So far, AI has not helped any major company outside the technology sector achieve a measurable increase in profits. This is the conclusion reached by Torsten Slok, Chief Economist at Apollo Global Management, in a recently published note. Comparing the seven major US technology companies with the rest of the S&P 500, he finds that profit margins and revenue per employee have stagnated. He also sees no evidence so far that AI has increased productivity growth.
Slok argues that while AI is highly capable when it comes to calculations, intelligence in business is primarily about judgement. This is highly context-dependent and can lead to different outcomes depending on the person or AI model being asked. In particular, AI lacks an understanding of accountability, trust, corporate culture, unwritten rules, ethical trade-offs and fairness. Yet all of these factors influence reputational risk and, ultimately, revenue growth and profit margins.
Academic research also suggests that the expected productivity gains have so far remained limited. A 2025 study by the National Bureau of Economic Research, for example, estimated that AI saved around 3% of working time, without finding any significant effect on income or hours worked.
So why has the use of AI so far translated into so little measurable productivity growth? One possible explanation is that AI does not simply speed up individual tasks. It also changes workflows, creates additional tasks and increases the amount of oversight required.
A survey commissioned by Monday.com suggests, for example, that managers increasingly struggle to distinguish between work produced by employees themselves and work generated by AI. Almost half of respondents therefore spend more time reviewing results and supporting their teams. At the same time, 46% of managers use AI to support decision-making and 39% use it to prepare performance reviews. This creates a risk that AI may reproduce existing human thought patterns and biases from its training data, while the context-sensitive judgement emphasised by Slok receives too little attention.
A study published in the Harvard Business Review also suggests that AI can increase workloads rather than reduce them. Employees complete tasks more quickly, take on additional work and sometimes extend their working hours on their own initiative. Initial efficiency gains can therefore be accompanied by greater work intensity, exhaustion and declining decision quality. Over the longer term, there is a risk that apparent productivity gains could actually undermine performance.
What matters, therefore, is not simply how often AI is used, but how it is integrated into existing workflows. This includes clear rules for its use, deliberate periods of reflection and critical review of AI-generated results. Especially when AI speeds up individual stages of work, it is important to leave enough room for independent assessment and decision-making.
What does this mean for the use of AI in professional research?
AI can also make many stages of research easier and faster, including information retrieval, analysis and the presentation of results. Genuine productivity gains, however, will only materialise if the time saved is not used solely to take on additional tasks. At the same time, systematic quality assurance and critical evaluation of results should be used to assess continuously where AI genuinely adds value. It would therefore be too simplistic to attribute the apparent lack of productivity gains solely to “AI”. The more important question is how people use it and, closely related to this, how work processes are deliberately redesigned.
By Ida Kandler