United KingdomPublic sector2016–2023Avon and Somerset Police and Bristol City Council
Child exploitation risk-scoring models
Bristol stopped using child exploitation risk models after staff questioned their accuracy
Problem
Child protection staff held separate records that could conceal signs of exploitation when considered individually. Police and council staff sought to identify vulnerable children earlier.
Approach
Police developed machine-learning risk scores using police and public-service records linked through the Think Family Database. After council data was removed, staff reported inaccurate rankings and missed vulnerable children. Reviewers could not locate the source code and variables needed to test the models. Both exploitation models had stopped being used by June 2023.
Results
No published figures
Takeaway
Removing council data changed the models' usefulness. Missing development records prevented reviewers from testing their performance.
Categories
- Size
- Large enterprise
- Industry
- Public sector
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Built in-house
- Goal
- Quality and risk
- Evidence
- Press reporting
Sources
Source summaries
An investigation with WIRED examined records and interviews about police predictive models. It reported staff concerns about exploitation scores after council data was removed, missing development records and the models' withdrawal by June 2023.
Full details are in the original sources.