A screening decision made by software is still your screening decision. That is the single most important thing for a property management company to understand about automated applicant screening, and it is the thing most vendor marketing quietly avoids saying.

The appeal of algorithmic screening is obvious: faster decisions, consistent criteria, less staff time per application. Those benefits are real. What does not come with them is a transfer of liability. If an automated system declines applicants in a pattern that violates fair housing law, the housing provider is the one facing the complaint. This article covers what that means operationally, and what a defensible process looks like.

A note before going further: this is general background rather than legal advice. Fair housing law is federal, state, and local, enforcement posture changes, and your specific obligations depend on jurisdiction and programme participation. Have counsel review your screening criteria.

What the Federal Guidance Established

In April 2024, HUD's Office of Fair Housing and Equal Opportunity issued two guidance documents addressing automated systems: one on screening applicants for rental housing and one on advertising through digital platforms. They were announced together as guidance on applications of artificial intelligence under the Fair Housing Act.

The documents are worth reading rather than summarising, but two principles from them shape everything a property manager should do.

1. The Algorithm Is Not a Defence

HUD's screening guidance states that a housing provider or a tenant screening company can violate the Fair Housing Act by using a protected characteristic, or a proxy for one, as a screening criterion, and that this holds true even where the decision about how to screen is made in whole or in part by an automated system, including one using machine learning or another form of AI.

Read that clause carefully, because it closes the escape route people assume exists. "We use a third-party scoring product and we do not see how it works" is a description of the problem, not an answer to it. The provider who acted on the output is in scope.

2. Proxies Count

The guidance is explicit that a criterion need not name a protected characteristic to create liability. It can serve as a stand-in for one.

This is where automated systems create risk that manual screening does not, because a model trained to predict tenancy outcomes will find correlations wherever they exist in the data, including correlations that track protected characteristics closely. Nobody has to intend that. It emerges from the optimisation, and the more variables a model considers, the harder it becomes to establish that none of them functions as a proxy.

3. Where the Guidance Sits Now

Both documents are currently hosted on HUD's archive server rather than its main site, which means the federal position in this area is in motion. Enforcement priorities shift between administrations, and there has been active rulemaking touching how discrimination claims are assessed.

Two things follow. Confirm the current status of any federal guidance before relying on it. And do not read federal deregulation, if it occurs, as the risk disappearing. State attorneys general, state fair housing agencies, and private plaintiffs all have independent enforcement routes, and several states have been active in this area regardless of federal posture. The exposure moves rather than evaporating.

Where Automated Screening Creates Exposure

The risk is not in using software. It is in a small number of specific practices that automation makes easier to do at scale.

Each of these predates AI. What changed is that a decision applied inconsistently by twenty leasing agents produces a messy pattern, while the same flawed criterion applied uniformly by a system produces a clean, documented, statistically visible one.

1. Blanket Criteria With No Individual Assessment

Automatic rejection based on any criminal record, any prior eviction filing, or a credit score below a threshold, applied without regard to circumstances, is the most examined practice in this area.

The concern is not that these factors are irrelevant. It is that categorical exclusions with no room for context tend to produce disparate outcomes, and that an applicant has no route to explain a dismissed case, a sealed record, or a filing that was resolved in their favour.

2. Records Without Dispositions

Screening reports have long included court filings without the outcome attached. An eviction filing that was withdrawn, dismissed, or decided for the tenant can appear identical to one that resulted in a judgment.

Under the Fair Credit Reporting Act, consumer reporting agencies are required to maintain procedures to include existing disposition information for court filings, a point HUD's guidance highlights by reference to a 2024 Federal Register advisory opinion. If your vendor's reports routinely show filings without outcomes, that is a question to put to them in writing.

3. Scores You Cannot Explain

Proprietary scoring products return a number and often a recommendation. Many do not disclose what goes into the number.

Ask yourself a simple test question: if an applicant asks why they were declined, can you answer beyond "the system scored you below our threshold"? If not, you have adopted a decision rule you cannot defend, and the fact that a vendor built it does not change who has to defend it.

4. Advertising Delivery

HUD's second 2024 guidance document addresses digital advertising, where automated ad delivery systems can direct housing adverts toward some audiences and away from others without the advertiser choosing that outcome.

Most property managers think of fair housing as a screening issue. Targeting and delivery sit upstream of screening and are covered too.

What a Defensible Process Looks Like

None of this argues against automation. It argues for automation you can explain.

The three practices below are ordered by how much protection each provides relative to effort. All three are things a management company controls directly, and none requires abandoning software.

1. Start From the Fundamentals

Written criteria applied consistently, adverse action notices with correct wording and timing, documented applicant communication, and a route for applicants to dispute inaccurate information are the baseline for any screening process, automated or not. Our tenant screening process guide covers those in detail and this article assumes them.

One addition specific to automation: check that the configuration in the software matches the written policy. Configuration drifts as people adjust thresholds, and the version governing decisions is the one in the system, not the one in the policy folder. That gap is the most common finding when anyone actually looks.

2. Interrogate the Vendor Before You Buy

The questions below are the ones that separate a product you can defend from one you cannot, and they should be answered in writing, in the agreement, rather than in a sales conversation.

• What data sources feed the product, and how frequently are they refreshed?

• Are court dispositions included where they exist, and what happens when they do not?

• How are records matched to an individual, and what is the false-match rate?

• Has the model been tested for disparate impact, by whom, and will you share the results?

• If we decline an applicant on this output, what explanation can we give them?

• If we face a fair housing complaint, what will you provide, and what does the agreement say about who bears the cost?

Vendors vary enormously in how they answer these, and the willingness to answer at all is itself informative. A vendor who treats the scoring logic as proprietary and will not describe its inputs is asking you to accept a liability you cannot inspect. Our guide to contract management in property management covers what else belongs in a vendor agreement.

3. Keep a Human Decision Point on Exceptions

Full automation is where most of the defensibility disappears. A defined route for an applicant to provide context, considered by a named person, applied consistently, is the practical answer.

Note the word consistently. An exception process used at each member of staff's discretion recreates the problem the automation was meant to solve, in a form that is harder to audit. Write down what qualifies as an exception, who decides, and what evidence gets recorded.

The Tension Nobody Resolves

There is a genuine conflict at the heart of this, and articles on the subject tend to pretend otherwise. Consistency and individual assessment pull against each other. Uniform criteria applied identically to everyone is the classic defence against a claim of intentional discrimination. Individualised consideration of circumstances is the classic defence against a claim that a neutral rule produced a discriminatory outcome. A process built entirely for one is weaker on the other.

The workable answer is neither extreme: objective criteria that are applied consistently, with a defined and documented exception process that is itself applied consistently. That is more work than a fully automated pipeline and it is the reason a purely hands-off screening system is a poor idea regardless of how good the software is.

The second tension is commercial. Faster screening wins applicants in competitive markets, and every human review step costs hours. Anyone claiming there is no trade-off is selling something. The judgement is about where to spend the time, not whether to.

Where the Technology Comes In

Most of what makes a screening process defensible is record-keeping: which criteria were in force on the date of the decision, what the applicant was told, what context they provided, who considered it, and what was decided. That evidence is what a complaint turns on, and it is usually scattered across an applicant tracking system, an email inbox, and somebody's memory.

That is the problem RIOO is built for, and the split is worth being precise about:

• Leasing, applications, lease administration, and tenant records run inside RIOO as a purpose-built property management layer, so the application, the decision, and the correspondence attached to it stay together as one record.

• Finance, multi-entity accounting, and reporting are handled by the NetSuite core RIOO is built on, which is where that depth is native.

• Both draw on one record, so an applicant file can be produced complete rather than reassembled from several systems months later.

The practical effect is that the evidence exists by default rather than by someone remembering to keep it. RIOO runs more than 180,000 units under management across residential and commercial portfolios on that architecture.

Where to Start

Print your current screening criteria, then open the configuration of whatever system applies them, and compare the two. The gap between the written policy and the live configuration is the most common finding in this exercise and the easiest to fix.

Then take ten recent declines and check whether you could explain each one to the applicant in plain language. Any you cannot explain is a decision rule you are relying on without understanding, and that is the list worth working through with counsel.

Book a RIOO Demo

RIOO keeps applications, decisions, and correspondence on one record, so the evidence behind a screening decision is available when it is needed. Book a demo and see how it works across your portfolio.

Frequently Asked Questions

1. Is a property manager liable for a screening decision made by software?
Yes. HUD's 2024 fair housing guidance on tenant screening states that a housing provider or screening company can violate the Fair Housing Act by using a protected characteristic or a proxy for one as a criterion, and that this applies even where the screening decision is made in whole or in part by an automated system including one using machine learning. Using a third-party product does not transfer the obligation to the vendor.

2. What is a proxy in the context of tenant screening?
A criterion that does not name a protected characteristic but functions as a stand-in for one, because it correlates closely with it. Automated systems raise this risk because a model trained to predict outcomes will use whatever correlations exist in its training data, including ones that track protected characteristics, without anyone intending it. The more variables a model uses, the harder it becomes to demonstrate that none acts as a proxy.

3. What should you ask a tenant screening vendor before buying?
What data sources feed the product and how often they refresh; whether court dispositions are included; how records are matched to individuals and the false-match rate; whether the model has been tested for disparate impact and by whom; what explanation you can give a declined applicant; and what the vendor will provide if you face a complaint. Get the answers in the agreement rather than in a sales conversation.

4. Are eviction records on screening reports reliable?
Not always. Court filings have historically appeared without the outcome attached, so a case that was dismissed or decided in the tenant's favour can look the same as one ending in judgment. Under the Fair Credit Reporting Act, consumer reporting agencies must maintain procedures to include existing disposition information for court filings. If your vendor's reports show filings without dispositions, raise it with them in writing.

5. Does fair housing law apply to housing advertising as well as screening?
Yes. HUD issued separate 2024 guidance on advertising through digital platforms, addressing automated ad delivery systems that can direct housing adverts toward some audiences and away from others, including where the advertiser did not choose that targeting. Fair housing obligations attach upstream of the application, not only at the screening stage.