What problem does it solve?
Landlords and property managers have always screened applicants on credit and criminal history, but traditional screening asks whether an applicant is able to pay, not whether they are willing to prioritize rent over other bills. RealPage, whose AI Screening product is built on a database of more than 36 million lease outcomes, argues that past rental payment behaviour predicts future payment better than a credit score alone, and that conventional screening turns away reliable renters who happen to have a thin or damaged credit file.
The stakes of getting the score wrong fall unevenly. A tenant screening algorithm decides who gets a home, and the same automation that promises lower vacancy losses and faster decisions can encode bias against a group already disadvantaged in credit history, which the Fair Housing Act's ban on race based discrimination does not allow, whoever or whatever makes the decision.
In Louis v. SafeRent Solutions, a federal court denied the defendants' motion to dismiss in July 2023, holding that the plaintiffs had adequately alleged that a screening score had a disparate impact on Black and Hispanic applicants who used housing vouchers; the case then settled for $2.275 million with injunctive relief in November 2024.
Voucher, or source of income, status is not itself a protected class under the federal Fair Housing Act; where it is protected, that protection comes from state or local law.
- When a housing voucher is used, on average over 73% of the monthly rental payment is paid by public housing authorities directly to housing providers, a fact the SafeRent Score's algorithm did not factor into the applicant's score, according to the plaintiffs' complaint in Louis v. SafeRent Solutions.Louis, et al. v. SafeRent Solutions, et al. (2022)
How does it work?
- Collect the application data. Credit history, rental payment history, eviction and criminal records, income and, for voucher holders, the voucher amount, come from the applicant and third party data providers.
- Score the application. A model trained on historical lease outcomes (paid on time, skipped, evicted) produces a score or a recommendation, weighing payment behaviour, income stability and credit history against the outcomes it has learned to predict.
- Recommend a decision. The score feeds a threshold the property manager sets (approve, approve with conditions such as a higher deposit or a guarantor, or decline), balancing occupancy targets against payment risk.
- Give the required notice. Where the score contributes to a decline or a less favourable term, the applicant gets an adverse action notice under the Fair Credit Reporting Act, naming the screening company and stating the right to a free copy of the report and to dispute it.
- Monitor for disparate impact. The score and its outcomes are tested periodically against protected characteristics, including voucher status where that is a protected class locally, and the model or its inputs are adjusted when a pattern of disparate impact appears.
- Audience
- Back office
- Autonomy
- Supervised agent
- Adoption
- Mainstream
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Risk and loss reduction, Lower cost to serve, Compliance quality.
Indicative value
A multifamily operator with 40,000 apartment homes
USD 1.2 million to USD 1.6 million
Net operating income gained from better screening per year
How this is calculated
Formula: units * savingsPerUnit. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Apartment homes managed units, homes | 40,000 | 40,000 | The reference operator, matching the scale RealPage's own case example uses. |
| Net operating income gained per unit per year from reduced non payment and default risk savingsPerUnit, USD per unit per year | 31 | 39 | RealPage reports AI Screening delivered an average savings of $39 per unit per year in the first full year of results in the field, exceeding an earlier, pre launch forecast of $31 per unit by 25%. Vendor reported, one platform, not independently audited. |
What it leaves out: RealPage's own reported average from the first full year of AI Screening in the field, not an independent study; the blog points to a white paper for methodology that is not cited here. It leaves out the cost of the screening service, of any compliance and testing programme, and of the harm and legal exposure a biased score can create.
Market estimates (analyst estimates, not deployments)
- One national REIT reported saving $1.1 million across its 52,000 units after adopting RealPage AI Screening, with early lease terminations down 24% year over year and skips and evictions down 38%; across 50,000 same store leases representing 487 unique assets and 130,000 units, RealPage reports AI Screening drove a 10.2% reduction in the average balance owed per move in. From "Ability to Pay" to "Willingness to Pay": AI-Based Tenant Screening is Boosting NOI (2021)
Who already uses it?
6 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Kane Realty Corporation
United States · Real estate · 2025
Kane Realty Corporation, a Raleigh, North Carolina owner operator with about 6,000 units, uses Funnel's AI enabled screening tool, FunnelSecure, to flag high risk rental applications for additional review before they can become costly delinquent leases. In a Funnel case study, Kane's Director of Marketing and Resident Experience, Sarah Allwardt, reports that from March to June 2025 total applications needing screening rose 25% year over year, applications that sailed through screening smoothly rose 49%, applications needing additional verification fell 30%, and the screening decision itself came back 47% faster than a year earlier.
- Cycle time reduction: 47%, one year, reported with March to June 2025 results
"Kane streamlined one of the most important steps in leasing: the move from application submitted to decision made. In just one year, they cut that stage nearly in half, delivering decisions 47% faster."
Claimed by: vendor
CF Real Estate
United States · Real estate · 2021
CF Real Estate implemented RealPage AI Screening, which scores rental applicants on their predicted willingness to pay rent using a database of more than 36 million lease outcomes rather than credit history alone. RealPage's own blog reports that CF Real Estate reported a 75% drop in skips and evictions after adopting the tool, as part of a wider set of customer results RealPage says followed the first full year of AI Screening being in the field. Skips and evictions are a broader measure than the standard 30 day delinquency rate, so this figure is not recorded as a delinquency reduction metric below.
No outcome disclosed.
JVM
United States · Real estate · 2020
RealPage's own blog quotes Kortney Balas, described as JVM's Vice President of Information Management, saying the Covid 19 pandemic was a major test of AI Screening's ability to predict a renter's willingness to prioritize rent over other bills, and that JVM ended April at an improved 1.7% delinquency rate with AI Screening in place. The source names the customer only as "JVM"; it is not independently confirmed which company this is.
No outcome disclosed.
R.A. Snyder Properties
United States · Real estate · 2020
R.A. Snyder Properties, a multifamily operator and a RealPage On-Site customer since 2005, rolled out RealPage AI Screening across nearly 300 rental communities as a beta customer. RealPage's own video case study quotes the company's Sandi Bonney describing the predictive scoring as helping the team understand both an applicant's capability and willingness to pay rent, and says the biggest payoff since rolling out to the entire portfolio has been finding quality applicants who would have been overlooked by traditional screening.
No outcome disclosed.
Camden Property Trust
United States · Real estate · 2019
Camden Property Trust, a multifamily owner and operator RealPage describes as a 56,000 unit company, uses RealPage AI Screening to score rental applicants. In a RealPage customer video, Roman Stevens, Camden's Regional Vice President of Operations, says the company has significantly lowered operational expenses and run rates since adopting AI Screening, alongside benefits to short and long term net operating income and property value. RealPage's own case listing groups Camden with CF Real Estate, JVM and R.A. Snyder as AI Screening customers.
No outcome disclosed.
Matrix Residential
United States · Real estate · 2019
Matrix Residential activated RealPage AI Screening, the willingness to pay scoring model, across its portfolio. Nicholas Smith, the company's Software Manager, said the transition was simple and its impact was already evident, and separately three members of the Matrix Residential team described how the model increased their confidence in the resident screening process and their expectation that it would help lift net operating income and cut delinquency.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Historical lease outcomes (paid on time, late, skipped, evicted) to train or validate the model
- Credit, criminal, eviction and income data from applicants and screening bureaus
- A record of protected characteristics or proxies for fair housing testing, held separately from the scoring process
- Published, consistent approval thresholds and override rules
Systems to integrate
- Property management system for applications and lease data
- Credit bureau and eviction and criminal record data providers
- Adverse action notice generation and delivery
- A compliance dashboard for periodic disparate impact testing
Complexity: High
The scoring model itself is usually bought from a specialist vendor, not built in house, but fitting it into a compliant process is hard: adverse action notices, dispute handling under the Fair Credit Reporting Act, and a genuine, tested process for detecting and correcting disparate impact by protected class and, in many US jurisdictions, by voucher or source of income status.
- 1
Separate the score from the decision
Treat the model's output as one input to a documented decision policy with human set thresholds, not as the decision itself, so the policy, not only the model, can be tested and changed.
- 2
Test for disparate impact before and after launch, and on a schedule
Run the score against a protected class and source of income breakdown of applicants before launch and on a recurring schedule after, using a method a fair housing lawyer has reviewed; the SafeRent case shows courts will look at outcomes, not intent.
- 3
Do not let the score penalize what it does not understand
A voucher, a co signer or a security deposit changes what an applicant will actually pay each month. Confirm the score or the policy around it accounts for these, rather than scoring on raw income and credit history alone.
- 4
Automate the adverse action notice, do not skip it
Every decline or less favourable term that relies on a consumer report needs a compliant Fair Credit Reporting Act notice naming the screening company and stating the right to a free copy of the report and to dispute it; generate it automatically from the same data the score used.
- 5
Give applicants and staff a way to challenge a score
A documented override and appeal path, reviewed by a person, catches cases the model gets wrong and creates a record that the process is genuinely supervised.
Guardrails
- A human set, documented decision policy sits between the score and the outcome
- Periodic disparate impact testing by protected characteristic and source of income, with a named owner
- Automated, compliant adverse action notices for every unfavourable decision that uses a consumer report
- A logged override and appeal path for applicants and staff
KPIs to instrument
- Approval rate and score distribution by protected characteristic and by source of income status
- Delinquency, skip and eviction rate versus the pre AI Screening baseline, on comparable properties
- Override rate and outcomes of appealed decisions
- Adverse action notices sent on time and disputes received
Human in the loop
Leasing staff and a compliance or fair housing officer own the approval policy, the thresholds and every override; the model informs, it does not decide alone. A cross functional review, including legal or compliance, signs off on the model or policy before launch and on any material change.
Common failure modes
- Disparate impact hiding in a proxy
- A model can discriminate on a protected characteristic through a correlated input, such as credit history, without ever using that characteristic directly. The SafeRent case turned on exactly this. Test outcomes, not just inputs.
- Voucher income treated as if it were not real
- Scoring only the tenant's own income when a housing authority pays most of the rent directly undercounts a reliable payer. Model or policy the voucher payment explicitly.
- A frozen threshold in a changing market
- An approval threshold tuned for one market or one point in the cycle keeps rejecting good applicants, or keeps approving bad risk, as conditions change. Revisit thresholds on a schedule against real outcomes.
- No one can explain a decline
- Staff cannot tell an applicant why they were declined beyond "the system said so", which leaves the applicant unable to challenge what drove the decision and weakens the override and appeal path. Keep the score's reasons in a form a person can read and repeat.
What are the risks and rules?
EU AI Act
High risk
A score used to decide whether a natural person is offered housing evaluates creditworthiness in substance, which is the likely reading of Annex III point 5(b) when the outcome is a rental decision rather than a loan, though the annex text itself only names creditworthiness evaluation and credit scoring. The provider of such a high risk system carries risk management, data governance and conformity assessment obligations; the deployer, the landlord or property manager, must use it according to its instructions, ensure human oversight, monitor its operation and keep logs under Article 26, and, where it falls under Annex III point 5(b), carry out a fundamental rights impact assessment under Article 27.
Rules that apply
Guidance
- 42 U.S. Code 3604, discrimination in the sale or rental of housing (Fair Housing Act) (United States Congress, North America). Prohibits discrimination in the terms, conditions or provision of a rental based on a protected characteristic. In Louis v. SafeRent Solutions, a federal court held in July 2023 that a tenant screening company is subject to this Act even though it is not itself a landlord, and let the claims against the landlord proceed too.
- Fair Credit Reporting Act (Federal Trade Commission, North America). Governs the accuracy, use and disclosure of consumer reports, including tenant screening reports, and requires an adverse action notice when a report contributes to a decline or a less favourable term.
- HUD Issues Fair Housing Act Guidance on Applications of Artificial Intelligence (May 2024, archived) (US Department of Housing and Urban Development, North America). In May 2024, HUD issued guidance on how the Fair Housing Act applies to tenant screening when AI and algorithms are used. The page was archived on February 3, 2025; the archive banner does not state that the guidance was withdrawn, and the underlying statute still applies.
- Annex III, high risk AI systems (point 5, essential private and public services) (European Union, Europe). Lists creditworthiness evaluation and credit scoring of natural persons as high risk.
Controls to put in place
- Documented, human set decision policy separate from the raw score
- Recurring disparate impact testing by protected characteristic and source of income status, reviewed by compliance
- Automated Fair Credit Reporting Act adverse action notices naming the screening company, with free report and dispute rights
- Logged overrides, appeals and their outcomes
- Change control and retesting whenever the model, its inputs or the approval thresholds change
When it went wrong elsewhere
- Louis v. SafeRent Solutions, tenant screening algorithm settlement. Plaintiffs alleged that SafeRent's tenant screening score gave disproportionately low scores to Black and Hispanic applicants using housing vouchers because its algorithm did not account for the portion of rent a housing authority pays directly, in breach of the Fair Housing Act. A federal court held in July 2023 that a screening company can be subject to the Fair Housing Act even though it is not a landlord, and granted final approval of a $2.275 million settlement with injunctive relief in November 2024, including a requirement that SafeRent stop using its scoring algorithm for voucher holders' applications in Massachusetts.
Frequently asked questions
- Does AI tenant screening actually reduce losses?
RealPage reports that its AI Screening delivered an average savings of $39 per unit per year in the first full year of results in the field, and that CF Real Estate reported a 75% drop in skips and evictions after implementing it. These are RealPage's own reported results, the $39 an average it gives with no disclosed method and the 75% one customer's report, not an independent audit.
- Can a tenant screening algorithm discriminate even without meaning to?
Yes. In Louis v. SafeRent Solutions, a federal court allowed claims to proceed that a screening score's algorithm, by not accounting for the share of rent a housing authority pays directly for voucher holders, produced a disparate impact on Black and Hispanic applicants, and the case settled for $2.275 million with injunctive relief. Fair housing law looks at the outcome, not whether the algorithm's designer intended it.
- Is a landlord responsible if a third party algorithm makes the screening decision?
In Louis v. SafeRent Solutions, a federal court in Massachusetts held in July 2023 that the screening company was subject to the Fair Housing Act even though it is not itself a landlord, and let the Fair Housing Act claims against the landlord proceed too. Keep a human set decision policy and a documented override path so no single score alone determines the outcome.
- What is the difference between this and an AI leasing chatbot?
A leasing agent answers questions, books tours and takes maintenance requests, and should stay out of screening and approval decisions entirely. This use case is the scoring and decision system itself, which carries the fair housing and credit reporting risk that a leasing chatbot is built specifically to avoid.
How to cite this page
Blits.ai AI Use Case Library, "AI tenant screening with fair housing safeguards", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/ai-tenant-screening-with-fair-housing-safeguards. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 1 October 2026: Added deployments: Kane Realty Corporation.
- 29 September 2026: First published