“How should we use AI?” sounds responsible. It is usually too large to answer well.
The question invites a tour of possibilities. Someone lists tools, demos automations, describes what competitors are doing, and produces a roadmap. The room feels productive because it contains many ideas. But the ideas are rarely attached to an owner, a number, an adoption condition, or a decision date. The list expands. The decision does not.
Ask a smaller question:
Where is the business experiencing a costly friction, and does AI deserve the right to be tested there?
That question begins with the business as it exists. It asks what is slow, expensive, inconsistent, risky, frustrating, or needlessly difficult today. It asks who feels the problem, what it costs, what has already been tried, and what evidence would justify doing more.
It also allows “no” to be a successful answer.
This matters because AI creates a peculiar kind of pressure. The technology is broad enough to touch almost every activity in a company. That does not mean every activity deserves attention. When everything is possible, prioritization becomes more valuable than imagination.
The executive's job is not to keep pace with every capability. It is to allocate attention and capital. That means deciding:
- which business problems deserve investigation;
- which possible uses are ready enough to test;
- which experiments are cheap enough to learn from;
- which results are strong enough to fund;
- which projects should stop;
- and which capabilities should remain deliberately ordinary.
The method has four movements:
- Map the business. See where value is created, delayed, lost, protected, or put at risk.
- Choose the strongest opportunity. Select one costly friction using impact, likelihood, adoption friction, time to evidence, cost, and risk.
- Buy the evidence cheaply. Run a small experiment in real work, with a real owner and a decision rule agreed in advance.
- Manage what follows. Stop, continue, invest, or industrialize. Then manage the set of decisions as a portfolio of business outcomes.
You can use this method without becoming an AI expert. You do need to know your business, invite uncomfortable facts into the room, and refuse to confuse activity with progress.
That is how to do all of this: not all at once, and not everywhere. One justified decision at a time.