Every marketing leader I talk to has done the same thing in the last eighteen months. They bought seats. ChatGPT for the team, maybe Claude, maybe a writing tool with a marketing name on it. Then they waited for the output to change.
It mostly did not.
The work got a little faster at the edges. First drafts arrive sooner. Somebody automated the weekly report. But the team is producing roughly what it produced before, at roughly the same quality, and the leader is left with an uncomfortable question they cannot ask out loud: was any of this worth it?
The answer is yes, but not for the reason the tool vendors gave you.
Access is not adoption
Here is the thing nobody says plainly. Giving someone an AI subscription is like giving them a library card. It grants entry. It does not make them a reader, and it certainly does not make them a researcher.
What I see inside teams is a spread, not a level. In the same department, on the same subscription, you will find one person who pastes a sentence in and asks it to "make this better," and another who has built something closer to a small production line: a working brief, a voice guide, three rounds of structured revision, and a check against the actual customer research before anything ships.
Both of them have AI. One of them has a practice.
The spread is the real finding, and most leaders cannot see it. They see the department average, which is usually "we're using it a bit." Averages hide the two things you need: how far your best person has actually got, and how many people are still at the library card.
The three things that block a team
When a team is stuck, it is almost never because the model is not good enough. It is one of three things, and they fail in a specific order.
Ingredients. The model does not know your business. It does not know who you sell to, what they are afraid of, what your competitors say, what your last twelve campaigns taught you, or how you are supposed to sound. People who get good output are people who feed it good input, and the input is not a prompt. It is the customer research, the positioning, the voice, the offer, the proof. Most teams have some of this written down somewhere and none of it where the work happens.
Tools. The model can write. It cannot see your analytics, read your CRM, look at the live site, pull the search data, or publish anything. A marketer working in a chat window is working with one hand. The teams pulling ahead have connected the thing to the systems where the work actually lives, so it can look something up instead of guessing.
Process. This is the one people skip. A good output is not a lucky prompt, it is a repeatable sequence: what gets briefed, what gets checked, who reviews, what "done" means. Without that, every person on the team invents their own method every time, and the quality swings wildly depending on who happened to do the task.
They fail in that order because each one depends on the last. Better tools cannot rescue missing ingredients. A tighter process cannot rescue a model that has nothing real to work with.
Why your best person is the most useful thing you have
Somewhere on your team is a person who has worked this out on their own. They are usually not the most senior. They are just the one who kept going after the first disappointing answer.
What they have built lives on their laptop and in their head. When they leave, it leaves. When someone new joins, they start from zero again, and you get another six months of "make this better."
That is the actual cost of doing nothing. Not the licence fee. The fact that your one genuinely fluent marketer is a single point of failure, and their method is not available to anyone else.
The work worth doing is taking what that person does, writing it down, connecting it to the things they need it to see, and making it the way the department works. That is a system, and it is the only version of AI adoption that survives a resignation.
Where to start
Find out where people actually are. Not what they say in a meeting, and not the department average. The spread, task by task, for the fifteen or so things your team does every week.
You will usually find three things at once. A couple of people much further ahead than you expected. A cluster who have tried it, got a mediocre answer, and quietly stopped. And a set of tasks where nobody is using it at all, some of which are exactly the tasks it would be best at.
Once you can see that, the next move is obvious, and it is usually smaller than you feared. It is rarely "buy a platform." It is more often "write down what Maria does, give everyone the customer research in a form the model can read, and agree on what a finished brief looks like."
That is the whole job. It is unglamorous, and it works.