ACF awards $6M for testing predictive analytics in 10 child welfare jurisdictions

The Administration for Children and Families is awarding $6 million to 10 child welfare jurisdictions to pilot predictive analytics tools designed to help caseworkers make faster, better informed decisions about children and families.

The 10 jurisdictions — Indiana, Missouri, Nebraska, New Jersey, North Carolina, Ohio, Oklahoma, Texas, along with the Muscogee Creek Nation and the District of Columbia — will each receive roughly $600,000 through the one-time grants. The three-year pilot projects are expected to begin Sept. 30.

The awards build on ACF’s broader “A Home for Every Child” initiative, which seeks to improve the ratio of foster homes to children in care through prevention, better placements and faster permanency. All 50 states, the District of Columbia and Puerto Rico have joined the initiative.

ACF Assistant Secretary Alex Adams said that the agency received 21 applications from 20 jurisdictions, with proposals spanning several child welfare use cases. Some will use predictive models to help triage hotline calls, while others will focus on foster care placements, permanency or reducing caseworker paperwork. The grants will also require jurisdictions to develop governance and quality-assurance processes, train staff, engage communities and evaluate results.

“Across all of the ones that we’re funding, the thing I’d really emphasize is that they focus on decision support, not decision making,” Adams said in an interview this week, emphasizing that the technology is intended to support, rather than replace, human judgment.

“Predictive risk modeling should help case managers and case workers have better information at their fingertips and allow them to make better decisions,” he said. “But it should never replace decisions that those highly skilled, highly valued humans are making.”

That distinction comes as states increasingly explore predictive risk models despite concerns about data quality, bias and potential impacts on families. Only a handful of child welfare agencies have adopted the technology, though proponents have argued that the tools can help workers identify cases needing intervention while avoiding unnecessary investigations or placements.

Adams hoped that more agencies will take advantage of ACF’s recent guidance on how to incorporate predictive software into their workflows. ACF also offers open-ended funding that matches 50% of state costs for certain expenses.

“The concept is rather than waiting, you know, for the next round of grants and sitting back to any refresh, every state has the ability to draw down federal funds to invest in predictive risk modeling today, and there will be a federal cost share for qualified state expenses,” Adams said.

ACF plans to track the pilot projects alongside “A Home for Every Child” data, allowing officials to assess the impact of the projects more quickly. Adams said the administration will “walk alongside these grantees” as they share what works, and what does not, with other child welfare jurisdictions.