South African companies They are facing immense pressure to implement artificial intelligence technology before they are left behind. Google Cloud 2026 AI Agent Trends shows that organizations are doubling AI spending, with CEOs increasingly taking personal ownership of AI strategy.
Yet the same research shows that most enterprises are nowhere near ready for agentic AI because their data is fragmented, inconsistent, and locked inside legacy systems.
This is the reality that many local organizations are experiencing, and accordingly Kim SchultzHead of Digital Advisory Practice Excelera Digital Group (ADG), the lesson is simple: “You can't build AI on a broken foundation. If your data is dirty, your AI will be dirty. If your systems are broken, your AI will be broken.”
Addressing decades of technical debt
While global enterprises are racing towards agentic AI (systems that can plan, perform tasks and execute tasks across all applications), South African businesses are facing another fundamental challenge – decades of technical debt.
The AI Agent Trends 2026 report highlights that agentic AI depends on clean, governed, connected data. It notes: “The real power comes from giving each employee agents based on the company's own enterprise context – its internal systems, knowledge base, customer data and past work.” Without that grounding, AI agents simply hallucinate increasingly.
“South African organizations want AI agents to automate workflows, but many can’t even trust their customer records,” says Schultz. “Before you can dream about autonomous agents, you need to get the basics right, including data quality, governance, integration and system modernization.”
Dirty data makes AI worse
AI agents don't magically clean up an organization's data; Whatever they are given, they make the most of it. If an organization's customer relationship management system has duplicates, missing fields, outdated contacts, or inconsistent naming conventions, an AI agent will largely automate those errors.
Similarly, if a company's ERP and finance systems don't talk to each other, an AI agent can't organize multi-step workflows, and if business rules aren't documented, an AI agent can't safely take actions.
Schultz warns that companies will hit a “data governance wall” because their data is too messy and lacks the write-back rules needed for secure automation. “AI agents don't fix inefficiency; they accelerate it. If your processes are broken, AI will break them faster,” says Schultz.
Modernization of the core: one board–level priority
For most global companies, modernizing legacy systems is no longer an IT project but a board-level imperative. Companies must break down monolithic systems into flexible, API-driven components so that AI can actually function.
“Executives are realizing that AI is not a layer you sprinkle on top. It's the result of a modern, connected, well-run digital core. Without it, AI is mere lipstick on a legacy pig,” says Schultz.
Organizations aiming to become AI-ready should adopt a practical and phased strategy to implement AI while avoiding unnecessary spend on trends. This approach involves several key steps:
- Data Quality Assessment: Identify duplicates, inconsistencies, missing fields and structural issues.
- Data Governance Framework: Define ownership, rules, lineage, and write-back policies that are important for agentic AI.
- System Integration and API Enablement: Break down silos so data can flow throughout the organization.
- Heritage Modernization Road Map: Prioritize which systems should be upgraded or replaced to support AI.
- aye–Readiness Verification: Make sure the organization has the clean, connected, and controlled data needed for secure automation.
risk of doing nothing
Although the risk of haste is high, the risk of not changing old systems will outweigh the risk of change. This is especially true for South African businesses. Those who delay data modernization will find themselves unable to scale AI, unable to compete, and unable to meet customer expectations.
“If you skip the groundwork, your AI project will fail,” says Schultz, “not because the technology doesn't work, but because your data doesn't work. Organizations will have to go through the messy, complex, but necessary work of getting ready for AI.”
“AI is not magic, it is mathematics, and mathematics requires clean inputs, so get your data house in order first, otherwise your AI investment will be a waste of time and money.”