Calling AI a risk is the right instinct. The trouble starts with what most leadership teams do next: some freeze and wait for the picture to clear, and others buy their way past the worry with a big rollout. Both moves feel like decisions, and neither one teaches you anything about whether AI works in your business.
That choice is in front of a lot of CEOs right now. In the Conference Board's third-quarter read on CEO confidence, the chief executives of large U.S. firms put cybersecurity and AI at the top of their risk list, ahead of geopolitics, as PYMNTS reported in August. The same leaders were noticeably more hopeful about the economy than they had been a quarter earlier. So they are more hopeful and more worried at the same time, and a mid-market owner who feels the same way has less room than a large company to absorb a wrong call.
All in and all out are the same mistake
Skepticism about AI is healthy. Skepticism with no motion behind it is how companies get stuck, because the risk does not shrink while you wait. Your competitors keep learning, your people keep using whatever tools they found on their own, and you have no evidence either way.
The opposite move fails for the same reason. A company that commits a year of budget to a platform before it has tested one workflow has bought a bigger version of the same uncertainty. My rule is trust but verify, and trust but learn. Verification is how you find out what to trust next.
Treat AI like every other growth lever
I spent two years founding and running a company that built, deployed and measured AI inside real businesses, and I hold AI to the same standard as any other growth lever: a clear question, a small test, a number you check, and then a decision about whether to scale.
In practice that means picking one workflow where the cost of a wrong answer is low and the benefit of a fast one is easy to see. Decide before you start what result would make you expand it and what result would make you stop. Name one person who owns the test and reports the number. Systems that work can scale, but you cannot scale a system you do not know works.
The person you tap has to bring everyone along
The leader you tap on the shoulder to look into AI has a second job, and it matters as much as the test. Whatever path they find, they have to bring the rest of the company with them. AI cannot be a separate journey that one team takes while everyone else watches from the side.
When it becomes one team's project, the other teams treat it as that team's tool, and the lesson stays in one corner of the business. Ask your owner to report what they learned to the whole leadership team, invite the people whose work it touches into the test early, and judge them on how many teams end up using what works as well as on whether the pilot passed.
Use the hiring plans as your test bed
A third of those CEOs (34%) expect to add people, up from 28% the quarter before. If your plan also adds people next year, I think your hiring plan is the best place to run your first AI test, because it ties the tool to growth you have already decided to fund.
When I look at an AI idea, the comparison I care about is not AI against a person doing the work. It is AI against no response at all: the follow-up nobody had time to send, the report nobody had time to build. Ask where AI lets each new hire cover work that is going undone today. That is a smaller, safer bet than replacing anything, and it is easier to measure.
When the risk is security, start with controls instead
This approach fits when the worry is whether AI will pay off. It does not fit when the worry is exposure: customer data pasted into public tools, or systems your team connected without anyone checking. The CEOs in this survey ranked cybersecurity right alongside AI, and for good reason. If that is your situation, the first step is finding out what tools are already in use and putting controls around them, with someone qualified in security, before you run any pilot.
Pick the one test you will run this quarter
Write down one workflow, one owner, one number and one date. If the test works, you have earned the right to spend more, and if it fails you have paid a small price for an answer that freezing or a big rollout would never have given you.
I have written before about whether your AI plan sits inside your strategy or beside it, and that is the next question once a test works. If you want help choosing where to start, here is how I work with leadership teams, and I am glad to talk it through.
John Wyatt · John Wyatt's page · All posts · RSS