Leveraging Data and Partnerships to Help States Navigate SNAP Policy Changes

Session Submission Type: Panel

Abstract

The passage of H.R. 1 (2025) introduced significant changes to the Supplemental Nutrition Assistance Program (SNAP), most notably by requiring state agencies to reduce payment error rates below six percent. For states exceeding this threshold, the fiscal impact would be immense. These looming penalties, combined with new work requirements, place immense pressure on state Departments of Social Services to rapidly adapt. Amidst all of this, these Departments are committed to continuing their pre-H.R. 1 priorities of identifying clients whose lives would be materially improved by SNAP but are not enrolled in the program. 

This panel explores how researchers and state agencies are collaborating to navigate these changes and improve program administration through the use of administrative data, predictive modeling, and artificial intelligence (AI). The papers in this session highlight the dual challenge facing state agencies: the need for sophisticated methods to respond to H.R. 1 and the simultaneous goal of minimizing the decline in overall participation triggered by the bill by increasing take-up among underserved groups. 

Three of the papers in this session focus on payment error rate reduction, using three different approaches. The first paper examines the use of generative AI to help caseworkers identify potential application errors before final determinations are made. The second paper leverages historical data to create a predictive model for high error risk cases in Connecticut, paired with a randomized trial to test the model’s effectiveness in the field. The third paper is similarly leveraging analytic models to help agencies identify process or system improvements and prioritize cases for review, and highlights how this work was operationalized through ongoing workshops with 15 states. The fourth paper focuses on how predictive modeling can help education systems identify eligible college students to increase take-up of SNAP, which has been a high priority across California’s government in recent years.   

Our discussants represent two state Departments of Social Services, who can speak not only to the quality of the research but its potential for real-world impact, and how researchers can support government agencies as they navigate urgent policy challenges.

Policy Area

• Poverty and Income Policy

Chair

• Carrie Welton, Trellis Strategies

Discussants

• Kelsey Pukelis, California Department of Social Services & Harvard University

• Daniel Giacomi, Connecticut Department of Social Services

Organizer

• Jennifer Hogg, University of California, Berkeley

Individual Presentations

• Using Generative AI to Help States Avoid SNAP Payment Error Penalties - Presenting Author: Alex Chohlas-Wood, New York University

• Targeting SNAP Quality Control with a Predictive Model: A Randomized Pilot in Connecticut - Presenting Author: Anthony Lollo, Yale University

• Simulating College Student Eligibility for SNAP to Close the Take-Up Gap - Presenting Author: Jennifer Hogg, University of California, Berkeley

• Working with state agencies to create public resources for modeling SNAP errors - Presenting Author: Eric Giannella, Georgetown University