The first ninety days of Lights-Out
Every executive who believes the thesis asks the same next question: Monday morning, what actually happens? Here is the answer we give design partners, without the mystique. Ninety days, three moves, one number at the end.
Days 1–30: pick the wedge, encode the model
Choose one process you already treat as an outcome — payables you’ve outsourced, reconciliations a shared-services center grinds through, audit preparation you buy by the hour. One process, one entity cluster, deliberately unheroic. Then do the work most programs skip: capture the Target Operating Model for that process as a runnable specification — thresholds, approvers, controls, escalation paths — and encode the Data Model it runs on: the accounts, the calendars, the policies. This is unglamorous, and it is the whole game. The agent is only ever as governed as the model it operationalizes.
Autonomy programs don’t fail at the model. They fail at the operating model.
Days 31–60: run at observe, then assist
Put the agent on the process at A0 — observing, matching, scoring, writing everything to the Evidence Ledger, touching nothing. This phase produces the two artifacts that decide everything after: a baseline (what touchless rate, what cost, what cycle time the process really has, measured rather than believed) and a divergence log (where the agent’s judgment differs from your team’s, adjudicated case by case). Promote to A1–A2 as divergence converges. Your people execute; the agent prepares. Nothing has been risked; everything is being evidenced.
Days 61–90: gate, act, and count
Set the materiality thresholds, turn on act-with-approval below them, and route everything above to the exception queue — with attribution and rationale required on every human decision. By day ninety you hold the only deliverable that matters: a measured delta. Touchless rate against baseline. Cost per unit against baseline. Cycle time against baseline. Chain-verified evidence for every action behind those numbers. Not a business case — a result, small enough to trust and large enough to fund the next process.
What the ninety days buys
Three things compound from here. The evidence base makes the second process faster than the first — the auditors have seen the ledger, the committee has seen the exception queue, the operating-model pattern is reusable. The autonomy ramp is now an institutional muscle rather than a leap. And the economics shift from projection to observation: you stop debating what agents might save and start deciding how fast to scale what they measurably did. The enterprises that reach Lights-Out first won’t be the ones with the boldest vision. They’ll be the ones that started with one process, ninety days, and a number.