Your AI training budget is a line item, not a strategy

Most AI training budgets go to three things: platform licenses (LinkedIn Learning, Coursera), lunch-and-learns, and prompt workshops. While 84% of companies plan to raise AI investment next year, only 35% have a mature, organization-wide program for teaching people to use it, and the distance between those two numbers is where the budget goes.

Platform licenses are the default move. L&D signs a deal, sends an email, and waits. Open online courses finish about 3% of the people who start them, according to an MIT and Harvard study of 5.6 million edX learners, and a corporate library runs on the same self-paced model. Live, instructor-led programs finish at a far higher rate, because someone in the room expects everyone back.

Six months in, in the rollouts I've seen, hardly anyone has logged in since week one. The license renews anyway, because canceling it would mean admitting the initiative failed.

Lunch-and-learns have a different problem. Whoever has a free hour runs them, and that person has usually never built AI into real work. I've talked to L&D directors who run these monthly. Not one has followed up to check whether anyone applied a single thing.

Prompt workshops are the newest thing: half-day sessions on crafting the perfect prompt. Teaching someone to prompt better without showing them where AI fits in their work is like teaching someone to type faster without saying what to write.

What all three have in common: nobody measures whether anything changed. Ask how training went and you hear "people seem more comfortable" or "engagement was positive." Those answers measure sentiment, and sentiment says nothing about whether anyone works differently.

A strategy starts with a specific pain point. Which tasks eat the most time for the least value? For an operations team it is usually the weekly status report; for legal, vendor contract review; for finance, the monthly commentary. The strategy begins with one of those tasks and nothing broader.

The measurement changes too. Completion rates show who clicked through and login rates show who remembered their password, which is about all they show. Whether anyone's work changed is a separate question, and almost nobody asks it.

The best programs track one metric: can someone now do in 30 minutes what used to take 3 hours? If yes, training worked. If they can't name a task that changed, it didn't.

Once AI sits inside an existing workflow, adoption stops being something anyone has to remember. The weekly report gets drafted from the tracker because that is now how the report gets made. Teams that reach that point stop saying "we trained on AI" and start saying "this is how we do the report."

Look at where AI training sits in your L&D budget. If it's next to compliance training and the annual leadership retreat, it will deliver the same results: a checkbox.

Compliance training exists to protect the company from liability, and leadership retreats exist to make people feel valued, so neither was ever designed to change how work gets done. AI training that sits beside them inherits the same expectation: attend, check the box, move on.

Framing sets the expectation. A budget line called an investment gets asked about returns at the next review, while a line called a perk never does.

I teach 75+ students at the University of Chicago and build AI automation systems at a $40B enterprise. The skill gap is the same on both sides.

A 22-year-old graduate student and a senior VP with two decades of experience make the same mistakes. They hit the same walls and have the same breakthrough moments. The VP just takes a little longer to admit they're stuck.

Everyone has heard of AI and most have tried ChatGPT, so the gap sits further along, at the point where somebody shows them where AI fits in their specific work: the report they write every week, the follow-up email they send after every meeting.

Until training gets that specific, it stays theoretical. And theoretical training is what a line item buys.