Start with one win, not a transformation.
The businesses that end up genuinely running on AI almost always started with something small enough to be slightly embarrassing. One narrow job, done properly, live in about a week.
The ones that stall usually started with a strategy. A workshop, a roadmap, a list of nine opportunities ranked by impact and effort. Six months later the roadmap is still a document, and the only thing that has changed is that everyone is now slightly tired of hearing about AI.
Why small wins compound.
A small win does three things a big plan cannot.
It produces evidence. Not a projection, a thing that happened. An inquiry that got answered at 9pm. A no-show that did not happen. Evidence ends arguments that opinion cannot.
It is reversible. If it does not work you turn it off and you have lost a week, not a quarter. That changes who is willing to say yes.
It teaches the business what it actually wants. Almost every business we work with revises its own priorities after the first thing goes live, because seeing one real system running is more informative than any amount of planning about systems.
Why big starts stall.
Not because the ideas are wrong. Usually the big plan is a perfectly good description of where the business should end up.
It stalls because nothing in it can be finished. Every piece depends on another piece. The budget is large enough to need real approval, so it gets scheduled rather than started. And staff, sensibly, do not invest energy in learning a system that might change again before it is finished.
Meanwhile the small thing you could have shipped in week one would have been running for five months.
What a first win should look like.
Four tests. It should be narrow enough to describe in one sentence. It should be live in about a week. Somebody should be able to point at a result without doing arithmetic. And it should be something the business already knows is a problem, so nobody has to be convinced it mattered.
If a proposed first project fails any of those, it is a second project.
The honest catch.
Starting small only compounds if you actually go back. A first win that nobody builds on is just a nice week. The point of starting small is not modesty, it is momentum: each finished thing makes the next one easier to approve, faster to build, and more obviously worth doing.
So the question we ask before anything gets built is not "what is the biggest thing AI could do here." It is "what is the smallest thing that would make the next conversation easier."