Agentic AI pilot delivered to production standards — on time
On time
Fixed deadline met
Production
Grade assurance
Red-teamed
Safety tested
“Can agentic AI create genuine value for citizens in a complex public service — and can it be delivered safely, at scale, inside government?”
The problem
A fixed financial-year deadline, with a plan that wouldn't hit it. No clear objectives, key results, or success criteria — and no template for governing an agentic pilot inside the department.
The work sat between discovery, alpha, and beta — but needed production-grade assurance because real citizens were going to use it.
What we built
- Re-planned the pilot on inheritance — structural changes, SRO sign-off, and a revised timeline delivered against precisely
- Set objectives, key results, and success criteria against the business case
- Led information security assurance end-to-end, identifying controls needed for public-facing launch
- Owned the data protection impact assessment, privacy policy, and user terms
- Participated in red-team exercises on the agent, contributing to model-safety evidence
- Configured survey tooling so user feedback flowed into the evaluation pipeline
Outcome
Delivered on time, to production-grade assurance, inside a fixed financial-year deadline — the kind of outcome most agentic pilots in government don't reach.
Generated credible, citizen-tested evidence to inform the business case and the department's next funding decision.
Recognise the problem?
Tell us the workflow or decision that's costing you the most.