A recent survey by the Boston Consulting Group (BCG) shows that while AI has improved efficiency within enterprises, many companies are not truly translating the time saved into business output. The more direct problem lies not in the tools themselves, but in the lack of guidance from management on how AI should be used and where the saved time should be invested.

42% of employees save 8 hours per week

BCG's "2026 Global Workplace AI Report" surveyed nearly 12,000 frontline workers. The results showed that 42% of respondents said that regular use of AI could save them about 8 hours per week, equivalent to one workday.

However, increased efficiency did not automatically translate into higher productivity. 66% of respondents said their companies provided almost no clear guidance on how to use the saved time; the other half said that the time was not invested in more strategic work.

David Martin, Global Head of People and Organization at BCG, told Fortune that many senior executives still struggle to clearly articulate their AI strategies and goals. This amplifies employee anxiety and slows down subsequent adoption and collaboration.

Businesses should first track usage and then calculate costs.

The article notes that in the past, many tech companies have focused on "increasing AI usage as quickly as possible," but this approach is exposing cost issues. Due to high model running costs and the continued increase in expenditures driven by token-based billing models, the growth rate of AI costs for enterprises has exceeded the output improvement in some scenarios.

The report cites several company examples, stating that Microsoft has reportedly canceled direct licensing of some of Claude coding tools; Uber used up its entire year's AI coding tool budget in the first four months of 2026. Microsoft's AI head, Mustafa Suleyman, also stated this week that Anthropic's services are too expensive and the company is looking for alternatives.

Bryan Catanzaro, NVIDIA's VP of Applied Deep Learning, also stated that for his team, computing costs have exceeded employee costs.

The incentive for "token farming" is starting to fade.

Amidst companies' increasing demands for employees to use AI more, some have even made usage volume itself a key performance indicator. The Financial Times reported last month that Amazon employees were engaging in "tokenmaxxing," meaning they were using models as much as possible to meet internal AI metrics. Companies like Meta have also adopted similar practices, such as creating AI usage leaderboards.

But this incentive approach is waning. Martin believes that companies previously distributed AI tools to almost everyone without differentiating between job roles or establishing clear business return standards. Now, companies are reassessing: who needs the permissions, whether the investment is worthwhile, and whether the goals are being met.

The Financial Times also reported that Amazon discontinued its internal AI usage tracking last week because some employees were using AI bots to complete tasks with no real value. Amazon executive Dave Treadwell reportedly told employees not to use AI for the sake of using AI.

Employees worry about delays in implementation

Besides cost, employee concerns about job replacement are also impacting AI adoption. Martin stated that if companies describe AI agents as "digital employees" rather than tools, employees are more likely to feel pressured to be replaced. This reduces team sharing, increases the use of AI privately, and ultimately slows down the organization's overall progress.

He believes a more effective approach is to integrate AI into the company's overall operating model, along with systematic training. Employees who are more aware of the limitations of tools and are better able to use new tools are generally more willing to share their experience and resources.

The article also cites the views of executives from Okta and Rakuten International, stating that many companies currently do not lack the slogan of "adopting AI," but what they truly lack is adjustments to their organizational structure, responsibility allocation, and management methods. AI can already save time, but whether companies can turn that time into new output still depends on whether management provides a clear path.