Every team has one AI person already. Deputize them.

In every company I talk to, the same person exists. She's in finance, or ops, or marketing, and months ago she quietly rebuilt chunks of her job around AI: the weekly report, the first-draft analysis, the inbox triage. She never asked permission. Her output got faster and cleaner, and a few desk neighbors noticed and started asking how. Leadership, meanwhile, is still debating which vendor to pilot.

That person is the most valuable asset in your AI rollout, and most executives can't name her. The instinct is to put out a call for volunteers. Resist it. Volunteers select for enthusiasm. You want the people who already have results, and there's one question that surfaces them: 'Who do people on your team already go to with AI questions?' Ask it in a few skip-levels and the same names come back fast.

Once you know the names, the highest-yield thing a champion can do costs almost nothing: a peer demo. Fifteen minutes of 'watch me do my actual job' moves a team further than any all-hands announcement, because the audience watches a colleague with the same tools and the same Tuesday produce real work in front of them. Nobody can dismiss that as a vendor pitch. It's also the fastest cure for the copy-paste commute, since people see work happening inside the file instead of shuttled through a chat tab.

But the moment you name champions, you owe them something. A title with no time attached is not a program, it's extra unpaid work for your best people. Deputizing means a written mandate, protected hours, and a leader who publicly backs the time spent. Without that, you've taxed the exact people you should be investing in.

The first trap is the lone hero. One champion for a whole company isn't a program, it's a bus factor. Every AI question in the building routes to one person, on top of their day job, with no backup and no relief. They burn out or they leave, and the capability walks out the door with them because nothing was written down. That's why playbooks matter: the workflow has to outlive the person who invented it. And it's why champions should be plural from day one.

The second trap is mistaking enthusiasm for judgment. The champion is not the person most excited about AI. It's the person whose AI-assisted output ships clean: the analysis that holds up, the draft that needs no rescue, the automation that doesn't break. Your loudest AI fan might be your sloppiest reviewer, and handing them a megaphone spreads bad habits at scale. Pick champions by inspecting work product, and judge the program by numbers you can defend: hours saved and workflows changed, not vibes.

The last trap is the biggest: treating champions as the whole strategy. Champions create pull. They prove the gains are real, locally, with someone the team trusts. What they can't do is give every employee structured practice on their own role and their own files, and that gap is why most corporate AI training fails when companies wing it. Champions accelerate training; they don't replace it. Run both: champions for social proof and momentum, structured role-based training to raise the floor for everyone else.