43% will fail. Nobody prepped the people.

43% of enterprise AI initiatives are expected to fail. Not because the models are weak. Because the people were never prepped.

That's from HCLTech's new report. 467 leaders, billion-dollar companies, an 18-month clock running on every AI dollar they spend.

The read is right. It's also incomplete.

What "prep the people" becomes

Watch what happens to "prep the people" once it enters a real company. It becomes "train the people." Workshops get scheduled. Licenses get bought. A certification path shows up in the LMS. Someone stands up a dashboard to track adoption. All of it real, all of it useful, and none of it answering the one question that decides whether the initiative survives contact with the business:

Usually no one. The AI got handed to IT to stand up, to a center of excellence to evangelize, to a steering committee to bless. Plenty of custody. No ownership. Custody keeps the thing running. Ownership is the person whose number moves, or doesn't.

Where it stalls

So the pilot works. The demo lands. Leadership nods. Then it stalls just past the pilot, because the first time it demands a hard call — kill this step, reassign this team, absorb a short-term dip — there's no one with both the authority and the accountability to make it.

That's not an enablement gap. It's an operating-model gap. And you can't train your way out of a structure problem. A trained team with no owner just reaches the dead end faster, with better tooling.

The cheap fix nobody reaches for

Here's the part the 18-month crowd won't like: ownership is the cheap fix. It's a decision, not a budget line. You could name an owner this week. The reason most companies don't is that naming one puts a real person on the hook in public, and that conversation is harder than approving another round of training.

So before the next rollout, the question isn't whether your people are ready. It's whether anyone owns the outcome by name — and whether the org chart actually lets them make the calls that owning it requires.