Where to start with AI

The hardest question in AI right now is not which tool to buy. It is where to point it. Most teams we meet are not stuck because they chose wrong. They are stuck because everything looks like a candidate and nothing feels safe to commit to. So they wait, or they try a little of everything, and a little of everything rarely adds up to much.

Here is what starting well actually looks like. Earlier this year Up Strategy Lab, the studio we build with, ran a workshop called AI for Non-Dummies. A room of professionals, most of them not technical, spent an afternoon moving from typing questions into a chatbot to building small automations around their own work. Nobody left with a grand AI strategy. They left having made one useful thing.

That afternoon is a good map for how any team should begin. It comes down to three decisions.

What to optimise.

In the workshop, people built what we call a skill, a short file that captures something specific about how they work, their tone of voice, their ideal customer, the way they describe their product. It takes about ten minutes to make, and once it exists the AI stops giving generic answers and starts giving yours. The tasks worth pointing it at were the boring, frequent ones. Meeting follow-ups. First drafts. The research pass before the real thinking starts. One example from the day: writing up meeting notes and next steps went from around thirty minutes to five. Not glamorous. Just an hour or two back, every week, for every person doing it.

What to hand over, and how much.

The workshop's core shift was moving from chatbot to automation, from asking the AI a question to letting it run a task. That is really a question of how much you delegate. There is a range, from the AI drafting while you decide everything, up to the AI running on its own with the occasional check. The skill is choosing where each task sits, on purpose. Start where a mistake is cheap and a person still signs off. Move a task further only once it has earned it. The teams that get caught out are usually the ones that jumped straight to the top because a demo looked impressive.

How to actually begin.

Pick one real task. Get the specific person who does it comfortable with the tool, on their own work, not in a generic training slot. Build the small thing. Check that it saved real time. Then move to the next one. By the end of that single afternoon, people who had never written a line of code had built a small working app from an idea in their head. The lesson was not that everyone should go and build apps. It was that the barrier to starting is far lower than it looks.

This matters for the budget, not only the calendar. The expensive path is buying a stack of tools before you know the job they are for, handing them to everyone, and hoping something sticks. Starting narrow and proving the return means you stop paying for things nobody uses, and the hours you win on the first task help pay for the next step.

We hold our own work to the same test. We keep only what earns its place, which is how we cut our own AI running costs by around 70 percent. The point is the discipline: aim AI at the right few things, and leave the rest alone.

If you are staring at a long list of maybes and unsure where to commit, that is the conversation we have most often, and the kind of session we run. A sensible first step is small. A short readiness check to see where you stand today, then one straight session to choose the one or two places worth starting. You leave with something you could act on next week.

Reach us at hello@oneaiadvisory.com.

#AIAdoption #AITransformation #AIStrategy #FutureOfWork