Half Your Team Is Already Using AI. Your Adoption Strategy Doesn't Know It.

A management team tells us their AI adoption is stalled. Usage on the company's licensed AI tool sits around fifteen percent. Training was delivered. A champion was appointed. And still, most people are not touching it.

Then we talk to the people doing the actual work. Half of them are using AI every week. Just not the tool leadership picked.

They are pasting client emails into a personal ChatGPT account to draft replies faster. They are using Claude on their phone to summarize a contract before a meeting. One operations manager has built a set of prompts in a personal notes app that she reuses every month for reporting, and has never told anyone about it. None of this shows up in any dashboard. As far as leadership can tell, adoption failed.

It did not fail. It went underground.

What is actually happening

This is not rare. It is close to the default state in most organizations right now. People adopt AI the same way they adopted spreadsheets and search engines before anyone approved a rollout: quietly, individually, because it solved a problem they had that day. The official tool, chosen by a committee and rolled out with a training session, competes with a personal account that already fits into someone's actual workflow. The personal account usually wins, because it asks nothing of the person except to keep using it.

The result is a strange split. Individual capability is often higher than leadership assumes. Organizational capability stays at zero, because none of that individual capability is visible, shared, or governed. Nobody knows which prompts work. Nobody knows which use cases are already validated by someone doing the job every day. And nobody has looked at what data is quietly leaving the building through a personal account with no enterprise agreement behind it.

The reframe

The real adoption problem in most companies is not resistance. It is invisibility. Leadership is trying to build adoption from zero, when the more accurate starting point is: adoption already exists, scattered across individuals, undocumented, and unmanaged. Training programs designed to convince people to start using AI often land badly, because the room already contains several people who have been using it for months and are being told, in effect, that their experience does not count.

The MAKIA lens is useful here. The people already experimenting are your Actors. What they have learned, informally, is Knowledge your organization does not yet own. Until you connect the two, no amount of top down Meaning setting will move the needle, because you are solving a problem that is not the one your people actually have.

What this means in practice

First, stop measuring adoption by license usage. A low number on the official tool tells you almost nothing about how much AI use is happening in your company. Ask people directly, in conversation, not in a survey nobody reads.

Second, treat your quiet power users as a resource, not a compliance problem. The instinct when you discover unauthorized tool use is to shut it down. Before you do, find out what they were solving for. That use case is often the most validated one in the building, because someone kept doing it without being told to.

Third, build the bridge from individual habit to shared capability. A prompt that works for one person in finance is worth documenting for the rest of finance. This is where most of the real value of AI adoption sits, not in the tool choice, but in turning private workarounds into shared workflows that survive when the person who built them changes roles.

Fourth, address the governance risk honestly rather than pretending it does not exist. If people are pasting client data into personal accounts, that is a real exposure. The fix is rarely a ban, which just pushes the behavior further out of sight. It is usually a sanctioned, well supported alternative that is at least as convenient as what people are already doing.

A question worth sitting with

Before the next AI training session gets scheduled, it might be worth asking a smaller question first: how many people on this team have already, quietly, figured something out. And what would it cost the organization to finally ask them.

Try this week

Pick five people on your team from different roles. Ask each of them, one on one, a direct question: have you used any AI tool for work in the last two weeks, which one, and for what. Do not survey them, talk to them directly. Write down what you hear, without judgment. That conversation, not the usage dashboard, is your real adoption baseline.