Chicago's model finds critical violations 7.5 days earlier

AI system

System: Food Inspection Forecasting (predictive analytics model, open-sourced in R)
Model: Not assessed
Release: Not assessed
Application developer: City of Chicago Department of Innovation and Technology with the Department of Public Health, Allstate Insurance data scientists and the Civic Consulting Alliance
Role: Not assessed
Configuration: Not assessed
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: Open inspection, 311 complaint, business license and weather data on the city's data portal. Fewer than three dozen inspectors for 15,000+ establishments, so ordering matters. A fixed two-month inspection record to re-score.
Human oversight: Not assessed
Relation to AI: Developer and deployer are the same body. The City of Chicago built the model in-house with pro bono partners and the Department of Public Health is the user. The evaluation re-scored inspections after they were completed, so the model did not change any inspection in the pilot. Model type is not stated in the sources.
AI contribution: Developer and deployer are the same body. The City of Chicago built the model in-house with pro bono partners and the Department of Public Health is the user. The evaluation re-scored inspections after they were completed, so the model did not change any inspection in the pilot. Model type is not stated in the sources.

Outcome identifiers: Not assessed
Benefit: 69% of critical-violation sites in the first half of the work versus 55%, 7.5 days earlier
Category: Improved Task Accuracy
Status: Not assessed
Beneficiary groups: Organizations. Society.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: 69% of critical-violation sites in the first half of the work versus 55%, 7.5 days earlier
Comparison: the Department of Public Health's normal inspection order during the September-October 2014 pilot
Results: Reference: 55 percent (141 establishments) of those with critical violations found in the first month during normal operations. Result: 69 percent (178 establishments) found in the first half of work; violations found on average 7 and a half days earlier.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Participants: Developer: City of Chicago Department of Innovation and Technology with the Department of Public Health, Allstate Insurance data scientists and the Civic Consulting Alliance. Operator: Chicago Department of Public Health.
Jurisdictions: Benefit: Chicago, United States. Developer: United States. Operator: United States.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Conditions for replication: The predictor set (prior violations, nearby sanitation complaints, facility type, days since last inspection, inspector) remains available. A critical-violation base rate near 15%.
Replication attempts: Not assessed
Later findings: 10 October 2019. An independent 2019 hindsight analysis states that CDPH deployed the model in 2015 to schedule inspections of Chicago's 15,000-plus food establishments and released the training data and code.
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Assessed benefits: Organizations. Society.
Depth: Not assessed
Reasons: Organizations: The sources document a retrospective simulation (69% vs 55% found in the first half; 7.5 days earlier) and no live-deployment outcome. Society: Earlier discovery “potentially after patrons became ill” is stated as a rationale. No illness outcome is documented.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Independently reported

Kind: Not assessed
Related entries: Not assessed
Field: Food safety / public health. Restaurant inspection targeting.
Purpose: Forecast which Chicago food establishments are most likely to have critical sanitation violations so that inspectors visit them first.
Dates: Date basis unassessed: 2015 (precision unassessed). Updated: 21 October 2026.
Stage: Not assessed
Version: SHA-256: eae4cef5a98e…

Report Developer

City of Chicago. Food Inspection Forecasting: Optimizing Inspections with Analytics. chicago.github.io (undated; copyright 2014-2017).
Preprint Independent author

Kannan V, Shapiro MA, Bilgic M. Hindsight Analysis of the Chicago Food Inspection Forecasting Model. arXiv:1910.04906, submitted 10 October 2019.
News report Independent author

Thornton S. Delivering Faster Results with Food Inspection Forecasting: Chicago's Analytics-Driven Plan to Prevent Foodborne Illness. Data-Smart City Solutions, Harvard Ash Center, 19 May 2015.