United KingdomPublic sector2025 – ongoingDepartment for Work and Pensions (DWP)
Universal Credit urgent journal message detection
DWP flagged possible urgent messages with 84.0% recall and 13.5% precision
Problem
Universal Credit agents manually identified messages indicating possible harm. The previous flagging process took over 24 hours.
Approach
DWP deployed a logistic regression model to flag potentially urgent journal messages. Agents saw red flags in their dashboards and decided how to respond. All other messages remained subject to normal review, and staff could correct flags.
Results
Takeaway
Low precision meant most flagged messages did not meet the urgency classification. Agents still reviewed unflagged messages because the model missed some urgent contact.
Categories
- Size
- Large enterprise
- Industry
- Public sector
- Business function
- Customer service
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Built in-house
- Goal
- Quality and risk
- Evidence
- Company's own claim
Sources
Source summaries
DWP documented daily production use of a machine learning model for message prioritization. It reported 13.5% precision and 84.0% recall for April 2025 to April 2026, with agents retaining control of responses.
Full details are in the original sources.