The 90-day rollout playbook

AI rollouts that succeed do not look like big-bang launches. They look like a small pilot that becomes obviously useful, an expansion that holds up, and a hardening phase that makes the system boring. 90 days is the cadence most execs can stay engaged with.

Days 1-30 — pilot

• Pick one team, one use case, the smallest valuable agent.
• Daily standups with that team for the first two weeks.
• Weekly metrics: usage, time saved, failures, NPS from the team.
• Exec readout at day 30: ship the metrics, name what's working, name what's not.

Days 31-60 — expand

• Onboard 2-3 more teams that match the original pilot pattern.
• Move daily standups → weekly. Keep the metrics dashboard.
• First platform investment: whatever the pilot taught you was missing (usually eval infrastructure or shared tools).
• Exec readout at day 60: how the pattern is generalizing, where it's not.

Days 61-90 — harden

• Cost model is real. Cost per task is measured.
• Runbooks exist for the top failure modes.
• Governance doc has been written, ratified, and used at least once.
• Exec readout at day 90: state of the platform, proposed 12-month investment, ROI to date.