Artificial intelligence models are becoming more powerful, more accessible and less expensive to use. While this is good news for businesses seeking to adopt the technology, it also creates a strategic challenge: when every competitor can access the same models, access itself cannot provide a sustainable advantage.
The enterprise AI conversation has often focused on which platform, provider or model an organisation should select. However, emerging research suggests that the technology choice is rarely the main factor determining whether an AI programme succeeds.
Rand Corporation’s analysis of enterprise AI initiatives found that more than 80% failed to deliver their promised business value. The leading causes were not weak models. They included poor-quality data, unclear ownership, shifting project scope and a lack of alignment between the technology and the business problem it was expected to solve.
Gartner has reported a similar pattern, with only 28% of AI use cases fully meeting return-on-investment expectations. Skill shortages and data quality were among the leading causes of underperformance.
According to Francois van der Merwe, CEO and founder of South African startup Otinga, these findings point to a fundamental misunderstanding of where AI value is created. “Access to large language models has been commoditised. When access is commoditised, access cannot be your differentiator,” he said.
Organisations can now use sophisticated AI models through cloud platforms, enterprise software products and application programming interfaces. Capabilities that once required significant infrastructure and specialist expertise can, in many cases, be accessed within days.
This has encouraged businesses to place a thin AI layer over existing systems or introduce off-the-shelf assistants into unchanged workflows. These initiatives may produce immediate productivity improvements, but they are also easily replicated.
“The value does not lie in using ChatGPT or Claude off the shelf. It lies in what you build around the model,” Van der Merwe said.
A durable AI advantage therefore depends on the assets and organisational capabilities that competitors cannot easily reproduce. These include proprietary data, knowledge of customer behaviour, specialised operational processes, internal expertise and the ability to redesign work around the technology.
The same publicly available model can produce very different results depending on the information, context and systems surrounding it. A generic AI tool may help an employee summarise a document. An AI-capable organisation, by contrast, could securely connect the model to approved internal information, embed it within a controlled workflow, assess the quality of its outputs and use the results to accelerate a customer or operational process. The model is only one part of that system.
“Feed the AI your private data and your customer insight, the ingredients a competitor cannot buy or copy. That is where differentiable advantage comes from,” Van der Merwe said.
This does not mean uploading confidential business information indiscriminately into public AI tools. It requires strong data governance, secure architecture, clearly defined access controls and certainty about how information may be used.
Companies must know which data is accurate, who owns it, where it is stored and whether it may legally and ethically be processed by an AI system. Without these foundations, even the most capable model could produce unreliable or risky results.
Organisational design is equally important. MIT CISR’s enterprise AI maturity research found that only a small proportion of companies had reached the highest level of AI readiness. Organisations with more advanced capabilities tended to outperform their industries financially, while less mature businesses performed below industry averages.
The difference was not based on technology alone. More mature organisations had clearer decision rights, stronger data foundations, suitable governance structures and employees who were equipped to use AI effectively.
“Roughly 80% of enterprise AI projects fail, and the cause is the organisation, not the model,” Van der Merwe said.
This has implications for how executives structure AI leadership. Responsibility cannot sit exclusively with the information technology department, innovation team or data office.
Business leaders must define the commercial problem. Data owners must ensure the relevant information is reliable. Risk and legal teams must establish appropriate safeguards, while employees must understand how the technology changes their work.
AI capability must also extend beyond a small group of technical specialists. Organisations will continue to require data scientists, engineers and governance experts, but much of the value will be created by employees who understand both the business process and how AI can improve it.
Training should therefore be connected to actual roles, decisions and workflows rather than delivered as a generic introduction to AI. Privacy and safety form another critical part of this organisational capability. As companies connect AI systems to customer, employee and operational information, the consequences of weak governance increase. A privacy incident, inaccurate automated decision or uncontrolled disclosure of confidential information could quickly undermine customer trust and the business case for AI.
“An AI-safety-first organisation, with privacy protection embedded rather than bolted on, is itself a moat: hard to build and hard to fake,” Van der Merwe said.
A strong AI safety posture should not be treated only as a compliance obligation. It can become a competitive asset, particularly in industries where customers are increasingly concerned about how their information is collected, interpreted and protected.
The companies most likely to build lasting value from AI will therefore not necessarily be those with early access to the newest model. Their advantage will come from combining widely available technology with capabilities that are specific to their organisations.
These capabilities take time to build. They include trusted data, accountable ownership, redesigned workflows, skilled employees and embedded safeguards. Unlike access to a model, they cannot be purchased instantly through a software subscription.
“Model access is not your moat. What you build around the model is,” Van der Merwe said.
Francois van der Merwe is the CEO and founder of Otinga.io and convenes the AI Strategy-to-Results Executive Bootcamp at Henley Business School Africa.
