Most AI plans describe a destination. Very few describe the road. This is the road: what the first quarter of a deployed AI operating system actually looks like, week by week, when the goal is a system in production rather than a report.
The shape below is the one Masar runs, Diagnose, Design, Deploy, Drive. The dates are illustrative, but the sequence and the discipline are not. The point of writing it down is simple: if your AI initiative does not have a week two, a week five and a day sixty that look like this, it is probably a strategy exercise wearing a delivery costume.
Weeks 0 to 2 — Diagnose
The first fortnight is not about AI at all. It is about where the business loses hours, money and quality today. We interview the people who do the work, map the real workflows rather than the ones on the org chart, and rank automation candidates by return, not by novelty.
What ships in this phase is a friction inventory and a shortlist: the two or three workflows worth automating first, each with an estimate of the hours or money it costs the business now. This is deliberately unglamorous. The most expensive AI mistakes are made in week zero, by starting with the workflow that is interesting instead of the one that is bleeding.
Weeks 2 to 5 — Design and first deployment
Design and build overlap on purpose. We architect the system around the specific business, connect it to the tools the team already uses, and get the first automation into production. Not a pilot in a sandbox, production, touching real work, watched closely.
A pilot in a sandbox proves the technology works. A system in production proves the business changed. Only one of those compounds.
By the end of week five there is a working system doing a job that a person used to do by hand, and the team has watched it run. This is the moment the initiative stops being a promise. It is also the moment resistance tends to fade, because the first thing a good deployment returns is time to the most senior person in the room.
Day 30 to Day 60 — Deploy the rest, then audit
With the pattern proven, the remaining automations from the shortlist go live. Then the cadence changes from building to improving. The first monthly audit scans what is running, surfaces new friction, and quantifies what the deployed systems have already returned in hours and cost.
Two numbers matter here. The first is time-to-first-value, measured in weeks, not quarters. The second is the cumulative hours the system has handed back, tracked from day one so the return is a fact on a tracker rather than a claim in a deck.
Day 60 to Day 90 — Drive
The final month is the one most AI projects never reach, because most AI projects are projects. An operating system is not. The top one or two findings from the audit ship as new automations. The KPI tracker updates with month-over-month deltas. The quarter closes with a short executive review: what shipped, what it returned, what compounds next.
What it costs
The honest answer is: less than the meetings it replaces, if the sequence is right, and pure loss if it is not. A system that saves a senior operator a day a week pays for itself long before the quarter ends. A tool subscription nobody adopts costs its licence fee plus the quarter you spent not deploying. The difference is not budget. It is whether, at day ninety, you can point to a cost line that fell or a revenue line that rose.
If you can, you do not have an AI experiment. You have an operating system, and it is worth more this month than it was last.