You Upskilled Your Marketing Team for AI. For What?

AI upskilling only creates value when your operating model can absorb it. Training marketers on AI tools before you redesign the workflows, decision rights, and data they will use produces capability the organization has no way to deploy.

You trained the team. What did the training produce?

Boston Consulting Group’s 2026 CMO survey found that 96% of CMOs report significant end-to-end AI transformation of their function. The same survey found that 42% still use generative AI only as an assistant for individual tasks in a handful of workflows. [1]

Both numbers describe the same people.

So when three CMOs tell BCG “the talent doesn’t exist, I have to create it,” and a B2B CMO upskills a 3,000-person marketing team through in-person training, hackathons, and a standing AI academy, the question the press release skips is the one that matters most. [1] You created talent. To do what?

That question sounds rhetorical. It isn’t. It has a specific answer, and whether you can give it decides whether the training was an investment or a line item.

What had to be true before anyone sat down to train?

Upskilling is the middle of a sequence. The work that comes before it decides whether the training lands.

Before the first session, did you redesign the work the trained marketer walks back into? Or do they return to the same intake form, the same approval chain, the same campaign calendar, now holding skills the workflow gives them no room to use?

Picture the concrete version. A marketer learns to generate 40 email variants and 12 audience-specific landing pages with a GenAI assistant in an afternoon. Then they open the CDP and find the segments those variants are meant to serve were never built, and the MAP still routes every send through a legal review queue that takes 9 days. The skill is real. The system around it never moved. So the 40 variants die in a queue, and the marketer goes back to the campaign cadence the system always allowed.

Who decides what an AI agent is allowed to do without a human signing off? If that question is open, your newly trained team keeps asking permission for everything, which is the process they already ran.

Does the data these skills depend on exist, clean and reachable? Or did you teach people to drive before anyone laid the road?

Here is the part nobody wants to hear. Doing this work first is slower and less visible than launching an academy. An academy photographs well. It has a launch date, a cohort, a hashtag. Rewriting decision rights and rebuilding data foundations is invisible, political, and slow, and it produces no photograph. Most leaders pick the photograph and call it progress.

BCG’s own read points the same direction: the CMOs pulling ahead invest in data foundations, orchestration, and talent they build themselves. [1] The talent is the part everyone copies, because it photographs. The infrastructure that makes the talent usable is the part that gets skipped.

What, exactly, are you training people to do?

There is a difference between training people on tools and training them into work that has been rebuilt around those tools. The first is software literacy. The second is capability. Programs sell the second and deliver the first.

A test cuts through it. Name one thing a trained marketer on your team does on Monday that an untrained one couldn’t do the Friday before.

If the honest answer names a tool, “they can run the new assistant,” you funded literacy. If it names an outcome the business can see, “they ship a segmented campaign in a day instead of two weeks, and an approval path exists that lets it go live,” you funded capability.

Most CMOs reporting end-to-end transformation would, pressed for that one sentence, describe a tool. That is the 42% hiding inside the 96%. [1]

What happens to the capability the day the academy closes?

The hackathon ends Friday. Monday, the trained marketer opens a workflow that hasn’t changed, reports to a manager measuring the same things as before, and works on a stack governed by rules written before anyone knew what an agent was. How long before the skill goes cold?

Who governs what the agents do once they run on their own? A trained team operating autonomous agents with no governance is a liability the organization hasn’t priced.

And the tools move under you. The model your team trained on in Q1 is deprecated by Q3. If the learning had an end date, so does the capability.

This is why upskilling behaves like maintenance: ongoing, never finished, funded every year rather than celebrated once. The achievement is the standing system that keeps capability current as the work and the tools change. The academy was just the kickoff.

The question to take into your next budget meeting

Before you approve the next training spend, answer the deployment question first. What redesigned work, with what decision rights, on what data, do these people walk into the day they finish?

If you can answer in a sentence, fund the training. It has somewhere to go.

If you can’t, the budget buys the illusion BCG named, one cohort at a time, dressed as transformation.

The survey says the talent doesn’t exist and you have to build it. Fair. Build the place to put it first.