Artificial intelligence has become the boardroom conversation of the decade. Across South Africa, organisations are investing in generative AI, copilots, predictive analytics and intelligent automation, driven by the promise of increased productivity, faster decision-making and competitive advantage. Yet many businesses are discovering that AI is not delivering the transformational results they expected.
According to Dr Pierre Le Roux, Managing Director of MOYO, the problem is seldom the AI itself. “Organisations are asking how to implement AI before asking whether their businesses are actually ready for it,” says Le Roux. “Every board wants an AI strategy and every executive wants to demonstrate progress. But if the underlying data is fragmented, inconsistent or trapped in disconnected systems, AI will not solve those problems. It will expose them.”
This is already happening across organisations that have spent years accumulating systems, databases, spreadsheets and applications that were never designed to work together. “AI is holding up a mirror to businesses,” says Le Roux. “It is exposing years of neglected data management, disconnected systems and weak governance. The technology is not failing. The foundations are.”
The rapid rise of generative AI has created the impression that implementing AI is primarily a technology exercise. Le Roux argues that the technology is often the simplest part. “The difficult work happens long before an AI model is deployed,” he says. “It involves understanding where your information resides, who owns it, whether it can be trusted and whether your systems can communicate with one another.” Every organisation generates enormous volumes of operational data through customer interactions, financial transactions, supply chains, HR platforms and production systems. Yet much of this information remains fragmented across departments and legacy applications. Different systems may contain conflicting versions of the same customer, product or business metric. Important data may be incomplete, outdated or duplicated. In many cases, employees still rely on spreadsheets and manual processes to reconcile information before decisions can be made. AI changes the consequences of these weaknesses.
“If different systems contain different versions of the truth, AI cannot magically determine which one is correct,” says Le Roux. “It produces answers based on the information it receives. Poor data does not become good intelligence simply because an AI model is analysing it.” This explains why organisations often see very different outcomes from similar AI investments. “AI is a force multiplier,” says Le Roux. “If your organisation has trusted data, disciplined processes and integrated systems, AI can create extraordinary value. If those foundations are weak, AI simply accelerates existing inefficiencies. That is why two organisations can deploy similar AI technologies and achieve completely different business outcomes.”
However, Le Roux believes the focus on clean operational data addresses only half of the real challenge. “Most conversations about AI readiness ask whether the data is accurate, complete and accessible,” he says. “Those are essential questions, but AI needs more than operational data. It also needs context.” AI does not only need information about what is happening in the business. It also needs to understand how the business itself works. That context includes the organisation’s structures, processes, rules, capabilities, systems and relationships. It explains how different business functions connect, how decisions are made and how information flows across the organisation. “AI does not just need access to your data,” says Le Roux. “It needs to understand your business. That is where many organisations fall short.”
In many businesses, this knowledge has never been captured in a structured, machine-readable format. It exists in employees’ heads, outdated process documents, presentation decks, spreadsheets and architecture diagrams that are seldom maintained. Le Roux describes this as the organisation’s architecture expressed as data, effectively, a digital map of how the business operates.
“An AI system may know that a shipment has been delayed, a customer has complained or a payment has not been processed,” he explains. “But without business context, it may not understand the relationships, dependencies and rules that determine what should happen next.”
The problem is therefore twofold. The operational data many organisations possess is often unreliable or fragmented. The contextual data AI requires to interpret that information is frequently missing altogether. “One dataset is dirty and the other simply does not exist,” says Le Roux. “You can clean operational data, but organisations also need to capture the architecture of the business in a way that technology can understand and use.”
Recently, the technology industry has begun referring to this challenge as ‘context engineering’, providing AI with the business context it needs to make better decisions. Le Roux believes the concept is important, but the underlying discipline is far from new. “The industry is presenting context engineering as though it is a completely new field created by AI,” he says. “In reality, it is architecture. Enterprise architects have always worked to understand how processes, information, systems and business capabilities fit together. AI has simply made that work impossible to ignore.”
This is where digital engineering and enterprise architecture become essential to successful AI adoption.
Modern integration platforms, APIs, cloud technologies and governance frameworks can connect operational systems and create reliable flows of data. Architecture provides the context that explains how that information relates to the broader business. Organisations need both. “Buying AI before fixing these foundations is like buying a Formula One car to drive on a gravel road,” says Le Roux. “The engineering may be extraordinary, but without the right infrastructure and an accurate map of the terrain, you will never experience what it is capable of.”
He says rather than beginning with the question, “Where can we implement AI?”, executives should first examine how information and decisions move through the business. Why do different departments report different figures for the same metric? Where does critical knowledge reside only in the minds of a few experienced employees? Which business rules have never been formally documented? Which systems cannot exchange information? And does the organisation have an accurate, current representation of how its processes, information and technology fit together?
“These questions reveal whether an organisation is truly ready for AI,” says Le Roux. “An AI strategy without trusted operational data and structured business context is ultimately little more than a technology ambition.”
For South African organisations operating amid rising costs, infrastructure constraints and global competition, getting these foundations right could become a significant competitive advantage. “The companies that lead in AI will not necessarily be those that buy the newest tools first,” concludes Le Roux. “They will be the organisations that have done the disciplined work of cleaning their operational data and capturing the context in which that data exists. The future will not belong to the businesses with the most AI. It will belong to those whose AI genuinely understands how their business works.
