gtmskills.com

Reference file

Ai rollout 5 phase playbook

ai-rollout-5-phase-playbook.md

AI Rollout: Five-Phase Playbook

On-demand reference for the revops-change-management skill.

This is your operational blueprint. Use it to sequence your AI adoption.

Phase structure: Discover → Pilot → Scale → Embed → Govern

Each phase has gates (go/no-go decisions), specific activities, and success metrics.

Phase 1: Discover (Weeks 1 to 3)

Goal: Identify which processes, decisions, and teams benefit most from AI. Build business case.

Activities:

  • Audit current processes: where's manual work, where do humans add low value, where's error-prone work?
  • Prioritize by impact and ease (impact/effort matrix)
  • Identify high-risk processes first (compliance-sensitive, data-heavy, error-prone)
  • Build financial case: time savings, quality improvements, cost avoidance
  • Check regulatory landscape (Works council notification if EU-based)

Duration: 2 to 3 weeks

Success metrics:

  • 3 to 5 priority processes identified
  • Business case quantified (conservative estimate OK here)
  • Regulatory blockers identified (early)
  • Executive alignment on first pilot

Gate: Proceed to Pilot when executive sponsor approves business case and pilot scope.

Phase 2: Pilot (Weeks 4 to 8)

Goal: Test one AI application in a controlled environment with a small, engaged team. Prove concept. Learn what doesn't work.

Two-week sprint structure (repeat as needed within pilot phase):

SPRINT TEMPLATE: Scope → Baseline → Build → Run → Evaluate → Document

Week 1: Scope & Baseline
Day 1-2: Scope
  - Define the specific decision/task: "Qualification scoring for inbound leads"
  - Define AI tool: vendor, model, integration point
  - Define success: "AI scores match human qualification 80%+ of the time"
  - Define constraints: data available, team capacity, regulatory requirements

Day 3-4: Baseline
  - Measure current state: How long does qualification take? Accuracy? Who decides?
  - Document decision logic: What criteria does a human use to qualify?
  - Establish comparison: Can we run manual and AI in parallel for a week?

Week 2: Build & Run
Day 5-7: Build
  - Configure tool: connect data, set parameters, define rules
  - QA: test on historical data, validate outputs match expectations
  - Safety check: confirm no data leakage, no PII exposure, no bias amplification

Day 8-10: Run
  - Live pilot: 100 to 200 records through AI + human decision
  - Log: inputs, AI output, human decision, time taken
  - Weekly standups: team feedback, quick fixes to parameters
  - Incident log: any errors, any weird outputs, any governance issues

Week 3: Evaluate & Document
Day 11-12: Evaluate
  - Compare: AI vs human accuracy, speed, consistency
  - Sentiment: team feedback, adoption friction, confidence in tool
  - Financial: hours saved, quality improvements, cost-per-decision

Day 13-14: Document
  - Pilot report: what worked, what didn't, learnings
  - Process map: AI's new place in the workflow
  - Team playbook: how to use tool for next phase
  - Recommendation: scale, refine, or kill

Pilot metrics (measure all):

  • Accuracy: AI decision vs. ground truth (human expert review). Target: 75%+ agreement.
  • Consistency: Same input yields same output. Target: 95%+ (identify edge cases).
  • Speed: Time per decision (AI vs. human). Target: 60%+ faster.
  • Adoption: Team using tool without prompting. Target: 80%+ of available opportunities used.
  • Confidence: Team trusts AI outputs (survey 1 to 5). Target: 3.5+.

Kill criteria (when to stop an initiative):

  • Accuracy <65% (tool not reliable enough)
  • Adoption <40% after 2 weeks (team doesn't want it)
  • Data quality issues discovered (garbage in, garbage out)
  • Governance blockers identified (regulatory issue, data sensitivity)
  • Cost-per-use higher than manual process
  • Team confidence score <2.5 (distrust too high)

Gate: Proceed to Scale when:

  • Accuracy ≥75% OR team confident in outputs + clear remediation plan
  • Adoption ≥60%
  • No unresolved governance issues
  • Financial case holds (ROI neutral or positive within 6 months)

Phase 3: Scale (Weeks 9 to 16)

Goal: Expand from one team to 3 to 4 teams. Roll out with clear playbooks. Build confidence.

Activities:

  • Expand to adjacent teams (similar roles, similar processes)
  • Recruit change champions from pilot team (they become trainers)
  • Weekly operational cadence: usage standup, issue triage, quick wins celebration
  • Build skill: structured training, hands-on practice, certification (optional but effective)
  • Gather feedback: monthly pulse survey, open suggestion channel

Duration: 6 to 8 weeks

Success metrics:

  • 3 to 4 teams active, 60%+ usage rate
  • Accuracy sustained (75%+) or improving
  • Incident rate <1 per 100 active users
  • Team confidence increasing (sentiment survey)
  • Time savings realized and quantified

Gate: Proceed to Embed when:

  • 60%+ of expanded user base actively using tool
  • Business case metrics met (time saved, quality improved)
  • No major governance incidents
  • Process standardization documented

Phase 4: Embed (Weeks 17 to 26)

Goal: Make AI usage standard operating procedure. Integrate into workflows, performance metrics, hiring.

Activities:

  • Integrate AI into formal workflows: job descriptions updated, training mandatory, metrics tracked
  • Add to onboarding: new hires trained on AI tools day 1
  • Performance management: AI productivity (time saved, quality) becomes KPI
  • Feedback loops: monthly review of outputs, retraining as needed
  • Build organizational muscle memory: "this is how we work now"

Duration: 8 to 10 weeks

Success metrics:

  • 80%+ usage rate sustained
  • New hires up to speed within 2 weeks
  • Accuracy stable or improving
  • Productivity metrics show sustained gains
  • Shadow AI usage declining (people using approved tools)

Gate: Move to Govern when:

  • AI-enabled workflow is default, not optional
  • Team competency normalized
  • Business value clearly realized
  • Regulatory compliance proven over time

Phase 5: Govern (Ongoing, starting week 26 and beyond)

Goal: Ensure sustained use, manage risk, optimize continuously.

Ongoing activities:

  • Monthly metrics review: usage, accuracy, incidents, cost-per-use
  • Quarterly business review: is AI still delivering ROI? Should we expand?
  • Continuous retraining: skill decay monitoring, refresher cadence
  • Incident management: protocol for errors, data breaches, user issues
  • Tool refresh: evaluate new models, update parameters, retire if outdated
  • Works council reporting (EU): quarterly updates on usage, incidents, any changes

Metrics (track forever):

  • Active user rate, engagement rate, output acceptance rate
  • Incident rate, data governance compliance, audit pass rate
  • Time saved per user, quality improvements (reduced rework)
  • ROI (realized + trending + capability)

Weekly Cadence During Implementation

Once you're in Pilot, Scale, or Embed, this is your rhythm:

WEEKLY STANDUP (30 min, Tuesdays 9am)
Attendees: Project lead, tool owner, team representatives, change champion

Agenda:
1. Usage: Are people using it? Adoption rate vs. target?
2. Issues: Any errors, data problems, user confusion?
3. Quick wins: What's working well? Celebrate it.
4. Blockers: What's slowing adoption? What needs fixing?
5. Forecast: What's coming next week?

Decision rights:
- Quick fix (parameter tweak, training gap): PM decides, implements by Thursday
- Larger issue (tool limitation, design change): escalate to steering, decide within 1 week
- Kill decision: steering committee vote (Weeks 1 to 4 of pilot)

OUTPUT:
- 1-pager: adoption %, issues, next week's focus
- Shared with sponsors, works council (EU), exec team

This framework takes 6 to 9 months from Discover to full Embed. That's realistic. GenAI adoption is not a 6-week sprint. It's a quarterly narrative.

Your biggest risk isn't the technology. It's abandoning the change process too early when adoption looks slow (weeks 4 to 6 is always slow), or trying to move faster than your organization can absorb.

Move at the speed of trust-building, not the speed of the technology.