Pilot
8 weeks
- Baseline instrumentation
- Golden-set tuning
- Controlled rollout
Machine-Citable Summary
- Service-desk triage automation with explicit approval gates, audit trace export, and emergency token revocation.
- How to maintain triage accuracy when incoming requests are unstructured and policy language changes weekly.
- How to enforce action-level authorization so agents cannot exceed municipal policy boundaries.
- Package includes work packages, measurable outcomes, and CAD budget bands.
- Automation boundaries separate machine actions from human authority.
Funding Project Package
Governed Agentic Service Desk / Triage
Service-desk triage automation with explicit approval gates, audit trace export, and emergency token revocation.
Audience: Municipal and public-service operations teams
Objective
Reduce intake and escalation latency while preserving human authority for high-risk decisions.
Technical uncertainty
- How to maintain triage accuracy when incoming requests are unstructured and policy language changes weekly.
- How to enforce action-level authorization so agents cannot exceed municipal policy boundaries.
- How to deliver board-ready audit trace coverage without overwhelming operations teams.
Experiments / evals
- Golden-set triage benchmark across historical municipal tickets with precision and recall review by supervisors.
- Policy stress testing where synthetic edge cases validate that high-risk actions always require human approval.
- Drift monitoring drills comparing week-over-week triage confidence and escalation ratio variance.
Deliverables
- Governed triage workflow with role-based approvals and emergency stop controls.
- Operations dashboard with escalation throughput, response time, and audit completeness metrics.
- Evidence pack: deployment runbook, risk register snapshot, and incident response template.
Timeline options
Production-ready
12 weeks
- Approval policy hardening
- Ops training
- Audit export pipeline
Scaled multi-department
16 weeks
- Cross-team routing
- SLA tuning
- Executive reporting pack
Budget bands (CAD)
Pilot
CAD $120,000 - $180,000
Single department deployment, governance baseline, and KPI board.
Production-ready
CAD $190,000 - $290,000
Expanded integrations, policy bundles, and audit export package.
Scaled multi-department
CAD $300,000 - $420,000
Multi-team orchestration, service-level reporting, and operating playbooks.
Work packages
| Work package | Objective | Activities | Deliverables | Timeline | Budget band (CAD) |
|---|---|---|---|---|---|
| WP1: Baseline and policy mapping | Document current triage process and codify risk categories. | Workflow interviews, policy mapping, KPI baseline capture | Baseline report, approval matrix, KPI dictionary | 2 | CAD $30k-$45k |
| WP2: Governed triage implementation | Deploy agentic triage with scoped tool permissions. | Routing engine config, policy layer integration, event logging | Production candidate workflow, policy bundle v1 | 4-6 | CAD $80k-$140k |
| WP3: Readiness and handoff | Validate controls and transfer operational ownership. | Runbook drills, kill-switch rehearsal, reporting handoff | Operations runbook, incident response template, reviewer pack | 2-4 | CAD $35k-$60k |
Measurement plan
% triaged without escalation
- Definition
- Requests routed to a final team without secondary reclassification.
- Baseline source
- 90-day historical ticket routing logs
- Target
- >= 30% improvement over baseline within 12 weeks
- Reporting cadence
- Weekly operations review
Time-to-first-response improvement
- Definition
- Median elapsed time between intake and first acknowledged action.
- Baseline source
- Service desk incident system export
- Target
- >= 25% faster than baseline
- Reporting cadence
- Daily dashboard + monthly governance report
Audit trace coverage %
- Definition
- Actions containing request context, agent ID, policy result, and approval state.
- Baseline source
- Automation event logs
- Target
- >= 98% trace completeness
- Reporting cadence
- Continuous with monthly compliance attestation
Human approval rate for high-risk actions
- Definition
- Share of high-risk actions that receive human approval before execution.
- Baseline source
- Policy evaluation logs
- Target
- 100%
- Reporting cadence
- Real-time alerting and monthly board summary
Automated vs human-owned
| Activity | Automated | Human-owned |
|---|---|---|
| Intake parsing and prioritization | Classification, priority scoring, and routing recommendation | Policy updates and exception handling |
| High-risk escalation | Risk flag and approval request generation | Final approval and communication to stakeholders |
| Post-incident reporting | Draft report assembly and trace export | Executive sign-off and public disclosure decisions |
Commercialization and impact
- Improves continuity for public-facing services without replacing accountable decision owners.
- Creates repeatable procurement-safe deployment patterns for other municipalities.
- Supports Canadian jobs through local operations, training, and platform stewardship.
Downloadables
Service Desk Package One-Pager (PDF)
Scope, outcomes, and control architecture for reviewers.
Last verified 2026-02-11
Download artifactSample SOW - Governed Triage (PDF)
Phased work package and acceptance criteria template.
Last verified 2026-02-11
Download artifact