Zero Recidivism Over 8 Years: Shanghai Xuhui District’s Electronic Monitoring Program

Case Study: How Xuhui District Achieved 8 Years of Zero Recidivism

The Challenge

Shanghai’s Xuhui District manages community correction for offenders on probation, parole, and supervised release within one of China’s most densely populated urban areas. The district’s correction authorities faced a challenge familiar to agencies worldwide: how to effectively supervise hundreds of offenders in the community while preventing recidivism and ensuring public safety.

Traditional supervision methods — periodic reporting, random home visits, manual check-ins — provided only intermittent snapshots of offender behavior. Between contacts, supervisors had no visibility into movement patterns, association behaviors, or compliance with geographic restrictions. Drug-related cases posed particular challenges: accomplice networks often spanned multiple districts, and traditional methods couldn’t detect cross-regional coordination patterns.

The Solution: CO-EYE GPS Ankle Monitoring + AI Analytics

Phase 1: GPS Monitoring Deployment

Xuhui District deployed CO-EYE one-piece GPS ankle monitors across its community correction population. The system was configured for 5-minute continuous positioning — far more frequent than most international deployments — combined with multi-source environmental data collection:

• GPS satellite positioning — Primary outdoor tracking at 5-minute intervals
• Wi-Fi positioning — Indoor location when GPS is unavailable
• Cellular LBS — Fallback positioning via cell tower triangulation
• Electromagnetic fingerprint data — Continuous environmental signal profiling that creates unique location signatures beyond coordinate positions

The 5-minute sampling rate was deliberate. While international standards typically require GPS storage every 3–15 minutes, Xuhui’s correction authorities determined that high-frequency data collection was essential for the next phase: behavioral pattern analysis. Every additional data point increases the resolution of behavioral models.

Phase 2: AI-Powered Behavioral Analytics

Using 10+ years of accumulated GPS data from CO-EYE devices, Xuhui’s correction team — in partnership with grassroots community correction cadres who provided domain expertise — developed an AI data analysis system. The investment was substantial (millions of RMB), but the resulting system provided capabilities that manual supervision could never achieve:

• Behavioral labeling engine — The AI system automatically categorized offender behavior patterns into hundreds of labels covering drug-related activity, alcohol consumption patterns, theft-associated movement, fraud indicators, gambling behavior, and other risk categories.
• Drug-related geographic database — The system automatically built a database correlating specific locations with drug activity, based on aggregated movement patterns of known drug offenders over years of data. When a new offender began frequenting these locations, the system flagged the behavior for investigation.
• Cross-regional accomplice tracking — By analyzing movement pattern correlations between offenders across different districts, the system could identify previously unknown associations. When two offenders from different districts repeatedly appeared at the same locations at the same times — a pattern invisible to any single district’s supervisors — the system detected the correlation automatically.
• Predictive risk scoring — The behavioral labels fed into a risk model that provided dynamic, continuously updated risk assessments — far more responsive than static instruments based on criminal history alone.

Training the System

A critical success factor was the involvement of frontline correction staff in training the AI system. Grassroots correction cadres — the officers who know individual offenders and local context — cleaned and labeled the raw positioning data. This human-in-the-loop approach ensured that the AI’s behavioral classifications reflected real-world operational knowledge, not just statistical patterns.

The Results

Zero Recidivism

Over 8 consecutive years, Xuhui District recorded zero recidivism among its community correction population monitored with CO-EYE GPS devices. This outcome reflects both the deterrent effect of continuous monitoring and the early intervention capability provided by the AI behavioral analytics — officers could intervene on risk indicators before they escalated to criminal behavior.

Why CO-EYE Was Retained

Deputy Chief Zhu of Xuhui’s correction authority offered a revealing explanation for why the district retained CO-EYE technology despite the availability of cheaper alternatives: “Our AI data analytic software automatically selected a tool that can match our system capabilities.”

The statement reflects a reality that procurement officials worldwide encounter: when an agency invests years of data collection and system integration around a specific device platform, switching to a cheaper device means losing the accumulated data and the AI models built on that data. The value isn’t just in the hardware — it’s in the years of behavioral data that the hardware collected and the analytical systems built on that data.

Key Takeaways for International Agencies

1. High-frequency data collection enables analytics. The 5-minute positioning interval wasn’t just about knowing where offenders were — it provided the data density needed for behavioral pattern recognition. Agencies considering AI-assisted monitoring should configure devices for higher data collection rates from the start, even if current workflows only use lower-frequency data.
2. Long-term data accumulation is an asset. Xuhui’s AI system required 10+ years of data. Agencies starting today are building the foundation for future analytical capabilities they may not yet envision. Choose data platforms and device vendors that support long-term data retention and export.
3. Human expertise trains the AI. The system succeeded because frontline officers — not just data scientists — shaped the behavioral classifications. Any AI deployment in corrections should involve domain experts in the training process.
4. Technology investment compounds. The initial GPS monitoring investment enabled the AI analytics investment, which produced outcomes (zero recidivism) that justified both. Agencies should evaluate monitoring technology not just on current use cases but on the platform’s ability to support future analytical capabilities.

Technology Used

• CO-EYE ONE — One-Piece GPS Ankle Monitor (5-minute continuous tracking, optical fiber anti-tamper)
• CO-EYE Monitoring Software (centralized tracking platform with data export capabilities)
• Community Corrections Electronic Monitoring Solutions

Source: CO-EYE Ankle Monitor: Better Tools, Better Job