United KingdomPublic sector2025 – ongoingHM Prison and Probation Service
Automated quality scoring of Key Work session notes
HMPPS used machine learning to score around 80,000 prison case notes monthly
Problem
Manual reviews covered only a sample of Key Work session notes. The sample supported annual prison comparisons but could not reliably track monthly changes.
Approach
The Ministry of Justice built a pipeline that removed names with BERT, represented text with Longformer and predicted quality with gradient boosting. Data scientists ran it monthly and uploaded scores to the HMPPS Performance Hub. Prison managers used the scores for quality reviews, with human reassessment available when prisons challenged a score.
Results
Takeaway
The system expanded assessment beyond sampled notes. Its scores measured documented session quality, rather than rehabilitation outcomes.
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
- Evidence
- Company's own claim
Sources
Source summaries
HMPPS described automated scoring of around 80,000 notes each month. It explained the text-processing models and how prisons could request human reassessment.
Full details are in the original sources.