5. 90-Day Launch Plan

Standing up an AI Centre of Excellence doesn’t happen overnight, but it doesn’t require years of planning either. This 90-day launch plan balances quick wins with foundational work, ensuring your CoE starts delivering value while building sustainable structures.

Phase 1: Foundations (Days 0–30)

Goal: Establish leadership, baseline current state, and define scope.

Week 1: Appoint Leadership and Secure Sponsorship

Actions:

• ✅ Appoint Head of AI CoE and Executive Sponsor(s)
• ✅ Define reporting structure (CIO, CDO, CTO, or Chief AI Officer)
• ✅ Secure initial budget and headcount approvals
• ✅ Communicate the CoE vision to the organisation

Outputs:

• Executive mandate and charter
• Initial team roster (even if not yet hired)
• Communication plan

Key risks:

• Lack of executive buy-in → Escalate early; show quick wins
• Unclear reporting line → Define before announcing CoE

Week 2–3: Baseline AI Readiness

Actions:

• ✅ Conduct AI Readiness Assessment across 15 dimensions
• ✅ Inventory existing AI initiatives (shadow IT, pilots, production systems)
• ✅ Identify gaps in governance, risk, and technology
• ✅ Interview stakeholders (IT, Risk, Legal, Business) to understand pain points

Outputs:

• AI maturity heatmap (where you are today)
• Inventory of existing AI projects
• Prioritised gap analysis

Key risks:

• Incomplete inventory → Engage Finance and Procurement to track AI spend
• Resistance to assessment → Frame as “improvement, not blame”

Week 3–4: Define Non-Negotiables and Initial Standards

Actions:

• ✅ Draft initial non-negotiables (identity, secrets, logging, data contracts)
• ✅ Define security baselines (threat modelling, least privilege, encryption)
• ✅ Establish privacy-by-design principles (DPIA, data minimisation)
• ✅ Create intake and prioritisation rubric (ROI, strategic fit, feasibility, risk)

Outputs:

• Non-negotiables document (v0.1)
• Intake form template
• Prioritisation rubric

Key risks:

• Standards too restrictive → Start lean; evolve based on feedback
• Ignored by delivery teams → Embed standards in golden paths

Week 4: Stand Up MVP Platform

Actions:

• ✅ Provision cloud environment (AWS, Azure, GCP)
• ✅ Deploy foundational tools: experiment tracking (MLflow), model registry, CI/CD
• ✅ Set up identity and access management (IAM, SSO)
• ✅ Implement logging and secrets management

Outputs:

• MVP AI platform (basic but functional)
• Access provisioning process
• Platform documentation (wiki or Confluence)

Key risks:

• Over-engineering the platform → Start with essentials; add features incrementally
• Vendor lock-in → Use open-source where possible; negotiate exit clauses

Week 4: Pick 2–3 High-Value Pilots

Actions:

• ✅ Review existing AI use cases and select 2–3 for pilot acceleration
• ✅ Criteria: High business value, manageable risk, sponsorship, data availability
• ✅ Assign Product Owners and delivery teams
• ✅ Define success metrics (business outcomes, not just technical metrics)

Outputs:

• Pilot project briefs
• Success criteria and timelines
• Resource allocations

Key risks:

• Picking too complex pilots → Start with medium complexity; demonstrate value
• Lack of executive sponsorship → Ensure each pilot has C-level backing

Phase 2: Enablement and Governance (Days 31–60)

Goal: Build golden paths, implement governance cadence, and run pilots.

Week 5–6: Build Golden Paths

Actions:

• ✅ Document reference architectures for common patterns (batch inference, real-time scoring, LLM applications)
• ✅ Create templates: data pipeline, model training, deployment, monitoring
• ✅ Implement evaluation gates (accuracy, bias, robustness thresholds)
• ✅ Publish golden paths to internal wiki or platform docs

Outputs:

• 3–5 golden paths (documented and tested)
• Template library
• Evaluation framework

Key risks:

• Golden paths ignored → Demonstrate value via pilot projects
• Documentation out of sync → Automate doc generation where possible

Week 6–7: Establish Governance Cadence

Actions:

• ✅ Schedule recurring meetings: Executive Steering, Portfolio Council, Architecture Review, MLOps CAB
• ✅ Create agenda and minutes templates
• ✅ Define RACI for key decisions
• ✅ Launch RAID log (Risks, Assumptions, Issues, Dependencies)

Outputs:

• Meeting calendar for next 6 months
• Decision log and action tracker
• RACI matrix

Key risks:

• Meeting overload → Keep meetings focused and time-boxed
• Low attendance → Ensure value is demonstrated; rotate participants

Week 7–8: Run Proof-of-Value (PoV) on Pilots

Actions:

• ✅ Pilot teams adopt golden paths for their solutions
• ✅ Conduct architecture and model risk reviews
• ✅ Gather feedback on golden paths and platform
• ✅ Track time-to-value and adoption metrics

Outputs:

• PoV results (technical feasibility, business value)
• Feedback for golden path improvements
• Lessons learned

Key risks:

• Pilot failures → Treat as learning; document what didn’t work
• Insufficient business engagement → Product Owners must drive adoption

Week 8: Launch Community of Practice

Actions:

• ✅ Schedule first Community of Practice (CoP) meeting
• ✅ Invite all AI practitioners (engineers, scientists, product owners)
• ✅ Agenda: Intro to CoE, demo of golden paths, Q&A
• ✅ Create Slack/Teams channel for async collaboration

Outputs:

• First CoP meeting (recorded)
• Community engagement plan
• Collaboration channel

Key risks:

• Low engagement → Showcase quick wins and celebrate contributors
• One-way communication → Make it interactive; solicit feedback

Phase 3: Production-Ready and Measurement (Days 61–90)

Goal: Launch first production solutions, implement monitoring, and start tracking benefits.

Week 9–10: Production-Ready 1–2 Pilots

Actions:

• ✅ Pilot teams complete production readiness checklist
• ✅ Conduct final architecture, security, privacy, and model risk reviews
• ✅ Deploy to production with monitoring and rollback plans
• ✅ Communicate launches internally

Outputs:

• 1–2 AI solutions live in production
• Monitoring dashboards and runbooks
• Case studies for future reference

Key risks:

• Rushed deployments → Enforce stage-gate discipline
• Lack of monitoring → No production deployment without observability

Week 11: Implement Monitoring and Playbooks

Actions:

• ✅ Deploy drift detection and alerting for production models
• ✅ Create runbooks for common incidents (model degradation, API outages, data quality)
• ✅ Establish on-call rotation and incident response process
• ✅ Test rollback procedures

Outputs:

• Monitoring dashboards (uptime, latency, drift)
• Runbooks and playbooks
• Tested incident response

Key risks:

• Alert fatigue → Tune thresholds; avoid noisy alerts
• Unclear escalation → Document on-call procedures and contacts

Week 12: Launch Benefits Ledger (v1)

Actions:

• ✅ Define value tracking framework (ROI, cost savings, time saved, revenue uplift)
• ✅ Establish baseline metrics before AI implementation
• ✅ Track actual outcomes vs. predicted benefits
• ✅ Create portfolio dashboard for Executive Steering

Outputs:

• Benefits ledger (spreadsheet or tool)
• Portfolio dashboard (value, risk, adoption)
• Quarterly business review template

Key risks:

• Benefits overstated → Use conservative assumptions; validate with Finance
• Lack of baseline → Capture “before” metrics for all future initiatives

Week 12: Retrospective and Roadmap Update

Actions:

• ✅ Conduct CoE retrospective: What went well? What didn’t?
• ✅ Gather feedback from stakeholders (delivery teams, executives, risk/compliance)
• ✅ Update golden paths, standards, and governance cadence based on lessons learned
• ✅ Draft 6-month roadmap: next use cases, platform enhancements, training plans

Outputs:

• Retrospective report
• Updated standards and golden paths
• 6-month roadmap

Key risks:

• Ignoring feedback → Act on lessons learned; communicate changes
• Roadmap overcommitment → Be realistic about capacity

90-Day Checklist

Post-90-Day: Sustaining Momentum

Once the CoE is stood up, focus shifts to:

✓ Scaling adoption — Onboard more use cases and teams ✓ Maturing the platform — Add advanced features (AutoML, feature stores, advanced monitoring) ✓ Building talent — Hire, train, and certify AI practitioners ✓ Measuring value — Quarterly business reviews with executives ✓ Evolving standards — Regular updates to golden paths and non-negotiables

Next Steps

With the CoE launched, the final step is understanding success metrics and avoiding anti-patterns that can derail even well-intentioned initiatives.