AI Risk Scoring for Child Welfare Prevention Services

What it is for

Risk Assessment & Triage

Predicts the likelihood of a child welfare case being involved in an indicated maltreatment investigation within 24 months from the start of a prevention service. Scores are generated on day 10 of service for each active prevention case.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Planning & Decision-making

Risk scores are aggregated to rank and group prevention service providers into four performance cohorts (Very High-Risk, High-Risk, Medium-Risk, Low-Risk) for fair comparative assessment of provider performance during the previous fiscal year.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

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

What it is for

Risk Assessment & Triage

Planning & Decision-making

This AI system predicts the likelihood that a family receiving child welfare prevention services will be involved in a future indicated maltreatment investigation within 24 months. It assigns a numeric risk score to each case on day 10 of service, and those scores are used to group prevention service providers into cohorts for fair performance comparison. No identifying personal information is included in the model's training, inputs, or outputs — only a case number is used.

What it collects and what happens to it

Data taken in

Operational data

Anonymized data

ACS administrative data about prior and current child welfare involvement at case start: SCR (State Central Register) investigations and time spent in foster care. Only a case number is used — no identifying personal information is included in the input data.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Processing

Classification & Prediction

Machine learning model trained on ACS administrative data (July 2010 – June 2019, ~338,467 training observations) that predicts the probability of future indicated maltreatment. Programs are then ranked and assigned to one of four quartile cohorts based on average case risk scores.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

What it does

Deciding (Analytical AI)

Human decides

The model produces a numeric risk score (likelihood of future indicated investigation) and a cohort ranking for each prevention provider. Human reviewers use these scores to assess program performance; the model does not make autonomous decisions about families or providers.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Outputs

A recommendation or prediction

Anonymized data

A numeric likelihood score for each prevention case, and a derived scorecard grouping each prevention service provider into one of four risk cohorts (Very High-Risk, High-Risk, Medium-Risk, Low-Risk). Output is keyed to case numbers only, with no identifying personal information.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Run by

Administration for Children's Services (ACS)

ACS, through its Division of Policy, Planning, and Analysis, deploys and operates the Prevention Score Card model to assess risk levels for families receiving prevention services and to evaluate provider performance cohorts.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Built by

Not stated by the NYC Office of Technology and Innovation.

Kept for

Retained Not specified in the register entry

The register does not specify a retention period for model outputs or input data. Training data covers closed investigations from July 2010 to June 2019, implying that historical administrative data is retained for at least the duration of the model lifecycle.
Duration: Not specified in the register entry

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Shared with

Not available to me

Individual families and children whose data informs the model's predictions do not have direct access to their case's risk score or cohort assignment. The register does not describe any individual access mechanism.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Available to the accountable organization

ACS and its prevention service providers have access to the scorecard outputs, which are used for annual provider performance assessments. Outputs are de-identified and keyed to case numbers only.

NYC Algorithmic Tools Compliance Report — Prevention Score Card

Stored

Not stated by the NYC Office of Technology and Innovation.

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 TransparencyMembers of the public and affected families may learn about the general operation of this model through NYC's public AI Use Cases register, published at data.cityofnewyork.us. The register describes the model's purpose, training data, input and output data, and autonomy level. No specific contact mechanism for further transparency inquiries is listed in the register entry.

Risks and safeguards

Civil liberties harmThe model scores vulnerable families involved in the child welfare system, and cohort rankings influence how prevention providers are evaluated — creating indirect downstream effects on families. Scores applied to historically marginalized communities may reflect and amplify systemic inequities in prior investigation data.Safeguard: ACS retrained the model on recent data and explicitly states that cohort assignment is not a performance assessment of programs but a grouping mechanism for fair comparison. Identifying information is excluded from training, input, and output data; only case numbers are used.
Reputational harmPrevention providers assigned to lower-risk cohorts could be perceived negatively despite the register's explicit clarification that cohort assignment is not a performance judgment. Families associated with high-risk scores may face stigmatization if scores are shared beyond intended use.Safeguard: ACS states that cohort assignment is solely for grouping providers for fair comparison, not for assessing program quality. Identifying information is excluded from all model stages; outputs are linked only to case numbers.