By now the problem is easy to see. AI that speaks with total confidence over records that are months out of date. Value that leaks away in the spaces between the systems you already own. Tools that get measurably faster while the one number you actually care about doesn't move at all. The diagnosis isn't the hard part anymore — most leadership teams can recite it. The hard part is the question that arrives the moment the diagnosis lands: so where do I actually start?
It is a fair question, and it has a bad default answer.
The instinct is to answer big
The instinct — reinforced by every vendor deck, every board slide, every peer who doesn't want to look timid — is to answer at scale. A transformation program. A two-year roadmap. AI rolled across the whole business at once, so that no one in the room can say you weren't ambitious. It looks like leadership. It photographs well. And it is the single most expensive way to discover that you started in the wrong place.
The reason has nothing to do with how good your plan is. A two-year program commits you to a long list of assumptions — about where the value actually sits, about what the work really involves, about how your people and your systems will behave once the AI is in the room. You cannot test most of those assumptions until you are deep into the build. And the ones that turn out to be wrong, you find out at full scale, after you have already paid to be wrong everywhere at once. Big bets fail big. The cost of being wrong is set the day you choose the size of the move, not the day the mistake finally shows itself.
The quieter answer that actually works
There is a different answer, and it fits in a single sentence: pick one corner and get it genuinely right before you touch the second.
Not the flashiest corner. Not the one in the headline, and not the one the loudest vendor is pushing. One flow that genuinely matters to the business — chosen on two plain tests, both of which an executive can apply without a single technical term.
Choosing the corner: two plain tests
The first test is legibility. Can you see exactly what is broken, and exactly what "fixed" would look like? "We should use AI in finance" is a category, not a corner. "Our invoices take eleven days to clear because the same figures get re-keyed by hand three times, and fixed means they clear in two" is a corner. If you cannot draw the before and the after in a sentence each, the problem is not yet legible enough to start in — however important it feels in the abstract.
The second test is measurability. Will you know, inside a quarter, in numbers a CFO will accept, whether it worked? A real corner has a number attached to it before you begin — days, dollars, error rates, hours returned — and a date by which that number will either have moved or not. If the only way to tell whether the effort succeeded is a survey of how people feel about it, you have chosen something that can never be proven. And something that can never be proven can never fund the next move.
If you cannot answer both questions plainly, it is the wrong corner, no matter how exciting it sounds in the planning session. The discipline is in the refusal: you do not start where you cannot see clearly and measure honestly, even when the unmeasurable thing is the one everyone is excited about.
What "getting it right" actually means
Once you have the corner, the work is unglamorous, and that is the point. Three things have to become true in that one place — and only in that one place, to begin with.
The records it relies on have to be current and trustworthy. An AI reasoning over stale or contradictory information will be confidently wrong, and no amount of model quality repairs a bad source. Getting the records right in a single corner is bounded, finite work you can finish. Getting them right across the entire business is the two-year program you are trying to avoid.
The connections into the rest of the business have to be clean and deliberate. The corner does not live alone — its output becomes someone else's input, and that handoff is very often exactly where the value was leaking in the first place. Closing it in one place, so the work crosses from this flow to the next without a person carrying it across by hand, is usually where the measurable win actually comes from.
And the authority has to be right. Decide, explicitly, what the AI is allowed to settle on its own and where a human still has to sign. Start that line tight — narrow authority you widen later, once you have watched the thing earn the trust — rather than broad authority you have to claw back after it has committed you to something you can't honor.
Records, connections, authority. That is not a platform, and it is not a strategy. It is one corner of the foundation, done properly — small enough that you can stand on it and feel that it holds your weight.
Why small beats big — and it isn't caution
It would be easy to hear all of this as a counsel of timidity: start small because big is risky. That is not the argument. Starting small is how you move fast, because it is how you learn the things no plan can teach you.
A two-year program tests its assumptions in year two, at full cost. One corner tests the same assumptions in a quarter, at almost none. You find out by doing — the real shape of the work, the place the records were dirtier than anyone admitted, the handoff everyone had forgotten was still manual — and you find it out while being wrong is still cheap. A team that has finished one real corner understands more about how AI behaves in your business than a team that has spent six months writing the strategy for all of it. They learned it in the only place it can be learned, which is in the doing.
The corner pays for the next one
There is one more reason to start in a single, measurable place, and it is the reason that compounds. A corner with a real number attached to it pays for what comes after — in two currencies.
The first is budget and mandate. A measurable win changes the conversation you have with the board. You stop asking them to fund a leap of faith and start showing them a result and asking to do the same thing again, somewhere else. That is a far easier sentence to say, and a far easier one to approve. Proof is cheaper to raise money against than ambition.
The second is direction. The corner you just finished tells you which corner to take next — because you now know, from having done the work, where the next legible problem and the next measurable win actually sit. You could not have known that from the plan; the plan was a guess made before you had touched anything real. Each corner you close sharpens your view of the one after it, until the sequence is no longer something you designed up front but something the work itself keeps revealing. That is what a foundation laid one corner at a time gives you that a big-bang program never can: it funds its own next step, and it points at where to take it.
The same discipline, seen from the other side
If you have watched a building go up, you have already met this idea from the builder's side. No one frames the top floor before the floor beneath it will carry weight; the structure rises in the only order that holds. And the teams that get AI right tend to grant it narrow authority first and widen it on purpose, rather than handing over everything at once and retreating after the first public mistake. Floor before floor, tight before loose — the same instinct each time. Fixing one corner first is simply what that instinct looks like when it is your move to make, with your budget, on your timeline.
The question to take into the room
Don't ask what your AI strategy is. Ask the smaller, harder question: of everywhere AI could go in your business, which one corner would you stake the first proof on — and what would it have to show you before you earned the right to widen?