What happened
It was reported that in May 2021, a woman in Massachusetts was denied an apartment after a tenant-screening system gave her a low score, despite having a government housing voucher that would have covered most of her rent.
The software, developed by SafeRent Solutions, helped landlords assess prospective tenants by producing a score based largely on factors such as credit history and debt. The problem, according to the lawsuit, was that the system did not adequately account for housing vouchers, which substantially reduce the amount of rent tenants need to pay themselves.
In May 2022, two women and a housing assistance organisation brought a class action against SafeRent, arguing that its scoring system unfairly disadvantaged Black and Hispanic applicants and people who relied on housing vouchers.
The case was settled in November 2024 for 2.275 million USD, without a court finding that the company had discriminated against applicants. Under the agreement, SafeRent also committed to stop providing scores or rental recommendations for housing voucher holders in Massachusetts for five years, unless its system was independently validated as non-discriminatory.
What an auditable version would have shown
The difficulty with automated tenant screening is that applicants may receive a rejection without understanding how the decision was reached or whether the same system consistently disadvantages people in similar circumstances.
An auditable system would have recorded which factors contributed to each score, how housing vouchers were considered and whether applicants receiving government assistance were rejected more frequently than those who did not need it.
Looking at these results across thousands of applications could help identify unfair patterns that would be almost impossible for an individual renter to recognise.
Where the gap was
Headlights identifies overconfidence, where organisations place too much faith in a scoring system without sufficiently questioning whether its assumptions reflect the circumstances of the people being assessed.
SafeRent’s software was designed to help landlords predict rental risk, but the plaintiffs argued that relying heavily on credit history while giving insufficient weight to housing vouchers could make applicants appear riskier than they actually were.
There is a familiar human tendency behind this kind of decision because numbers often create an impression of objectivity, particularly when they come from software designed to simplify complicated judgements.
A landlord looking at a low score might assume the applicant represents a financial risk, without realising that a government voucher could already guarantee a substantial portion of the rent.
The case illustrates how an apparently neutral scoring method can be challenged for producing unequal outcomes, even when discrimination is not an explicit part of its design.
What governance should have looked like
A system influencing whether someone can secure housing should be tested to understand how its recommendations affect different groups of applicants, particularly when the consequences of rejection can be significant.
A ConductRecord could document the factors used in each assessment, while a MetricRecord could measure whether rejection rates differed between voucher holders and other applicants, allowing potential disparities to be investigated before they became widespread.
Independent testing would also help establish whether the information used to calculate risk genuinely reflected a person’s ability to pay rent, rather than relying on assumptions that might disadvantage particular groups.
The broader lesson is that reducing a person to a number can make decisions easier for organisations, but it does not necessarily make those decisions fairer or more accurate.
Failure Pattern: SafeRent faced allegations that its tenant-screening system unfairly disadvantaged housing voucher holders and minority applicants, resulting in a 2.275 million USD settlement and restrictions on scoring voucher holders in Massachusetts.
Governance Principle: Automated screening systems should be tested for unequal outcomes, with transparent assessment criteria and independent validation where decisions affect access to essential services such as housing.
The reference implementation of ConductRecord and MetricRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.
Sources
- Rental Applicants Using Housing Vouchers Settle Discrimination Class Action Against SafeRent (Cohen Milstein)
- Louis, et al. v. SafeRent Solutions, et al. (Cohen Milstein case page)
- Louis v. SafeRent Solutions, LLC, 1:22-cv-10800 (D. Mass.), Civil Rights Litigation Clearinghouse