level-1-essentials/company/10-rollout-playbook/README.md

10 — Rollout Playbook

How to go from Day 0 to a company-wide AI program in 90 days, then continue.

Files

• 30-60-90-plan.md — the day-by-day rollout plan
• pilot-selection-criteria.md — how to pick first workflows and departments
• expansion-decision-framework.md — when and how to expand from pilot → dept → company

The three phases

• Days 1-30 — Foundation: governance, accounts, prompts, training basics
• Days 31-60 — First workflows: pilot with 2 departments, launch first agents
• Days 61-90 — Expand and measure: roll to more departments, measure ROI

Success criteria at Day 90

• 60%+ of company using AI weekly
• 10+ production workflows
• Cost within budget
• 3+ measurable ROI examples
• Employee NPS ≥ 40

level-1-essentials/company/10-rollout-playbook/30-60-90-plan.md

30-60-90 Rollout Plan

The company's rollout playbook. Assumes you're starting from ~0 AI maturity.

Day 0 — Prerequisites

Before Day 1: - Executive sponsor identified and committed - Steering Committee members named (5-7 people) - Initial budget approved (see budget-and-billing-model.md) - Legal + Security + IT engaged - Communications plan drafted

Days 1–30 — Foundation

Governance

• [ ] AUP drafted and reviewed (company/01-governance/acceptable-use-policy.md)
• [ ] Data classification matrix approved (company/01-governance/data-classification-matrix.md)
• [ ] Vendor DPA checklist ready
• [ ] Incident response addendum in place
• [ ] Steering Committee charter signed

Accounts & procurement

• [ ] ChatGPT Enterprise procured (or equivalent enterprise chat)
• [ ] Claude Team / Enterprise procured
• [ ] Perplexity Enterprise procured
• [ ] Copilot / Cursor site license procured
• [ ] LLM gateway selected (Portkey / LiteLLM / OpenRouter)
• [ ] Vector DB selected (pgvector for start, or Pinecone if scale is high)
• [ ] SSO integration confirmed for each

Prompts & personas

• [ ] Company-wide system prompt drafted (company/03-system-prompts/company-system-prompt.md)
• [ ] Legal disclaimer library published
• [ ] 15 company-wide prompts loaded to enterprise workspaces
• [ ] Individual master system prompt template distributed

Training

• [ ] Onboarding curriculum ready
• [ ] AI Champions named (one per department)
• [ ] Champions trained on the kit
• [ ] Employee kickoff communication sent

Success criteria at Day 30

• 100% of employees have accounts and have completed Module 1
• 100% of managers know the AUP
• Steering Committee met at least twice
• Zero data classification incidents

Days 31–60 — First workflows

Deploy first workflows

• [ ] Individual prompt library rolled out to all employees
• [ ] Department copilot launched for 2 pilot departments (usually Sales + Support)
• [ ] Meeting-notes-to-CRM agent piloted for sales team
• [ ] Inbox triage agent piloted with 10 volunteers

Evaluation

• [ ] Eval sets built for the first two department copilots (25 cases minimum each)
• [ ] Baseline scores captured
• [ ] Weekly regression running

Observability

• [ ] All LLM calls flowing through gateway
• [ ] Logging + redaction verified
• [ ] Cost dashboards live
• [ ] Adoption dashboards live

Communication

• [ ] "AI at [Company]" internal newsletter launched (biweekly)
• [ ] First "what shipped" showcase completed
• [ ] Employee feedback survey run

Success criteria at Day 60

• ≥ 50% of pilot department employees actively using their copilot weekly
• 3 real production workflows live
• Cost tracking working; monthly spend within 80-120% of forecast
• Zero critical incidents

Days 61–90 — Expand and measure

Expand

• [ ] Department copilots launched for 3 more departments
• [ ] Weekly report generator rolled out company-wide (opt-in)
• [ ] Second wave of 5 department-specific workflows shipped

Measure

• [ ] ROI measurement framework applied to first workflows
• [ ] Adoption goals set per department
• [ ] Cost per resolved-item baselined
• [ ] Executive dashboard reviewed

Iterate

• [ ] First model bake-off completed
• [ ] Prompt library reviewed and pruned
• [ ] Retired workflows documented
• [ ] Rules and policies updated per learning

Success criteria at Day 90

• ≥ 60% of company using AI weekly
• 10+ production workflows live
• Cost within budget
• One completed model bake-off with a decision
• Executive team can point to 3 measurable ROI examples
• Employee NPS on AI program ≥ 40

Days 91+ — Steady state

Enter continuous mode: - Monthly steering committee - Quarterly refresh + bake-offs - Biannual department deep-dives - Annual audit + strategy refresh - Continuous eval regression - Continuous adoption expansion

Rollout risks and mitigations

Executive checkpoints

• Day 30: Governance + accounts confirmed
• Day 60: First workflows shipping
• Day 90: ROI evidence + roll-forward plan
• Quarterly thereafter: dashboard review

level-1-essentials/company/10-rollout-playbook/expansion-decision-framework.md

Expansion Decision Framework

When and how to expand from pilot to broader rollout.

The decision moments

Three natural checkpoints:

1. Pilot → Department: After 30 days, expand to the full pilot department?
2. Department → Adjacent Departments: After 60 days, roll to more departments?
3. Adjacent → Company-wide: After 90 days, roll to everyone?

Do not skip stages. Each stage teaches you something you can't learn from the previous.

Gate criteria — Pilot → Department

Must have all:

• [ ] Quality: aggregate eval score ≥ 80% of maximum
• [ ] Safety: zero P1 incidents in pilot period
• [ ] Adoption: ≥ 70% of pilot users active weekly
• [ ] User NPS: ≥ 40 among pilot users
• [ ] Cost: within budget
• [ ] Owner ready: department Champion recommends expansion
• [ ] Docs ready: SOP, prompts, training material updated

Should have most:

• [ ] Time-saved estimate confirmed (rough ROI)
• [ ] At least 3 case studies from real users
• [ ] Retrospective completed with lessons captured

Gate criteria — Department → Adjacent

Same as above, plus:

• [ ] Champion identified in target department
• [ ] Adjacent department's data and permissions ready
• [ ] Playbook adjusted for new department's context
• [ ] Steering Committee approval

Gate criteria — Adjacent → Company-wide

All previous gates, plus:

• [ ] Cross-department feedback synthesized
• [ ] Training scaled (recording available, self-serve)
• [ ] Support model defined (help channel, escalations, office hours)
• [ ] Enterprise licensing at scale
• [ ] Formal exec sponsor for full rollout
• [ ] Comms plan for company-wide announcement

Warning signs — Do NOT expand

Any of these = pause:

• Any P1 safety incident
• Adoption < 50% at expected marker
• Quality trending down
• Cost trending above 130% of forecast
• Employee sentiment negative
• Champion resigns or is deprioritized
• Change in exec sponsor commitment

The rollback playbook

If a workflow fails after expansion:

1. Contain — pause the workflow immediately if quality/safety broken
2. Diagnose — analyze eval regression, user complaints, incident logs
3. Fix — revert to last known good version, or re-do the workflow
4. Communicate — tell affected users honestly what happened
5. Post-mortem — blameless, structural fixes
6. Re-launch — after fixes, run pilot-scale again

Communication cadence during expansion

• Weekly: operational update to Steering Committee
• Bi-weekly: all-hands mention of AI wins
• Monthly: exec dashboard review
• Quarterly: full program review with board or leadership

Expansion staffing

Successful expansion needs staffing. Typical model:

• 1 AI Champion per department (partial FTE — usually 20%)
• 1 AI Program Manager central (full FTE)
• 1 AI Platform Engineer central (full FTE, more as scale grows)
• 1 Security/Legal partner central (partial FTE)
• 1 L&D partner central (partial FTE)

Do not attempt company-wide rollout without this staffing.

Long-term ambition setting

After 90 days, the Steering Committee sets 12-month and 24-month ambition:

• Coverage goal: % of workforce using AI weekly (target: 90% in year 1)
• Depth goal: average workflows per user (target: 3 in year 1)
• Cost goal: cost per user per month (target: within enterprise SaaS band)
• Value goal: measured productivity or revenue impact (target: 10-20% team-relative)

Sunset criteria

Retire workflows that: - Fall below quality threshold and can't be fixed - Have adoption < 20% after 6 months - Are superseded by better tools - Violate updated policies - No longer have an owner

Sunset formally: announce retirement, provide alternatives, archive assets, remove from launchers.

level-1-essentials/company/10-rollout-playbook/pilot-selection-criteria.md

Pilot Selection Criteria

How to pick which workflows and departments go first.

The scoring model

Score each candidate on 1-5:

Total 40. Pick candidates scoring ≥ 30.

The archetypes of good first pilots

Archetype 1 — High-volume drafting

• Sales follow-ups, support responses, marketing content
• Well-defined output format
• Human always reviews before send
• Easy to measure time-per-task

Archetype 2 — Research and synthesis

• Weekly reports, competitive research, meeting briefs
• Reads a lot, writes a summary
• Low-risk (informational)
• Individual productivity gain

Archetype 3 — Data extraction

• Contracts, receipts, resumes, forms
• Structured output
• Ground-truth easy to check
• Bounded blast radius

Archetype 4 — Q&A over docs (RAG)

• Employee handbook Q&A, policy Q&A, product-doc Q&A
• Frequency: many small asks per day
• Reduces load on subject-matter experts
• Confidence-scored answers

Archetypes to AVOID first

• Autonomous external communication — sending emails, posting to social
• High-stakes irreversible decisions — approvals, denials, pricing
• Regulated workflows without human review — HIPAA, GLBA, GDPR-restricted
• Anything requiring long-horizon planning — plans over 5-10 steps are still fragile
• Novel domains without eval data — no way to know if it works

Department readiness scoring

Before picking a department for pilot:

Departments scoring ≥ 20 are ready. Below 15, wait or invest in readiness first.

The recommended first-pilot mix

Two departments, chosen to demonstrate range:

• One customer-facing (Sales or Support) — validates external-quality workflow
• One internal-facing (Ops, Finance, or HR) — validates safer internal workflow

Total 4-6 workflows across those two departments in the first pilot.

Success bar for the pilot

At the end of 60 days: - ≥ 60% of pilot department using ≥ 1 workflow weekly - Quality metrics ≥ baseline - No critical incidents - Champion recommends expansion - Sponsor recommends expansion - Steering Committee approves expansion

If any of these fail, do not expand. Iterate first.