Full Report
As UK police embrace the AI revolution, a WIRED investigation reveals the messy inside story of one region’s experiment with predictive analytics.
Analysis Summary
# Industry News: UK Predictive Policing Under Scrutiny Following Failed Regional Trial
## Summary
A multi-year investigation into the Avon and Somerset Police’s use of predictive analytics reveals significant failures in model accuracy, data transparency, and ethical oversight. Despite high-profile ambitions to utilize machine learning for risk-scoring citizens, several key models were quietly abandoned after officials deemed the results untrustworthy and potentially biased.
## Key Details
- **Date:** Investigation published June 25, 2026
- **Companies Involved:** Avon and Somerset Police, Bristol City Council, College of Policing (UK)
- **Category:** Field Analysis / Predictive Analytics Implementation
## The Story
Between 2016 and 2024, the Avon and Somerset Police, in partnership with the Bristol City Council, developed the "Think Family Database" and an "Offender Management App." These systems integrated sensitive data—including mental health records, school meal status, and police intelligence—to assign risk scores to nearly 500,000 citizens. The goal was to predict domestic abuse, burglary, and child exploitation.
However, a joint investigation by WIRED and Liberty Investigates uncovered that these models suffered from "genuinely poor predictive performance." Internal documents suggest that at least two models were scrapped because their outputs were unreliable. Furthermore, the lack of transparency meant that individuals were "scored" by algorithms without their knowledge or any path for recourse, leading to potential legal challenges regarding data privacy and human rights.
## Business Impact
### For the Companies Involved
- **Public Sector Reputational Damage:** The Avon and Somerset Police and Bristol City Council face significant scrutiny over "black box" decision-making and the potential for algorithmic bias.
- **Liability Risks:** The involvement of groups like Liberty suggests impending litigation regarding the Processing of Data under GDPR and the UK Data Protection Act.
### For Competitors
- **Opportunity for "Ethical AI" Providers:** Vendors focusing on explainable AI (XAI) and rigorous bias-testing frameworks will likely gain a competitive edge over companies selling opaque "black box" predictive models.
- **Stricter Procurement Standards:** Competitors in the GovTech space must now prepare for more intensive audits and validation requirements from public sector clients.
### For Customers
- **Civic Trust Erosion:** Citizens (the "end users" of public safety services) face a loss of trust in law enforcement, which can lead to decreased cooperation in community policing efforts.
- **Due Process Concerns:** Individuals may be unfairly targeted for police intervention based on flawed data inputs.
### For the Market
- **Market Deceleration:** The failure of this high-profile project may slow the adoption of predictive analytics in other UK regions as police forces grow wary of legal and public backlash.
- **Shift Toward National Standards:** The move by the College of Policing to set national standards suggests the market will shift from "wild west" regional experiments to a more regulated, centralized procurement model.
## Technical Implications
The failure highlights the "Big Bucket" fallacy: the assumption that aggregating disparate datasets (housing, health, police records) and applying machine learning will inherently yield actionable insights. Problems identified include:
- **Data Quality Issues:** Sensitive data from social services often lack the structured consistency required for accurate ML training.
- **Feedback Loops:** Using historical crime data to predict future crime often reinforces existing biases in policing.
## Strategic Analysis
- **Market Positioning:** The UK is attempting to position itself as a leader in AI governance; however, these revelations suggest a gap between high-level policy and local implementation.
- **Competitive Advantage:** Agencies that prioritize transparency and "human-in-the-loop" systems will be better positioned to survive the upcoming regulatory tightening.
- **Challenges:** The primary obstacle remains the "Data-Science Spatula" approach—oversimplifying complex human behaviors into binary risk scores.
## Industry Reactions
- **Analyst Opinions:** Independent data analysts described the performance scores as "startlingly lacklustre."
- **Expert Commentary:** Privacy advocates cite this as a primary example of why predictive policing should be strictly regulated or banned under human rights frameworks.
- **Market Response:** There is a growing demand for independent third-party auditing of algorithms used in the criminal justice system.
## Future Outlook
- **Predictions:** We expect a wave of "Right to Explanation" requests from UK citizens attempting to find their scores in similar databases.
- **What to Watch For:** New guidelines from the UK College of Policing that may restrict the types of data (e.g., school meals, mental health) allowed for use in crime-prediction models.
## For Security Professionals
Cybersecurity and data privacy practitioners should take note of the **data minimization** and **transparency** failures here. This case serves as a warning for those building internal risk-scoring models (e.g., Insider Threat detection): if the underlying data is flawed or the model's logic is opaque, the resulting "risk scores" can lead to discriminatory outcomes and significant legal liability. Always ensure a clear path for data subject access requests (DSARs) and regular model validation.