LI GARDNERResearch • Design • AI experience
AI practice / Public sector / Agentic workflow

Applying AI-first service design to complex public sector work.

I designed a Microsoft Copilot-based system with specialised agents and reusable skills for the end-to-end service-design process, keeping human judgement at critical decision points.

A team works through a service design workshop while an agentic workflow of Strategy, Stakeholders, Research, Design and Implementation plan is overlaid
>50%Less research and synthesis time (user-reported)
80Stakeholders mapped
4Government departments
My role
Designer & Engagement Lead
Type
AI Practice + Consulting
Duration
11 weeks (2026)
Focus
Service design · Research · Government · Agentic AI
AI
Microsoft Copilot · Multi-agent workflow · Human-in-the-loop
Research
24 people interviewed · 18 sessions · 17 survey responses

What I did, in four moves

  1. 01 ChallengeSituation

    Large-scale discovery was rigorous, but slow.

    Cross-government discovery was manual, with fragmented evidence, multiple teams and difficult stakeholder access, especially over the holiday period.

    17 teams across 4 government departments to engage.

  2. 02 DefineTask

    Decompose the work into agents and skills.

    I decomposed the engagement into five connected stages and defined specialised agents with reusable skills for each stage.

    Strategy, Stakeholder, Research, Design and Report agents.

  3. 03 MakeAction

    Design the agentic workflow in Microsoft Copilot.

    I connected the agents through a shared knowledge layer, with clear human checkpoints at every stage.

    AI supports structuring, synthesis and draft artefacts; judgement stays human.

  4. 04 TransformResult

    More time for judgement, less for processing.

    The agents structured evidence, synthesised research and prepared design artefacts, while I led stakeholder engagement, interpretation and final recommendations.

    Research time more than halved across a larger stakeholder network (user-reported).

Agentic workflow architecture: five specialised agents, human judgement at every stage

The agentic workflow I designed

InputProject briefContext, goals, constraints
Orchestrator agentManages project context, delegates tasks, coordinates agents, ensures hand-offs and quality
OutputDecision-ready outputsReports · Recommendations · Service design artefacts
1Project strategyDefine direction, scope and research approach
Strategy agentInterprets the brief and plans the engagement
Skills
  • Document analysisAnalyse background documents
  • Brief decompositionExtract objectives, scope and assumptions
  • Kick-off workshop planningGenerate agenda and materials
  • Project plan generationCreate timeline, activities and resourcing
  • Research strategyDefine methods and success measures
  • Risk & dependency mappingIdentify risks and dependencies
Human role
  • Frame the real problem
  • Challenge assumptions
  • Set scope and ambition
  • Approve project strategy
2Stakeholder mappingIdentify, engage and manage stakeholders
Stakeholder agentMaps, analyses and coordinates stakeholder engagement
Skills
  • Stakeholder identificationExtract and enrich stakeholder list
  • Stakeholder mappingMap influence, interests and relationships
  • Stakeholder communicationDraft tailored communications
  • Interview bookingCoordinate calendars and send invites
  • Topic guide generationCreate role-specific interview guides
  • Engagement planPrioritise and track engagement
  • Coverage analysisMonitor representation and gaps
Human role
  • Read organisational dynamics
  • Build trust and access
  • Prioritise key stakeholders
  • Approve engagement approach
3User researchCapture, analyse and synthesise insights
Research agentProcesses research evidence and extracts insights
Skills
  • Topic guide reviewRefine and adapt by stakeholder type
  • Interview note processingTranscribe and structure notes
  • Evidence codingCode data and tag key themes
  • Theme clusteringIdentify patterns across interviews
  • Cross-research consolidationCombine interviews, survey and document insights
  • Contradiction & gap detectionIdentify divergent views and evidence gaps
  • Evidence traceabilityLink insights to source data
Human role
  • Conduct and adapt research
  • Probe nuance and context
  • Judge evidence quality
  • Challenge weak conclusions
4Future-state designIdeate, evaluate and define the future state
Design agentTurns insights into opportunities and service concepts
Skills
  • Current-state synthesisSummarise key issues and opportunities
  • Ideation workshop planningGenerate agenda, prompts and materials
  • Idea clusteringGroup and synthesise ideas
  • Idea evaluationAssess against agreed criteria (e.g. value, feasibility, impact)
  • Future-state journeyGenerate customer / user journeys
  • Service blueprint draftingCreate end-to-end service blueprint
  • Requirements analysisExtract, analyse and prioritise requirements
Human role
  • Facilitate co-design and workshops
  • Make trade-offs visible
  • Select and prioritise direction
  • Resolve conflicting views
5Report generationCreate evidence-based recommendations
Report agentAssembles findings and generates decision-ready outputs
Skills
  • Report architectureCreate structure and chapter outlines
  • Evidence-to-finding mappingLink findings to source evidence
  • Recommendation generationDevelop options and recommendations
  • Narrative structuringWrite and refine content
  • Chart & table generationCreate visualisations and data summaries
  • Citation checkingValidate sources and ensure traceability
  • Quality reviewCheck consistency, completeness and formatting
  • Executive summaryGenerate decision-ready summary
Human role
  • Challenge the argument
  • Own the recommendation
  • Tailor message for decision-makers
  • Approve final report
Shared project knowledge & evidence layerCentralised, structured and traceable project knowledge
  • Client documents
  • Research data (interviews, surveys)
  • Project artefacts (plans, outputs)
  • Templates & methods
  • Previous research
  • Real-time context
Tools & integrationsLeverage Microsoft 365 and other tools
  • Microsoft Copilot
  • SharePoint / OneDrive
  • Teams / Outlook
  • Forms / Surveys
  • Calendar
  • Other systems (e.g. CRM)
Governance & controlsEnsure security, quality and responsible AI
  • Permissions & access control
  • Source citation & traceability
  • Versioning & audit trail
  • Human approval gateways
  • Responsible AI & data privacy
Simplified from the design specification. Time saving is user-reported against a similar earlier project, not a controlled benchmark.
My contribution

What I owned

  • Designed the end-to-end agentic workflow
  • Defined specialised agents and reusable skills
  • Designed shared knowledge and hand-off structures
  • Built human checkpoints into every stage
  • Led cross-government stakeholder engagement
  • Connected research evidence to future-state design and reporting
Methods, frameworks & toolkit

How I worked

  • Agentic workflow design
  • Microsoft Copilot
  • Stakeholder mapping
  • User research
  • Evidence synthesis
  • Future-state design
  • Human-in-the-loop
  • Evidence-based reporting