In boardrooms across South Africa a quiet crisis is unfolding. There is a palpable pressure to innovate and capital is flowing into AI initiatives at an unprecedented rate. Yet, considering the bottom line of most organisations that huge investment isn’t translating to big returns. We are caught in what I call the “AI paradox”.
Globally, corporate AI investment reached more than $250bn in 2024, yet only a tiny fraction of companies are seeing a meaningful impact on their earnings. This isn’t a simple case of early-stage jitters. It is a structural failure.
We are now seeing a second wave of capital flooding into AI, hitting the same hurdles we saw a few years ago. Despite the billions spent, AI projects are failing at a rate double that of traditional IT projects.
Why? Because many of us have fallen into the technology-first trap. Too many organisations are treating AI like a static IT purchase — a piece of software you simply plug in and expect to deliver magic. But AI isn’t a product you buy; it’s a structural capability you build.
When you try to force an advanced AI engine into a legacy business process without preparation, the result is predictable: it’s like putting a Ferrari engine into a donkey cart. You won’t get to your destination faster; you’ll simply break the cart.
For South African firms operating in a market governed by strict regulations such as the Protection of Private Information Act (Popia), strategic abandonment ― dropping projects before they ever hit production ― isn’t just a wasted expense it is a critical loss of competitive standing.
Myth of the successful pilot
There is a dangerous gap between C-suite ambition and operational reality. We’ve all seen the impressive lab projects ― AI models that work perfectly in a controlled, clean-data environment. But these pilot projects are often a mirage. They create a false sense of security, masking the technical debt, old mainframes and manual processes lurking in the basement of the organisation.
Let’s be straight-talking. If your pilot projects are running in a curated bubble you are essentially lying to your board about your AI readiness. True maturity isn’t about winning a small test; it’s about scaling that success across a complex, legacy-driven infrastructure.
Scaling exposes every fracture in your data architecture, security protocols and governance models. While “AI emergent” leaders operate on hope, “AI mature” leaders operate on the realism that the physical reality of their legacy core is the ultimate gatekeeper of transformation.
The legacy elephant in the room
South Africa’s core economy often runs on older, monolithic systems, the heavy-duty mainframes that power our banks, insurers and retail giants. These systems are incredibly stable, but also stubborn. They perform critical transactions, but they are inelastic, and they weren’t built for the rapid, real-time data needs of modern AI.
The challenge isn’t just connectivity; it’s a language barrier. Most of the organisational effort now is lost to manual data preparation, trying to translate siloed, ancient formats into something an AI can read. We cannot solve this with a simple lift-and-shift approach.
You don’t need more complex tech jargon; you need smarter ways to bridge the gap. We must focus on building adapters, smart layers that allow our AI to understand and talk to our legacy systems without forcing us to rip and replace the core of our business overnight.
A roadmap for real value
To move from firefighting to orchestrating, I propose a three-phase shift that keeps regulatory compliance and human expertise at the centre:
- The Foundation (data hygiene). If your data isn’t clean, connected and contextual, it isn’t AI-ready. Stop chasing complex algorithms and start organising your data fabric. Use metadata tagging to ensure that when your AI looks at a file it actually understands what it is seeing.
- The Bridge (decoupling). We need to break down our heavy, monolithic processes into smaller, independent services. Consider the insurance sector: by extracting monolithic functionality into smaller, automated steps ― such as automated document retrieval and policy lookup ― firms are reducing processing times by more than 50%. This architecture also provides the “forensic traceability” Popia demands.
- The Future (orchestration). Finally, we scale. By using intelligent orchestration we create a continuous loop where our data informs better decisions in real time. This allows us to grow without the structural friction that usually stalls AI projects.
AI is a people problem
Ultimately, AI readiness is not an IT problem; it is an organisational learning problem. You cannot buy your way to an intelligent enterprise. I would recommend the following to South African C-suites:
- Move AI out of IT. Captured value in IT is minimal, while the real value sits in operations, sales and strategy. You need a joint leadership coalition (CEO, CIO and HR strategy) to own this transformation.
- Keep humans in the loop. Responsible AI requires explainable logic. Every decision an AI makes must be traceable, and human oversight is non-negotiable ― not just for regulation, but for good business.
- Reimagine the workforce. Stop thinking of AI as a tool for replacement. AI agents are your new digital coworkers. Pivot your focus toward onboarding these agents, managing their performance and redesigning workflows so that your human talent can focus on the high-value problem solving they were hired for.
The intelligent enterprise isn’t built by chasing the shiniest algorithm of the week. It is built by modernising your core, respecting your people and teaching your organisation to learn at the speed of light.
• Pillai is CEO of Leantechnovations and AIEISA.
Business Day
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.