under the organisation’s policy. This is a test of the implemented boundary, not an exercise in obtaining real competitors’ confidential information.
Agent-to-agent communication deserves its own record. Identify the counterparty, purpose and authorised fields. More messages do not automatically mean better market information, and a machine-generated recommendation remains part of a business decision with accountable owners.
What a useful vendor response looks like
A strong response names the product and version, describes the relevant configuration and distinguishes facts the supplier can demonstrate from matters requiring customer judgement. It provides evidence for the data categories used, the deployment boundary and the change process. It should be possible to connect the response to the system the buyer is actually purchasing.
A weak response repeats that the product is “AI powered”, cites a generic privacy policy or points only to the ability to override recommendations. Those statements can be relevant, but they do not answer the complete set of questions. The buyer should return a specific evidence request rather than convert the unanswered question into a positive assumption.
Confidentiality can be handled through an appropriate professional review arrangement. A vendor may reasonably decline to publish proprietary code. The purchasing decision still needs enough reliable evidence to evaluate the intended use. Record any resulting limit on what the business can verify.
DN Alpha Thesis
As pricing software becomes more capable, evidence about inputs and decisions may become part of the product’s commercial value. Buyers could increasingly favour suppliers that make their data boundaries understandable, retain useful records and support independent governance without preventing legitimate analysis.
This is a testable procurement thesis. It would weaken if buyers consistently obtain the same assurance from cheaper generic tools, if the evidence imposes disproportionate operating costs or if adoption decisions show no preference for verifiable controls. DN is not predicting a vendor winner or treating compliance documentation as proof of economic performance.
Methodology and limits
The source review uses the relevant US government record, the consent judgment, the defendant’s stated position and a separately identified UK regulator discussion. Allegations are labelled as allegations, agency arguments as agency arguments and the final judgment as a consent resolution. The article does not determine obligations for a particular business, infer that every landlord uses the same system or describe all rental litigation as resolved.
The proposed register captures documentary completeness, with no invented numerical risk weights. The optional coverage calculation uses a reader-defined inventory and cannot reveal undisclosed sources. It should support a concrete conversation with operational and legal reviewers, not replace that conversation.
Recheck the record when a vendor changes its model, input categories, adviser process, automatic-publication settings or relevant contractual terms. Preserve the prior review so that a new product description does not erase what the business previously approved. Review jurisdiction-specific developments before relying on a historical order or guidance document.
For the separate operational question of handling rent receipts, repairs and contractor payments, read DN’s AI Property Management guide. This article concerns pricing inputs and competitive independence. It includes no affiliate recommendation or paid product ranking.
Frequently asked questions
Did the RealPage case remain only a proposed settlement in November 2026?
No. The inspected final judgment was filed on 19 May 2026. This article identifies that consent resolution while distinguishing it from other parties and separate rental litigation.
Did the consent judgment establish every allegation as fact?
No. The judgment states that there was no trial or adjudication of fact or law and no admission by a party. Its operative requirements still have a defined, binding scope.
Is every use of pricing software illegal?
No. Software use alone does not establish a violation. Applicable law, market context, information flows, arrangements and actual behaviour need appropriate review.
Does an override button settle the competition question?
No. It is one operational fact. The sources, agreements, configuration and way discretion is actually exercised also matter. The article explains this distinction without judging a particular deployment.
What does the documentation-coverage percentage prove?
Only the share of the reader’s declared source inventory with documented origin and use. It is not a compliance score, does not detect undisclosed sources and does not estimate overcharges.
Can I copy a threshold from the RealPage judgment into another product policy?
Do not treat a defendant-specific definition or exception as a general safe harbour. The complete order, the proposed use and the relevant jurisdiction must be reviewed together.
Primary sources and review record
- US v RealPage: Final Judgment, filed 19 May 2026. Consent status, absence of trial/admission, defined runtime/training, features, monitoring and compliance provisions; not general safe harbour. Primary court document hosted on court-authorised settlement site. Reviewed 2026-11-22.
- DOJ: RealPage proposed settlement announcement, 24 November 2025. Allegations concerning nonpublic information and price-alignment features; announcement is historical proposal, superseded procedurally by final order. Reviewed 2026-11-22.
- RealPage: Statement on DOJ settlement. Company position, no admission, acknowledges final approval 19 May 2026. Reviewed 2026-11-22.
- FTC: Cornish-Adebiyi v Caesars statement of interest. Agency position in a separate US case: no necessity for direct communication or binding recommendation in alleged illegal agreement. Not presented as judgment on every algorithm. Reviewed 2026-11-22.
- CMA: AI and collusion, 4 March 2026. UK guidance on sensitive competitor data, indirect exchange, auditing data/algorithms and benefits/context; jurisdiction expressly scoped. Reviewed 2026-11-22.




