The Allegheny Housing Assessment — Pittsburgh's Machine-Learning Tool Improves Housing Prioritization, But an Independent Study Finds Racial Disparities Persist

Allegheny County's Department of Human Services (DHS) needed to decide who gets scarce, priority access to permanent supportive housing and rapid re-housing among people experiencing homelessness. Its previous tool, the VI-SPDAT survey used nationwide, had been shown to score white clients as higher-risk than Black clients with similar circumstances, and front-line staff found it slow to administer.
Since August 2020, DHS has used the Allegheny Housing Assessment (AHA), a predictive model trained on 18,008 historical housing assessments (2016–2024) from the county's integrated data warehouse. AHA predicts the likelihood of four adverse outcomes within 12 months if someone remains unhoused — a psychiatric inpatient stay, four or more emergency-room visits, a jail booking, or a further episode of homelessness — and converts this into a 1-10 priority score. A related Mental-Health AHA model, trained on 13,673 assessments, was added in February 2023 for Medicaid-enrolled clients. In its most recent (2025) validation, AHA's four component models show AUCs of 0.66-0.72 and positive predictive values of 29%-58%.
County staff report the tool makes prioritization decisions faster and more consistent than the survey it replaced, and a peer-reviewed 2024 study (AAAI/ACM Conference on AI, Ethics, and Society) found that, after deployment, actual service decisions became better aligned with the model's own risk scores, and the score distribution is now similar across racial groups.
The same peer-reviewed study found that despite these improvements, the disparity in who actually receives services by race did not shrink. The researchers trace this to continued use of a survey-based fallback tool (Alt-AHA) where data quality is poor, and to demographic differences in eligibility criteria such as chronic-homelessness and veteran status — a reminder that a fairer-looking algorithm does not by itself fix unequal outcomes.

Read the full analysis: https://analytics.alleghenycounty.us/2026/01/16/improving-prioritization-of-housing-services-implementation-of-the-allegheny-housing-assessment/