An adoption percentage can combine occasional users, power users, sanctioned tools, shadow AI, and sharply different spending patterns. The number may be accurate, but it still may not tell leaders what they need to know.
What an AI adoption rate tells you depends on how the organization defines it.
A license-based rate shows how many employees have access. A login rate shows who opened a tool. A weekly active-user rate shows who met the organization’s activity threshold during that week. Each measure can be useful, but none provides a complete picture of adoption on its own.
Which employees are included in the calculation matters too. A rate based only on employees assigned AI licenses answers a different question than one calculated across the full workforce. A single-vendor dashboard also can’t show activity in other enterprise tools, personal accounts, or embedded AI features it doesn’t track.
That doesn’t make the adoption rate wrong. It makes the label and measurement scope essential. Leaders should be able to explain exactly who and what the percentage includes before using it to make budget, training, or governance decisions.
In one Larridin customer environment, 92.4% of employees tracked were active AI users during the week. One engineer still generated 65% of the organization’s AI spending during that period.
Both findings can be true. The organization had high weekly participation and highly concentrated spending. The adoption rate alone couldn’t show how usage and cost were distributed.
Engagement can vary just as widely. In another view of the same customer data, employees using ChatGPT in a web browser averaged 18 events per user over four weeks, while those using the installed desktop app averaged 130. The roughly 7x difference shows why counting users can hide substantial variation in how often they engage.
That channel difference doesn’t show whether desktop or browser use was sanctioned, productive, or risky. It’s a signal that leaders should investigate the account, tool, tasks, data, cost, and outcomes behind the activity.
Higher engagement can be valuable. It can also generate waste, rework, excess cost, or governance exposure. Low-frequency use can still create risk when an employee enters sensitive information into an unapproved tool.
That’s why adoption breadth and depth need to be paired with two separate measures:
A highly proficient employee working through approved systems may have practices worth scaling. A high-volume user producing weak outputs may need workflow changes or stronger review. An occasional user working outside approved channels may need access to an approved tool or clearer guidance.
Usage depth helps leaders find the questions, but it doesn’t answer them.
Writer’s 2026 survey of 2,400 global employees and C-suite leaders who use AI at work found significant organizational strain around adoption. Among C-suite respondents, 54% said adopting AI was tearing their company apart, while 67% of executives believed an unapproved AI tool had already contributed to a data leak or security breach.
Those findings don’t prove that a high adoption rate causes conflict or security incidents. The survey also doesn’t independently verify the reported breaches. It does show that leaders perceive major governance and organizational problems even as AI use becomes widespread.
Writer also found that 92% of C-suite respondents were cultivating an “AI elite,” while 60% planned layoffs for employees who couldn’t or wouldn’t adopt AI. Those are reported management intentions, not a measure of actual usage distribution. But they raise the stakes for getting adoption measurement right. A vague percentage is a weak foundation for decisions about workforce investment, performance, or employment.
A board-ready adoption view should answer six questions:
These measures should be examined together. A team with high adoption and low proficiency needs a different response than one with low adoption but strong results among a small group. A team with high adoption through sanctioned tools needs different support than one relying on shadow AI.
The goal is to stop asking one percentage to answer six different management questions.
It tells you that a high share of the measured population met the organization’s chosen adoption threshold. Its meaning depends on the definition, time window, population, tools, and data sources included.
No. Adoption measures participation or activity. It doesn’t automatically establish proficiency, business impact, governance compliance, or appropriate spending.
Track engagement frequency, feature and task breadth, workflow integration, and consistency over time. Then connect those patterns to proficiency and business outcomes rather than treating more activity as proof of value.
Not automatically. Risk depends on the tool, account, data, workflow, and controls involved. Deep use through approved systems may be both productive and well governed. Even occasional use of an unapproved tool can create exposure.
It should clearly define the adoption rate and show participation trends, engagement depth, proficiency, sanctioned versus unsanctioned usage, cost concentration, and business outcomes.
Larridin’s AI Adoption platform shows adoption trends across tools and teams, identifies differences in engagement, surfaces sanctioned and unsanctioned usage, and tracks cost. Pairing those signals with proficiency and impact data helps leaders separate participation from performance and visibility from assumption.
Book a discovery call to see what is underneath your AI adoption rate.