Article written by Dr Shaheen Khan (Senior Academic & Programme Co-ordinator at Regent Business School)
Artificial Intelligence (AI) is no longer an emerging technology confined to research laboratories or technology companies. It has become an integral part of managerial decision-making across organisations worldwide, including South Africa. According to PwC’s Africa Workforce Hopes and Fears Survey (2025), 91% of South African senior executives who used AI during the previous year reported that it had improved the quality of their work, while 64% of employees across Africa reported using AI in the workplace, exceeding the global average of 54%. These statistics demonstrate that AI has moved beyond experimentation and is becoming embedded in everyday business practice.
What has not kept pace, however, is the development of AI literacy, governance and critical evaluation skills required to ensure that AI-assisted decisions remain accurate, ethical and accountable.
The public debate surrounding AI often centres on whether it will replace human jobs. While this remains an important discussion, it overlooks a more immediate organisational risk. Managers are making strategic, operational and financial decisions based on AI-generated recommendations without fully understanding how those recommendations are produced, their inherent limitations or the circumstances under which they may be inaccurate. This is not primarily a technology problem; it is a management problem.
The Technology Acceptance Model (TAM), developed by Davis (1989), provides a useful explanation for why AI has been adopted so quickly. As discussed in my doctoral research, TAM proposes that individuals are more likely to adopt new technologies when they perceive them to be useful and easy to use (Davis, 1989; Lai, 2017; Granić & Marangunić, 2019). Subsequent developments of the model by Venkatesh and Davis (2000) and Venkatesh and Bala (2008) further demonstrate that behavioural intention is strengthened by organisational context, social influence, experience and facilitating conditions. These developments have made TAM one of the most influential theoretical frameworks for explaining technology adoption across multiple industries. Generative AI satisfies these conditions exceptionally well.
Managers can generate reports, analyse financial data, draft policies, develop marketing strategies, summarise research, write presentations and produce business correspondence within seconds using natural language prompts. For organisations operating in competitive environments, AI offers immediate productivity gains while requiring relatively little technical expertise. From a TAM perspective, its widespread adoption is therefore unsurprising. However, TAM explains why managers adopt AI, not whether managers understand AI.
Large Language Models (LLMs) such as ChatGPT do not reason in the same manner as experienced managers or subject matter experts. Rather, they generate statistically probable responses based on patterns learned from vast quantities of training data. They do not possess organisational memory, contextual understanding, intentionality or professional accountability. Although AI frequently produces convincing outputs, those outputs remain susceptible to hallucinations, incomplete reasoning, outdated information and inherited biases within the underlying training data.
Consequently, AI should not be viewed as an autonomous decision-maker but rather as a sophisticated decision-support tool.
This distinction is reinforced by Herbert Simon’s theory of bounded rationality. Simon (1957) argued that managers rarely make decisions under conditions of complete information. Instead, they operate within environments characterised by uncertainty, competing organisational priorities, stakeholder expectations and limited time. Rather than identifying theoretically optimal decisions, managers use professional judgement to arrive at decisions that are practical, ethical and strategically appropriate.
AI can dramatically improve the availability and processing of information, but it cannot replace managerial judgement.
Strategic decisions concerning acquisitions, restructuring, regulatory compliance, organisational culture or corporate reputation require contextual understanding that extends beyond pattern recognition. They require experience, ethical reasoning and accountability, qualities that remain uniquely human.
Research into automation bias further highlights why organisations should approach AI adoption with caution. Parasuraman and Riley (1997) describe automation bias as the tendency for individuals to place excessive confidence in recommendations generated by automated systems, particularly when those systems have previously demonstrated high levels of accuracy. Similarly, Goddard, Roudsari and Wyatt (2012) found that decision-makers frequently overlook contradictory evidence once an automated recommendation has been presented.
AI will occasionally produce incorrect recommendations. The real risk is that managers gradually lose the habit of questioning the ones that look plausible. This represents an important limitation of existing technology acceptance theories. Managers may readily adopt AI because it is useful and easy to use while simultaneously lacking sufficient AI literacy to critically evaluate the outputs it generates. Technology acceptance should therefore never be confused with technology understanding.
Stakeholder Theory provides an additional perspective on this challenge. Freeman (1984; 2004) argues that organisations exist to create value for multiple stakeholders to maximise organisational efficiency alone. Donaldson and Preston (1995) further demonstrate that effective management requires balancing competing stakeholder interests through informed managerial judgement rather than mechanistic decision-making.
AI can undoubtedly assist managers by analysing stakeholder information, identifying trends and evaluating alternative scenarios. However, it cannot determine which stakeholder interests should take precedence within a particular organisational context. Nor can it assume responsibility for the ethical or legal consequences of those decisions. That responsibility remains with managers.
For South African organisations, the question is no longer whether managers will adopt AI. That has already been answered. What needs to be addressed now is whether organisations are developing managers who understand both the capabilities and limitations of AI, which calls for three practical interventions.
First, AI literacy should become a core component of management education. MBA programmes, executive education initiatives and postgraduate business qualifications should equip future leaders with an understanding of how Large Language Models generate outputs, where their limitations lie and how AI-generated recommendations should be critically evaluated. Managers do not need to become data scientists, but they must become informed users of AI.
Second, organisations should establish Human-in-the-Loop governance. High-impact decisions involving finance, recruitment, healthcare, legal compliance, strategy or enterprise risk should require documented human review before implementation. AI should support managerial decision-making rather than replace managerial accountability.
Third, boards should recognise AI governance as a strategic organisational risk. Just as boards oversee cybersecurity, enterprise risk management and ESG performance, they should also receive regular reporting on AI governance frameworks, executive AI capability, AI-related risks and organisational safeguards. AI governance should sit within corporate governance as a whole not be treated as an information technology matter alone.
AI is undoubtedly one of the most significant technologies of this generation. Organisations that fail to embrace it risk falling behind competitors. However, organisations that embrace AI without simultaneously investing in governance, critical thinking and managerial capability may expose themselves to an entirely different form of competitive disadvantage.
AI can generate answers in seconds. Producing sound judgement still takes longer, and it still depends on people. For South African organisations, that is where the competitive advantage will sit: with managers who know when to trust AI’s recommendations, when to challenge them and when to override them altogether. Information is now abundant and easy to reach. What remains scarce is the discipline to think critically about what it produces.
