A wave of reports released in late January has delivered sobering news for business leaders who have bet big on artificial intelligence: most companies are seeing lots of activity but little improvement to their bottom line. Studies published in January 2026 by consulting giant PwC, along with tech companies Anthropic, OpenAI, and Google, paint a consistent picture. Workers are using these tools more than ever. However, the expected cost savings and revenue bumps have not shown up for most organizations. Majority of executives report no financial gains The numbers tell a blunt story. PwC’s 2026 survey of chief executives found that 56% saw neither lower costs nor higher revenue over the past year. Just 12% reported gains in both areas. That gap matters. Businesses have spent heavily on software licenses and training. The survey suggests the problem isn’t the technology but how companies are deploying it. Executives who did report financial benefits were two to three times more likely to have woven these tools deeply into their operations and customer-facing activities, rather than just handing out software accounts. Simply adding more users doesn’t translate to better financial performance. Companies need to redesign how work gets done, not just distribute new tools. So if counting active users doesn’t work, what should companies measure? Anthropic released findings on January 15 that propose tracking what it calls “economic primitives”, the type and difficulty of tasks people assign to these systems. The difference between task types matters. Having a system summarize an email requires little sophistication and saves minimal time. Delegating a complex, multi-step coding project represents genuine labor replacement. Anthropic’s research show s so ftware development requests average 3.3 hours of equivalent human work, while personal administrative tasks clock in at just 1.8 hours. Business managers need to look beyond simple headcounts of who logged in. They need to know what kind of work is actually getting done. High usage of trivial tasks equals wasted money. Focused use on complicated tasks equals real productivity gains. OpenAI’s analysis, publishe d Ja nuary 21, backs up this argument. The company identified what it calls a “capability overhang”, a mismatch between what these systems can accomplish and how people actually use them. Tw o fi ndings stand out. First, the heaviest users tap into advanced features, particularly sophisticated reasoning capabilities, seven times more often than typical users. Second, when OpenAI examined usage patterns across more than 70 countries, they found a threefold difference in how intensively people employ these advanced features. This creates a new competitive dynamic. Companies operating in regions where workers know how to leverage full capabilities will outperform rivals using the same software in less sophisticated ways. Digital literacy alone isn’t enough. Workers need what researchers call “agentic fluency”, the ability to delegate complex, multi-step tasks. Google’s January 20 update to its Workspace software addresses another measurement challenge. The business now shows comprehensive usage analytics, including which teams are using features and how frequently, directly in administrator dashboards. This change is important. It transforms spending into a category that finance departments can monitor and audit. The dashboard offers utilization data to support or refute a manager’s claim of increased efficiency. Five priorities for business leaders What actions should executives take differently? Five priorities are suggested by industry analysts: Stop confusing usage with value Audit projects for task complexity rather than user counts Put in place tracking systems that link usage to business outcomes Budget for workflow redesign rather than just software purchases Research the 12% of businesses that report real financial gains. Financial officials will likely need uniform reporting on profit-and-loss effects during the next three months. There will probably be competition among software suppliers to make their measurement techniques industry norms. Regulators may also request data on how autonomously these systems operate and what safety measures are in place. The message from this batch of studies is clear. The experimental phase is over. Companies now face pressure to show concrete returns on their investments. If you're reading this, you’re already ahead. Stay there with our newsletter .
