In the second half of last year, the operations team at a midsize Korean retail company rolled out an AI writing tool across the organization. The company ran three rounds of internal training, with a completion rate of 89 percent. Six months later, the team lead asked the room: "Who here is actually using this tool in their day-to-day work?" Three hands went up—out of seventeen. Between the training completion rate and the actual usage rate sat a gap of nearly 70 percentage points.
I've been hearing versions of this story a lot lately. The adoption budget got spent, the official training got delivered, the showcase use cases got documented—and six months on, the way the team works hasn't meaningfully changed. Is it strange that a tool arrives and the organization stays the same? Or was the expectation that a tool would change the organization overblown from the start?
Take that question seriously and you run into a fundamental misconception about AI adoption. And once you understand where that misconception comes from, where to spend your AI budget first starts to look very different.
Between adoption and actual use lies an invisible distance
In its 2024 global survey on the state of AI, McKinsey reported a figure worth pausing on. Sixty-five percent of responding companies said they had deployed at least one AI solution in their operations. Yet among them, no more than 30 percent had seen tangible financial results. That's a gap of more than two to one between adoption and payoff. The report attributed the gap less to the limits of the technology than to the problems organizations run into when embedding AI into actual work—workforce redeployment, process redesign, and the absence of systems for measuring results came up again and again.
The situation in Korea isn't far from the global data. Starting in 2023, generative AI tools spread rapidly through Korean companies. From executives at large conglomerates to team leads at small firms, people began writing internal usage guidelines, appointing owners, and opening channels to share use cases. Procedures emerged for quarterly check-ins and tallying usage counts. From the outside, adoption looks like it's proceeding on schedule.
But even in organizations with all these procedures in place, talk to frontline staff and the usage pattern is remarkably consistent: a small minority who use AI actively, coexisting with a majority who signed up but rarely touch it. And often, the majority's reason for not using the tool isn't that the tool is clunky. It's that the gain from using it doesn't yet feel bigger than the friction of changing how they already work.
What creates this distance is usually not the quality of the tool but the environment it lands in. Even if AI drafts the report, if the approval chain stays the same and the review criteria don't budge, workers end up running the old process alongside the AI draft. Everything gets checked twice, duplicated work piles up, and usage quietly tapers off.
The case for "buy the tool and the organization will follow"—and its limits
AI optimists push back on all this. A sufficiently powerful tool, they argue, beats organizational inertia in the long run. When spreadsheets arrived, companies started abandoning paper ledgers; when email spread, fax-dependent workflows died out. Give it time, the argument goes, and AI will restructure how work itself gets done. Low usage in the early phase is simply the natural early stretch of the adoption curve, and once key competitors start raising productivity with AI, everyone else will have no choice but to follow. Some leading companies do report cutting the time spent on certain documents by 30 to 40 percent.
This rebuttal has real historical grounding. Spreadsheets and email broke through early resistance and remade entire ways of working. Predicting that AI will travel the same path is not an unreasonable inference.
But there's a difference between how spreadsheets and email changed organizations and the nature of today's AI tools. Spreadsheets and email were structured so that not using them made collaboration itself difficult. Once the whole team works in Excel, the one holdout keeping books by hand can no longer share data. Once email becomes universal, the person communicating only by memo slips drifts out of the information flow. The cost of refusing to change kept rising.
In most organizations, today's AI tools exert no such pressure. A colleague writing reports with AI doesn't prevent me from sticking with my old method. Used well, the tool is faster and more convenient—but skip it and you're not excluded from anything. Under those conditions, change depends far less on any current the technology generates by itself, and far more on deliberate organizational design.
Evaluation structures and tool usage move together
HR researchers who have spent careers studying how organizational change happens converge on one point: for a new way of working to take root, individual willpower isn't enough. Role definitions, evaluation systems, and decision-making structures have to shift along with it. If AI usage isn't part of performance reviews, if the report templates are unchanged, if the approval steps are identical, then nothing in the structure gives workers a reason to use the tool any differently than before it arrived. Inertia isn't a bad habit—it's a product of structure.
This is where the role of middle managers and solo product leads comes into play. When large companies evaluate AI adoption, they tend to focus on IT budgets and tool selection. But whether a tool survives inside a team comes down to how the manager places it in the team's daily life. Does the team make AI drafts the default for meeting prep? Are AI-written drafts reviewed and given feedback in the open? Small choices like these, one by one, determine whether the tool actually lives or dies.
If you want AI usage to rise, there's something to check first: within your team, is there any difference in evaluation or recognition between the people who use AI and the people who don't? If there isn't, then tool usage remains purely a matter of personal taste. When finishing work faster with the tool is never made visible in any way, the tool stays the property of an enthusiastic few.
Managers visibly using the tool themselves matters more than you might expect. When a team lead shares an AI-drafted agenda the day before a meeting, or mentions that a feedback memo was written with AI's help, team members register that this is an acceptable way to work. Someone has to send the first signal that using the tool isn't awkward.
For solo operators and small teams, this kind of design can happen much faster. Because you are both the evaluator and the executor, you can simply decide that certain tasks start with an AI draft by default. Not "I'll use it when I can," but "this type of document begins with an AI draft"—apply that rule to yourself and your usage rate changes. I tend to see this less as a tool-usage problem than a routine-design problem. Structure has to be built deliberately, at any scale.
The expectation that AI will change your organization isn't wrong in itself. But if you assume the change starts automatically the moment you buy the tool, you'll find yourself in the same place six months later. For a tool to survive inside a team, someone has to deliberately place it within the structures of everyday work. Planting a tool in a team takes far more handwork than clicking a subscribe button—and for now, that work remains squarely human.
