LI GARDNERResearch • Design • AI experience
AI + Public safety / Policing

Turning a GenAI demo into an agentic pilot.

GenAI could already synthesise location-based data. The harder question was what to do with it. In two weeks, I moved the work from a broad AI ambition to a defined human–AI pilot: one operational problem, one bounded workflow, clear evidence measures.

Illustrative scene: a police officer studies a hotspot on a holographic city map beside a SARA workflow panel showing AI and human roles
2 weeksFrom broad ambition to a defined pilot proposition
4Party ownership and partnership for one thin-slice pilot to scale
Agentic MVPDefined, with a proposal for scaling
My role
Engagement Lead · Service Designer
Type
Consulting
Duration
2 weeks (2025)
Partners
UK regional police · AWS · University evidence-based policing research partner
Focus
AI-assisted crime prevention
AI
GenAI · Agentic pilot · Human–AI workflow · AI evaluation & assurance

What I did, in four moves

  1. 01 ChallengeSituation

    The technology was not the blocker. The target was.

    GenAI’s potential was proven, but there was no clear next move. The challenge: focus on one operational problem, show real value and build a case for investment.

    Reframed “how could we use GenAI?” into one testable proposition.

  2. 02 ApproachTask

    Make the whole decision flow visible.

    I brought officers, analysts and technical teams together to map the decision flow end to end, from data and context to briefing and action.

    Clear lines on where AI helps and where human judgement is essential.

  3. 03 ScopeAction

    Reduce the scope without breaking the value chain.

    • One location, one crime type
    • A minimum viable dataset
    • A realistic operational workflow
    • Governance kept intact

    A thin-slice pilot that tests the human–AI system in practice and generates evidence quickly.

  4. 04 OutcomeResult

    From framing to something teams could build and test.

    A defined pilot proposition: human–AI workflow, evaluation criteria, technical scope and a clearer view of value and next steps.

    In two weeks: a proposal ready for further development, with potential to extend and scale.

From question to action: the human–AI workflow

The workflow I shaped

  1. Data & contextRelevant data sources and local insight
  2. AI analysisRetrieve, analyse and surface patterns
  3. Human judgementValidate, add context and decide
  4. Operational actionTargeted patrols and interventions
  5. OutcomesMonitor → improve → extend → scale
AI: retrieve, analyse, draftHuman: validate, decide, act
My contribution

What I owned

  • Crafted the partnershipFor a well-sponsored programme.
  • Framed the opportunityTo deliver value earlier.
  • Defined the MVPIn two weeks, ready for the next phase.
  • Designed the evidence modelReady for design and test.
  • Shaped the proposal and business caseTo support the long-term strategy.
Methods, frameworks & toolkit

How I worked

  • Executive leadership workshop
  • Stakeholder & RACI mapping
  • Assumption & risk mapping
  • Value hypothesis framing
  • SARA workflow modelling
  • Service blueprint / swimlane mapping
  • Thin-slice MVP definition
  • AI evaluation & assurance framework