Hidden supervisory burdens eat productivity gains, survey of 1,700 leaders reveals employee evaluation crisis and software fragmentation

Outputs reach managers' inboxes polished and free of glitches, but this sleek appearance now hides a growing administrative dilemma. A survey conducted by Nielsen on behalf of the Mandi.com platform, involving 1,700 senior executives and leaders across seven countries, found that more than half of leaders, over 50 %, experience severe difficulty evaluating their employees' performance after AI tools were introduced. The core issue is the blurring line between pure human effort and what the machine produces, which has led roughly half of managers to spend more time reviewing and auditing their teams' outputs than they previously did.

This confusion does not reflect a waning enthusiasm for technology, but rather reveals a widening gap between the rapid evolution of models and the sluggishness of corporate processes. Survey results show that 94 % of participating managers feel enthusiastic about advancing their businesses with AI and are confident in pinpointing where to use it for maximum return. However, Michelle Miller Levy, head of market research and customer insights at Mandi.com, a platform, explained that the situation resembles the early days of the Internet at the turn of the millennium, when everyone was dazzled by the tool’s capabilities individually, while leaders now face a different challenge: steering the entire organization amid an innovation pace that outstrips the ability of organizational structures to adapt.

Administrative rules of the game have changed in practice78 % of respondents said they have adjusted employee evaluation criteria after adopting AI, giving greater weight to skills in drafting directives, the ability to delegate tasks precisely to intelligent systems, and experimenting with new work methods. Yet the paradox emerges as management turns to the same technology to judge teams: 46 % of managers use AI to support their decisions, while 39 % of leaders rely on generative models to draft annual performance evaluation reports, a practice that raises genuine concerns about replicating biases embedded in training data. The Luxembourg Institute of Social and Economic Research, a research institute, documented in a study earlier this year that language models, in the absence of clear objective standards and when evaluating individuals independently, tend to reproduce patterns of human judgment. A study published in early 2025 showed an explicit bias in models against specific groups such as male candidates of dark skin and marked disparities among different social groups.

At the level of daily work, auditing intelligent systems has become what the report describes as a new layer of “hidden work,” where selecting tasks, reviewing outputs, and intervening on failures consumes additional hours that were not anticipated. In parallel, expectations of software bundle reduction have fallen short; 57 % of managers said their companies are seeing more digital tools in circulation than before, 51 % observed that their tools have become more interconnected and intertwined, while only 29 % described their infrastructure as centralized and optimal. Levy confirms that the bet on reducing software resulted in expanded unauthorized usage and an accumulation of applications, with more than half of leaders acknowledging that software bundles are on a continuous upward trajectory.

For chief technology and executive officers in the Gulf, Egypt and the broader Arab region, where the adoption of intelligent transformation is accelerating unusually, this survey provides a critical indicator for recalculating operating costs. The true cost of automation is not limited to subscription fees and cloud infrastructure, but also includes the human review hours lost in sorting polished outputs and curbing the random use of tools within departments. Unleashing independent agents inside banks or regional service firms without strict governance frameworks shifts the administrative burden into the unknown, especially as 58 % of managers worldwide anticipate supervising intelligent agents that will outnumber human employees within a year, at a time when 46 % fear losing control and 35 % express concern about bearing legal and corporate responsibility for unexpected automated decisions.

The field summary is clear:Automation success is not measured by the volume of texts and reports flowing into your inbox, but by your organization’s ability to keep human decision-making at the helm, unify work platforms to prevent software chaos, and establish transparent standards that separate employee competence from the capabilities of the tools they use.