Everyone in Marketing Is Trying AI — But Nobody Tells You About the Step That's Actually Hard

A while back, I was chatting with a few friends who work in marketing, and I noticed something genuinely interesting.

Every company is buying AI tools. Every company is pushing its team to get on AI. Every company is chanting at the weekly meeting: "We're going AI-first!"

But ask them one question: what work has actually been replaced by AI?

Silence.

"Tried it" and "actually using it" are two completely different things

There's a set of data that hits close to home. A time-tracking company ran a study and found that marketing people spend roughly twice as much time on AI tools as people in other roles.

Sounds like marketing is moving fast, right?

But a more substantial report cut straight to the truth: everyone is trying it, yet at most companies, once the trial phase is over, the work still gets done the same old way, the process is still the same old process — there's just one more tool sitting on the desk, gathering dust.

What does "tried it" mean? Tried it means somebody opened it a few times, thought "wow, that's amazing," and then went back to doing things the old way.

What does "actually using it" mean? Actually using it means your team starts rethinking a question from scratch: how should this piece of work really be done, the right way?

Between those two things lies an entire transformation journey.

And the hardest step on that journey isn't which model to pick, or how many accounts to buy.

It's changing how a team thinks, how it picks up work, how it reviews copy, how it delivers.

Get people wanting to move first — then we can talk training

When I drove this inside my own department, the first thing I figured out was this: don't start by scheduling training courses.

People need a reason to want to use it first.

How? In our regular weekly meetings, I'd carve out a few minutes and let the people who'd actually used it talk: which process saved time, which little research trick pushed the quality of a draft up, which agent was almost usable but still one step short.

Don't just talk about prompts. Talk about concrete scenarios. Talk about the pitfalls you stepped in. Talk about the little experiments other people could copy.

After they speak, recognize them publicly. Make sure the team knows that trying something imperfect is safe — and encouraged.

There's one more thing. It looks small, but it's make-or-break. You have to strip away, one by one, the little obstacles quietly dragging everyone down.

For example, the accounts you bought don't have enough tokens, so everybody winces with every sentence they type. In that state, nobody dares to experiment.

You need to buy enough. Let people experiment freely, without worrying about wasting tokens when they get it wrong.

Give it three months, and you'll have a handful of genuine seed champions and a few real, concrete project wins under your belt. That's when you take the best cases and lock them in as standard processes and automation.

At that point, you're not forcing a new process on anyone. You're formally confirming the good things people are already using.

It feels completely different.

Learn it Monday, don't use it by Friday — it's as good as never learning it

My team has one ironclad rule: every training course must come paired with a real project, and that project must kick off that same week.

Why so strict?

Because humans, as a species, lose most of what they learn if they don't use it within a week.

We set a minimum learning path — three courses: one on how to talk to AI, one on how to use the skills of agents, and one on the underlying cognitive framework of AI.

These three courses solve one foundational problem: the team at least shares a common language — how to give AI a task, how to break a piece of work down cleanly, and which parts a human absolutely has to watch.

Once the basics are done, each role digs deeper into whatever it needs most.

But the core comes down to one line: what you learned this week has to turn into a deliverable this week.

Learning without delivering doesn't count.

If the process is a mess, AI just magnifies the mess

This is a lesson a lot of people don't want to hear.

If your process is already a disaster, AI won't clean it up for you. It'll take the existing chaos, blow it up tenfold, and serve it back to you ten times faster.

So before you bring in AI, do one thing first: look at where your team's time is actually going.

We use time-tracking data to see how the team's hours really flow each week. Not gut calls — evidence. Then, based on that evidence, we pick the few processes most worth operating on.

Every quarter, make a short list. For each process, write down three things clearly: what it looks like today, what it should look like, and who owns it.

You'll notice the team's attention quietly shifts from "where can we use AI" to "which bottleneck can AI help us break through."

That's the question a CMO should actually be pushing the team to wrestle with.

Leaders model it from the top, champions emerge from the bottom — push from both ends

Transformation can't be driven from the top alone — it stalls. It can't survive on grassroots enthusiasm alone either — it won't last. You have to push from both ends at once.

What does the top do? Leaders have to be visible practitioners themselves. I use AI to write, to build agents, to produce first drafts — and then I drop those into the team channel. Not so they copy me, but so they see a signal: this isn't just talk. I'm in the trenches too, and I might not even be doing it better than you.

And the bottom? You have to find the people on the team who are genuinely lit up. The ones willing to tinker on their own time, who can't resist sharing the moment they find something that works.

These people — praise them publicly, give them time on the team meeting to present. They'll set the standard for "what good AI practice looks like" on your behalf.

My goal is that by year-end, this group turns the best of what they've learned in the field into an in-house marketing AI playbook that belongs to our company. It'll have prompts, processes, review rules, the logic behind agents, the rules for how to use data, and a record of every pitfall we stepped into.

That playbook is worth more than any course on the outside. Because what it captures is how your company actually gets work done.

When you hire a consultant, don't hire a generalist

At a certain stage, you might need an outside consultant.

But don't hire a generalist who's "built every kind of agent." The person you want has to genuinely understand the marketing tech stack, understand what marketing processes actually look like, and know what kind of daily grind the marketing team faces.

Ideally they've pulled off a few of these things: optimized a marketing workflow, closed the gaps between tools, built automations that actually run, and helped a team go from "flailing in trials" to "able to repeat it."

Plain and simple, you don't want somebody who can build agents. You want somebody who's solved problems just like yours.

What you choose to measure decides where everyone will pour their effort

Last, let's talk KPIs. This is the easiest place to fool yourself.

If your metrics are "how many prompts were run" or "how many hours of training," then everyone will absolutely game those numbers. And once they're gamed, the work is still the same old work.

Vanity metrics give you a fake feeling of winning.

The scorecard I use looks roughly like this:

Training — the pilot team completes the foundational courses at 100%.

Real-world delivery — after everyone finishes learning, each person ships at least one project genuinely built with AI.

Process inventory — the most common core processes are mapped out cleanly and prioritized.

Consultant landing — at least two hands-on build sessions are completed.

Agents in production — at least three agents are live in a production environment.

Efficiency — the top three daily grind tasks have their time cut by sixty percent.

Engagement — over seventy percent of the pilot team is genuinely using it each week.

Report delivery — the turnaround on campaign retrospectives is cut in half.

Output — content output goes up fivefold, without adding headcount.

Quality — eighty percent of AI-assisted output needs only light edits to ship.

Playbook — one reusable marketing AI playbook, captured and locked in.

Business impact — AI's contribution makes it into the year-end ROI retrospective.

See? This set of metrics forces everyone to do the right things: cut the manual toil, produce more of what's actually useful, and ultimately prove the results with data.

The single most important line

Why does AI transformation fail?

Because it's sitting at number eleven on your ten-item priority list.

When does it succeed?

When the CMO actually carves out time, budget, and training resources for it — and treats "the first three agents that can run" as a real deliverable, not a side project for spare time.

Spark curiosity, build confidence through real-world practice, clean up the processes, nurture the champions who step up, and set the KPIs clearly.

A marketing team grown this way isn't "using" AI.

It runs on AI, end to end.