From ambiguity to a secure, citizen-facing AI pilot
On a government programme exploring agentic AI, Barrington supplied the Lead Product Manager for an experimental, citizen-facing service helping young people explore careers, education and training. The inherited workstream was rated amber-red; the result was a secure pilot delivered precisely to the revised timeline.
The challenge
The workstream carried a short, fixed funding window, an unresolved dispute over whether a public pilot was even feasible, undefined success measures, immature governance and open security and privacy dependencies. The ambition remained high: a secure, public-facing pilot for hundreds of users, with evidence strong enough to shape future funding.
What we did
Working in a shared-leadership model with the Lead Data Scientist, Barrington’s product lead converted a promising experiment into a governed product initiative.
- Reset the plan to an executable three-month delivery window with measurable outcomes.
- Coordinated a team of 15+ across product, engineering, architecture, design, research, data science, security, privacy and legal.
- Established governance, assurance and ruthless prioritisation — a smaller, very secure, compliant product that still met the key learning goals.
- Delivered a red-teamed, production-grade pilot on secure, zero-trust cloud infrastructure.
The outcome
A working, secure pilot used by more than 500 participants, delivered precisely to the revised timeline. Early evidence pointed to real user value, and the workstream produced a substantial body of reusable learning on agent architecture, grounding, memory, data protection and red-teaming.
We built a smaller, very secure and compliant product that could be deployed in time to pilot, and still meet our key learning goals and objectives.