AI Risk Scoring for Child Welfare Investigations

What it is for

Risk Assessment & Triage

Scores open child welfare investigation cases to identify those with the highest likelihood that a child will experience severe harm (substantiated physical or sexual abuse) within the next 24 months. Used exclusively to prioritize which cases the Quality Assurance unit reviews; not used to determine services or interventions for children or families.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Run by: Administration for Children's Services (ACS)
Where: No fixed location
Kept: Retained not specified in the register entry
Shared with: Accountable organization, Not available to vendor
Your copy: You cannot see the data it holds about you. What you can do

What it is for

Risk Assessment & Triage

This tool uses machine learning to score open child welfare investigation cases by their likelihood of severe harm — defined as substantiated physical or sexual abuse within the next 24 months. Each day, the highest-scoring cases are surfaced for a Quality Assurance review team at the NYC Administration for Children's Services, which can review roughly 3,000 of the approximately 50,000 annual investigations. The model does not determine services or interventions for any child or family; it only helps prioritize which cases receive additional internal oversight. Scores are not shared with frontline investigators.

What it collects and what happens to it

Data taken in

Sensitive personal information

Pseudonymous data

ACS administrative data about children involved in active investigations, including prior and current child welfare involvement (investigations triggered by a New York State Central Register call and time spent in foster care). The only identifier retained is the case number; no directly identifying information about individuals is collected or disclosed.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Operational data

Anonymized data

ACS historic and current administrative records about child welfare investigations, including investigation case records from the New York State Central Register and foster care records. Only ACS administrative data are used; no external data sources are incorporated.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Processing

Classification & Prediction

A machine learning model trained on approximately 815,284 historical closed investigation cases (2012–2022) predicts the likelihood that a child in an open investigation will experience substantiated severe harm (physical or sexual abuse) within 24 months. Each investigation is assigned the score of the child with the highest individual likelihood within that case, producing a ranked list for QA selection.

NYC Algorithmic Tools Compliance Report — ASAP Tool

What it does

Deciding (Analytical AI)

Human decides

The model produces a numeric likelihood score for each open investigation case and ranks cases by that score. The ACS Quality Assurance unit uses the ranked list to select which cases to review; all decisions about services or interventions remain with human staff.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Outputs

A recommendation or prediction

Pseudonymous data

A rank-ordered list of open investigation cases involving children with the highest predicted likelihood of future severe harm, for use by the ACS Quality Assurance unit in selecting approximately 3,000 cases for review annually. Scores are not shared with field investigators or frontline staff; the output is advisory to the QA selection process only.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Run by

Administration for Children's Services (ACS)

ACS develops and operates the ASAP Tool through its Division of Policy, Planning, and Analysis. The ACS Quality Assurance unit in the Division of Child Protection uses the tool's outputs to select approximately 3,000 active investigation cases annually for additional review.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Built by

Not stated by the NYC Office of Technology and Innovation.

Kept for

Retained not specified in the register entry

The register entry does not specify a retention period for model outputs (ranked case lists and scores). ACS administrative case records are subject to separate city and state records retention requirements not detailed in this entry.
Duration: not specified in the register entry

NYC Algorithmic Tools Compliance Report — ASAP Tool

Shared with

Available to the accountable organization

Output scores and ranked case lists are available only to the ACS Quality Assurance unit within the Division of Child Protection. Scores are explicitly not shared with field investigators conducting the active investigations.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Not available to me

Individuals and families involved in ACS investigations do not have access to the risk scores assigned to their cases, nor are they informed that a predictive score has been calculated or used to prioritize their case for QA review.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Not available to vendor

The ASAP Tool was developed internally by ACS with no external vendor. There is no third-party vendor with access to model outputs or case data.

NYC Algorithmic Tools Compliance Report — ASAP Tool

Stored

Stored locally

The tool was developed internally by ACS using ACS administrative data. No vendor or third-party cloud storage is mentioned; storage is presumed to be within ACS and NYC government infrastructure.
Duration: not specified in the register entry

NYC Algorithmic Tools Compliance Report — ASAP Tool

How to read the colours

Can it identify you?

Anonymized data: Data about people with the link to who is broken. Stripped of identifiers, blurred, aggregated, or noised so this system can’t reasonably tie a record back to an individual.
Pseudonymous data: Each person’s data is tied to a token (hash, ID, template) that lets this system recognise the same person across events, but the token itself doesn’t reveal a name. Reidentification is possible with extra information.
Identifiable data: The data either contains a direct identifier (name, address, account name, recognisable face or voice, plate number) or carries a token this system uses to look up legal identity during processing.

Who completes the loop?

Human decides: This mode suggests; a person decides what to do next. The AI is always advisory — a human is in the loop on every decision. Example: a triage tool ranks cases for a clinician who chooses which to see first.
Human executes: This mode decides; a person carries out the result. Example: an optimizer plans the day’s trash-collection routes, and drivers run them.
Autonomous: This mode decides and acts on its own. No person reviews each decision or carries out the resulting action.

Definitions from the DTPR standard. Amber is about your data, violet about who decides. The fuller the shape and the deeper the colour, the more identifying the data or the less a person is involved.

What you can do

Ask about this system

Questions go to the NYC Office of Technology and Innovation, not the vendor.

Your rights

Right to Algorithmic TransparencyACS publishes information about the ASAP Tool in the NYC Algorithmic Tools Compliance Report, which describes the model's purpose, inputs, outputs, and limitations. Members of the public may access this register at https://data.cityofnewyork.us/City-Government/Algorithmic-Tools-Compliance-Report/jaw4-yuem. The register entry discloses that the tool uses machine learning to rank child welfare cases for internal QA review and that it does not influence intervention or service decisions.
Right to Be Informed of AI UseFamilies and individuals involved in ACS investigations are not individually notified that an AI system has scored their case. General disclosure occurs through the public NYC Algorithmic Tools Compliance Report. There is no described mechanism for an individual to learn whether their specific case was scored or selected for QA review.

Risks and safeguards

Civil liberties harmPredictive scoring of child welfare cases may systematically disadvantage families already over-represented in the child welfare system due to structural inequities, potentially amplifying existing biases through algorithmic recirculation. Individuals and families affected by ACS investigations are not informed that a predictive score influences which cases receive additional QA scrutiny.Safeguard: The tool's use is strictly limited to internal QA case selection and explicitly does not influence decisions about services or interventions. Scores are withheld from frontline investigators. The model was retrained on newer data (through 2022) to address data drift from policy changes, reducing the risk of decisions based on outdated patterns.
Reputational harmFamilies whose cases are flagged by the model for additional QA review may be subject to greater scrutiny without being aware that an algorithmic score influenced their selection, raising concerns about stigmatization if the system's use becomes known without appropriate context.Safeguard: Scores are not shared with field staff conducting investigations, limiting the direct labeling effect on individuals. The tool is scoped exclusively to internal QA prioritization, not to any external-facing decision about the family.