Most AI training teaches people what the tool can do. They sit through a session, learn the features, maybe pass a quiz, and go back to their desks no better at the part that matters: telling when the answer in front of them is wrong, and knowing what to do about it. That gap is a judgment gap, and a course on features will not close it.
Mid-market leaders feel it. Nearly half (46%) of the finance leaders Capital One surveyed this summer named AI literacy as a key skills gap at their companies, and Capital One's own conclusion is that training on its own does not get people using new technology. If the training is aimed at the tool instead of the work, it will not.
Companies say balance, then spend on tools
Two thirds of those leaders say they want to split their investment evenly between buying tools and building their people's skills. Their budgets say otherwise. More than half report putting roughly two dollars into technology for every dollar they put into training their own staff.
I have sat on the buying side of this. When I led digital marketing and analytics at World Market, every tech pitch started the same way: someone walked in with a platform and a savings number. The first question from the table was never about the software. It was what we were actually trying to fix, and how we would know it worked. A powerful tool aimed at a goal nobody chose still gets you nowhere, just faster. Spending two to one on tools before your people know what the tool is for repeats that mistake.
AI literacy is three questions your team can answer
I would call a team AI-literate when the people using the tool can answer three things about their own work. When is this output wrong, and how would I catch it? Who is accountable for acting on it once it is right? And what is this tool for here, in this workflow, as opposed to everything it could do?
None of that comes from a features session. It comes from people running their real work through the tool, side by side with how they do it today, and seeing where the two answers differ. At the AI company I founded and ran for two years, the standard was that every tool we recommended had been tested first. Your team needs to do that testing themselves, on their own work, because that is where the judgment gets built.
The cheaper the tools get, the more the judgment costs. Anyone can buy the software now. Knowing when to trust it is the part your competitors cannot buy off the shelf.
Point the training at better work
People learn a tool fastest when it is clearly there to help them do their job better. If the first announcement about AI sounds like a headcount plan, the people who know the workflow best will stop showing you how it runs, and those are exactly the people you need teaching everyone else. My rule is together, not instead. Measure AI against the follow-up nobody had time to send, or the analysis nobody had time to run, because that work was never getting done.
When more training will not fix it
Training will not help if nobody owns the workflow it is supposed to improve. It spreads thin across a company that has not decided what AI is for, and people go back to their old way within weeks.
It also will not create judgment that nobody in the building has yet. If no one on your leadership team has made these calls before, the faster route is to borrow that experience for a while from someone who has, and have them leave the skill behind with your team.
Start with one team's own week of work
Pick one team whose output your leadership already watches. Have them run a normal week of their real work through the tool, and mark every output they would send as is, every one they would fix, and every one they would throw out. Then talk through the fixes and the throwaways together. That conversation is the training.
If you want a second set of eyes on where to start, here is how I work with leadership teams on AI. For what comes after the first win, I have written about who runs the AI once the vendor leaves.
John Wyatt · John Wyatt's page · All posts · RSS