HUD’s mission is to create strong, sustainable, inclusive communities and quality affordable homes for all. By leveraging the benefits of AI, the Department can amplify our positive impacts on the people and communities that we serve.

2024 AI Inventory

Automating Draft Counterparty Credit Narrative Reports - Ginnie Mae performs counterparty credit reviews of mortgage issuers who participate in Ginnie Mae’s program. The reviews are based on written analysis and financial data. As an initial step in counterparty credit reviews, Ginnie Mae uses Natural Language Generation (NLG) to implement coded rules and generate draft narratives. This application of AI enables processing efficiency and reduces errors that can occur in a manual process.

Counterparty Risk Anomaly Detection - Ginnie Mae is responsible for analyzing counterparty risk profiles of mortgage issuers who participate in Ginnie Mae’s program. Ginnie Mae analyzes data from multiple sources to identify potential risks and areas of focus. To enhance the identification of data patterns, Ginnie Mae uses machine learning algorithms, specifically clustering and genetic techniques. These algorithms detect potential risk areas, enabling a focused approach to subsequent analysis.

Subledger Data Quality Machine Learning - Ginnie Mae analyzes Master Sub-Servicer (MSS) transaction data on a monthly cadence. Through its use of machine learning models, Ginnie Mae has enhanced its ability to identify data inconsistencies and exceptions associated with its MSS transaction data. By automating the analysis of a significant number of MSS transactions, the machine learning algorithms enhance the efficiency and accuracy of key reporting processes.

Quantitative Text Analysis - The Office of the Chief Risk Officer (OCRO) has created a Fraud Risk Analysis Report using Quantitative Text Analysis (QTA). QTA uses machine learning algorithms, which were applied to publicly available reports published by the Office of the Inspector General, Department of Justice, U.S. Government Accountability Office, as well as other public and non-profit organizations, such as the Internet Crime Complaint Center (IC3) and Identity Theft Resource Center (ITRC). The application of the algorithms identified clusters of words and phrases related to fraud across this collection of publicly available reports. A human analyst subsequently reviewed the output to discern key trends and themes across the documents. The report of fraud trends and themes is being used to enhance awareness and training materials to strengthen HUD’s Enterprise Fraud Risk Management program.

Voice of the Customer - The Office of the Chief Financial Officer, Customer Experience Team works across the department to develop a deep understanding of who our customers are and how we can best serve their needs. The Voice of the Customer application brings the best-in-class voice transcription, speech, and text analytics to customer feedback surveys, contact center calls and chats. It allows HUD to unlock critical insights that provide deeper understanding of how to support and manage our program/service delivery and contact center providers, and ultimately to improve customers’ experiences.

Google Translate - The Office of Public Affairs is using Google Translate to translate information on public-facing websites. On every HUD.gov webpage, there is an “Español” button that calls Google Translate to translate the page into Spanish. From there, a Google Translate banner appears on the page allowing for the translation into any available language. This is a free, off-the-shelf solution with no customization. It is also an interim solution to improve language access at HUD while a permanent solution is being developed in line with HUD’s Language Access Plan in response to Executive Order 13166. 

Consolidated Plan Analysis - In March 2023, the Office of Policy Development and Research began a proof of concept to analyze aspects of HUD's Consolidated Plans. HUD requires grantees of its formula block grant programs to submit Consolidated Plans, which are meant to identify and assess affordable housing and community development needs and market conditions. These plans are publicly available via HUD's website. HUD staff currently review these plans for compliance, but HUD lacks the capacity to do in-depth analysis of commonalities or trends contained within plans. This proof of concept will explore creating a database and chatbot that will enable HUD staff to query features of the nearly 1,000 active Consolidated Plans. This exercise has the potential to inform grantees, technical assistance, and other programmatic tweaks, as well as inform how advanced data science tools can benefit our programs and operations.