A working field guide for whoever just got handed the job of building an AI and data capability from nothing — not one model, but a program that creates value and an office that keeps creating it after you leave.
Someone has just been handed a hard job: walk into a company that has never built an AI and data capability, and build one. Not a single model. Not a proof of concept that dies in a slide deck. A living program that creates value, and a standing office that keeps creating it long after the founding team has moved on.
This is the shape of that job. It holds across industries — a bank, a hospital system, a retailer, a manufacturer all walk roughly the same path. What changes is where the money sits, which data fights hardest, and what the culture resists. The shape stays the same.
The program is the portfolio of work that creates value — the fraud model, the demand forecast, the maintenance predictor, the service copilot. It is the reason anyone funds you. The office is the standing organization that delivers and sustains it — the people, the operating model, the governance, and the platform.
Most efforts that fail pick one and neglect the other. Chase only use cases and you end up with a drawer of demos and no way to run any of them in production. Build only an office and you end up with a large team, a lovely platform, and nothing that shows up in the numbers.
The program produces the proof and the funding that justify the office. The office is what keeps the program alive after the first few wins. Hold both in your head at all times.
Two questions settle first, with the CEO and the sponsor. What kind of bet is this — cutting cost, growing revenue, or keeping pace with competitors already moving? The honest answer sets which use cases you favor and how fast you need to show results. And where does the office report? That sounds like an org-chart detail; it quietly decides whether you succeed a year out. The value lives in the business, not the technology function, so the office has to sit close to where the value is. Bury it inside IT and it will produce clever work nobody adopts.
Spend the first six to eight weeks on diagnosis: where the company makes and loses money, what data exists and how good it is, where the pockets of hidden skill are, and what governance will allow. Then score every candidate on value and feasibility — and treat adoption as part of feasibility, because a model nobody will use is not feasible, it is a science project.
Choose the portfolio on purpose. A few genuine quick wins to build momentum and earn the right to ask for more. One to three lighthouse bets that justify the whole effort. The quick wins buy you credibility; the lighthouses buy you a future.
The single most important lesson lives in delivery: what separates success from failure is adoption, not modeling. A churn model is worthless if the retention team never acts on it. So do the unglamorous work — embed with the people who will use the output, redesign the workflow around it, put it inside the tools they already open. Then baseline the before, measure the after, and get finance to sign the number. Finance-validated value is the currency that funds the next wave.
One use case working in one place is a pilot. The same one running across every branch, plant, or hospital is a program, and that is where the value compounds. You get there by moving from hand-built to manufactured, and by treating change management across dozens of teams as the main event rather than an afterthought.
The engagement ends when the office can run without you. If it keeps creating value after you stop paying attention, you built a capability. If it stalls the moment you look away, you built a dependency. Aim for the first.
Bring us where you're stuck — a mandate, a stalled pilot, or the whole build. We'll tell you where we'd start.