90-Day AI Rollout · 40-Person Company
The First 90 Days
What a governed AI rollout actually looks like inside a 40-person company. Education, governance, deployment, and adoption run in parallel from day one.
Most AI projects don't fail because of the technology. They fail because of poor deployment, governance, and adoption.
Agilitec
Days 1 to 30
Establish the floor
Days 31 to 60
Ship the first wins
Days 61 to 90
Prove it, then expand
Governance
Days 1 to 30 · Establish the floor
- Shadow AI scan names what people are already using. No guessing.
- Approved tool list and acceptable use standard signed and trained on.
- Data labels applied, prohibited data published, executive sponsor named.
Days 31 to 60 · Ship the first wins
- Human in the loop defined per workflow before anything is automated.
- Prompt standard and output validation rules live.
- Unapproved GenAI tools blocked at the endpoint. Exceptions logged, not argued.
Days 61 to 90 · Prove it, then expand
- Configuration drift review on a monthly clock.
- Audit evidence path tested against a real request.
- Agent inventory with a named owner for every agent.
Education
Days 1 to 30 · Establish the floor
- AI literacy baseline on all 40 people. Comfort, usage, and quality scored.
- Three champions named across departments.
- Role-based learning paths built from real jobs, not generic curriculum.
Days 31 to 60 · Ship the first wins
- Department training in 60-minute blocks, tied to the work in front of them.
- Prompt library seeded from actual deliverables the company produces.
- Champions run weekly office hours. Questions get answered in-house.
Days 61 to 90 · Prove it, then expand
- Every seat trained and validated, not just licensed.
- Champions certified. Skills gaps measured, not assumed.
- Internal admin owns configuration and first-line support.
Deployment
Days 1 to 30 · Establish the floor
- Identity, MFA, and device baseline confirmed before a license lands.
- Permissions and oversharing cleaned on the top five repositories.
- Pilot of eight named users with a written rollback plan.
Days 31 to 60 · Ship the first wins
- Approved tools live for the pilot cohort. Grounding scoped to allow-listed sites.
- Two out-of-the-box agents in production inside the first 45 days.
- First custom workflow prototyped against a measured baseline.
Days 61 to 90 · Prove it, then expand
- Remaining seats enabled in two waves, each gated on training.
- Custom workflow shipped with a test and promotion path.
- Support runbook handed over. Escalation path per workflow.
Adoption
Days 1 to 30 · Establish the floor
- Launch communication comes from the sponsor, not from IT.
- Outcomes defined per role, measured in hours returned.
- Top workflows inventoried by department. Highest-pain work goes first.
Days 31 to 60 · Ship the first wins
- Usage telemetry reviewed weekly. Quiet users get a call, not a reminder email.
- Output quality sampled, not assumed.
- First hours-back captured and published internally.
Days 61 to 90 · Prove it, then expand
- Cost and value scorecard at the 90-day review.
- 12-month roadmap with the next three use cases priced.
- Monthly AI council owns the backlog from here.
Maturity arc
1 Discovery
2 Onboard
3 Generative AI
4 Agentic AI
5 Autopilot AI
Ninety days clears the first three. Agents scale only once the floor holds.
Agilitec · Advisory, Enablement, Cybersecurity · agilitec.com
Illustrative rollout for a 40-person organization. Governance scored against the Magic Framework, 114 controls across 10 areas.