Child welfare is at an inflection point. Agencies are now being asked to use data in ways that go beyond system modernization: to improve safety, strengthen family-based care, reduce unnecessary system involvement, and show measurable results.
Federal momentum is accelerating this shift:
- A 2026 issue brief from the Administration for Children and Families (ACF) highlights predictive risk modeling as a way to generate real-time, data-driven risk assessments that support more accurate and efficient decision making.
- The Children’s Bureau is funding Predictive Analytics in Child Welfare Demonstration Grants to help jurisdictions design, implement, test, and replicate responsible strategies for predictive analytics.
- New Child Welfare Information System (CCWIS) Guidance on Predictive Analytics from the Children’s Bureau encourages agencies to consider how they can responsibly use predictive analytics to modernize their child welfare technology. The costs of predictive analytics and related data integration might be included in CCWIS projects.
These signals represent an important opportunity and a challenge: Predictive analytics can enhance child welfare, but successful use will require more than just building a model.
Using predictive analytics to support judgment, not replace it
Discussions about predictive analytics often begin with, “What analytical model can we build?” when they should really begin with, “What decision, practice-related challenge, or system problem are we trying to improve?”
For a model to create value, child welfare agencies must first select the right use case. After all, even a technically strong model can fail if aimed at the wrong problem. A model that cannot reliably distinguish risk, is poorly calibrated, performs inconsistently across groups, or drifts over time should not be used to inform high-stakes decisions.
After selecting the right use case based on expertise in child welfare practice, agencies can then develop an accurate and reliable model, integrate insights into real workflows, and evaluate whether practice and outcomes improve. Responsible predictive modeling equips workers, supervisors, and agency leaders with insights they need to make frontline decisions about screening, service targeting, matching, case planning, and other aspects of service delivery. It can also inform larger system strategies related to placement capacity, policy reform, and program integrity.
Implementing predictive analytics for better child welfare decisions, practices, and outcomes
Predictive analytics should be designed around the decisions agencies seek to make and should be tested before, during, and after implementation. Agencies can build a structured process that helps them turn analytic possibility into practical improvement by:
- Exploring and prioritizing opportunities. Input from agency leaders, staff, and community partners can shed light on top priorities, decision points, data readiness, workflows, and potential use cases for predictive modeling.
- Defining the selected use case. After a use case is selected, agencies will need to clearly define what decision it is intended to inform, the population and outcomes it will affect, and the requirements for implementation. Different use cases will require different data, model designs, safeguards, and implementation supports, so it is important to tackle this step early.
- Developing and validating the model. Agencies should use child welfare and, where appropriate, cross-system data to build or adapt and validate the selected model. Validation will include testing performance, calibration, stability, subgroup differences, and usability.
- Implementing, monitoring, and scaling. Implementation pilots with guardrails for safety and transparency help agencies determine if staff understand and use the predictive analytics tool, workflows support the intended decision, supervisors have the capacity to respond, and the model is ready to scale.
Understanding the problem before deploying a solution
Responsible predictive analytics requires partnership with those affected most by the use case: frontline staff, supervisors, agency leaders, community partners, and people served by the child welfare system. Their perspectives can help address the questions that technical analysis alone cannot answer: What problem matters most? What information do child welfare workers need? How should outputs be presented? What language could stigmatize families? What safeguards are needed to monitor disparate impact?
Mathematica has engaged state and local agencies to help make sure their predictive analytics efforts are technically sound, useful in practice, trusted, and sustainable. Our work with the Arizona Department of Child Safety and Los Angeles County Department of Children and Family Services shows how predictive insights can translate into practical strategies that support different child welfare goals. These strategies work best when they are grounded in the right question, informed by the people closest to the work, and connected to implementation and learning.
- In Arizona, our Next Event Study relied on predictive modeling for policy analysis and strategic planning, helping leaders examine how different responses to child welfare reports might affect children and families and informing efforts to reduce unnecessary involvement with child protective services. This work drew on both data analytics and input from community advisory councils and helped Arizona direct its prevention and treatment efforts to substance-exposed newborns and families experiencing domestic or intimate partner violence.
- In Los Angeles, we supported implementation monitoring for a risk-stratification pilot designed to strengthen supervision, early engagement, and collaborative practice in high-stakes investigations. Staff feedback helped identify practical tool refinements, workflow needs, fairness concerns, and protocols needed for scaling.
Supporting decisions at all levels
Predictive risk modeling can support individual decisions as well as system-level strategies. It can assist in areas ranging from screening, prevention targeting, matching, and permanency planning to placement demand forecasting, caseload monitoring, hotline reform, and program integrity. Predictive modeling can advance efforts such as ACF’s A Home for Every Child initiative by helping agencies expand foster care capacity and reduce unnecessary entries into care or time spent in care. It can also help agencies modernize their child welfare technology and CCWIS by using data to strengthen their practices, better target their resources, improve permanency, and enhance accountability.
The question is not is whether child welfare agencies need better ways to support high-stakes decisions. They do. The question is whether the field will modernize responsibly, with accurate models, transparent safeguards, strong supervision, meaningful engagement, and the workforce and families at the center. Predictive analytics should be treated as a disciplined strategy for improving the decisions that shape children’s and families’ lives.
