National Water Safety Agency
From complex reports to clean water decisions
Project type: MVP
Team: 11
Sprint time: 4 weeks
Project summary
Problem
Every year, the agency receives between 500 and 600 incident reports from water companies.
These reports can be up to 60 pages long and must be carefully reviewed and classified by a small team of inspectors.
They need to determine the severity of each incident and whether intervention is required – often against tight time pressures.
Manual classification was slow, repetitive, and resource-heavy.
It also delayed response times for high-severity events.
Process
I facilitated workshops with inspectors to understand how they reviewed reports and where delays happened.
Together, we mapped the current process, defined what data mattered most, and shaped a future-state flow using AI to classify and highlight key details.
I worked with AI engineers and the client to prioritise what to test first, then designed and refined the tool based on real feedback.
Early testing showed that inspectors found the summaries useful, the report highlighting intuitive, and the classifications "more accurate than expected."
Solution
A working AI-powered tool that reads, classifies, and explains incident reports in minutes, not hours – highlighting critical risks and surfacing supporting evidence for each classification.
Impact
Cut report review time from 1.5 hours to 30 minutes — a ~66% reduction
Improved classification accuracy from 90% to 100% — an ~11% improvement
Reduced delays in intervention decisions
Validated in early testing by inspectors as “very promising” and “exactly what we want”
My role
I co-led discovery workshops, defined the user journey, and designed the user interface.
I worked closely with AI engineers to shape how model outputs were surfaced to inspectors — from summaries and reasoning trails to risk categories and visual clarity.
Ran design thinking workshops with inspectors to understand how reports were received, reviewed, and classified
Documented the key data they needed to make decisions and spotted where delays happened
Co-created a “to-be” scenario with the client,
their ideal future processExplored how generative AI could help classify and highlight important parts of each report
Prioritised a solution to test with the client and AI engineers
Designed and refined the tool through several rounds of feedback