South African technology leaders are facing the same reckoning playing out in boardrooms globally: the CFO is now in the AI conversation. AI initiatives that were technology-led are failing. Those that are business-led − anchored to defined outcomes, measured returns and value that a board can verify − are consistently delivering. The difference is not the technology. It is the discipline.
In the last quarter of 2024, something shifted in the conversations I was having with enterprise leaders. CFOs and finance executives started showing up where previously only technology leaders had sat. That shift is now permanent. The AI experimentation era is over. The AI accountability era has begun.
The numbers defining this moment are uncomfortable. Global AI spend reached $644 billion in 2025; 89% of enterprises have adopted AI tools. Yet only 23% can accurately measure whether those tools are generating a return (Larridin State of Enterprise AI, 2025).
The common thread in the failing 77%? AI adopted as a technology initiative, not driven as a business one. In South Africa − across financial services, telecommunications and the public sector − the same gap applies, and the consequences are compounding.
This pattern is not new
I have watched this pattern before. Internet, mobile, cloud − every major enterprise technology wave followed the same arc: uncritical adoption, CFO scrutiny, then the discipline that separated leaders from laggards.
The measurement problem is solvable, and solving it is the highest-leverage action available to South African technology leaders right now.
S&P Global data shows 42% of companies abandoned most AI projects in 2025, up from 17% the year before. That is not a failure signal − it is a correction. The organisations building AI value infrastructure now will be the AI leaders of the next decade.
The measurement problem is solvable, and solving it is the highest-leverage action available to South African technology leaders right now.
Business-led AI outperforms tech-led AI by nine to one
What I see consistently in my advisory work is confirmed by the data: AI delivers strong returns for organisations that approach it rigorously.
JOIN THE CONVERSATION
Register for the ITWeb AI Summit 2026 on 22 April in Johannesburg and hear from industry experts on how organisations are turning AI into measurable business impact.
IBM’s Institute for Business Value is unambiguous − organisations with formal AI value frameworks achieve 55% ROI. Those without achieve 5.9%. A ninefold gap, driven entirely by governance discipline, not by the quality of the tools being used.
The Google Cloud 2025 ROI of AI Report − surveying 3 466 senior leaders globally − confirms this. Organisations deploying agentic AI, where AI executes multi-step tasks and produces countable outcomes, report dramatically better results. C-suite sponsorship alone accounts for a 35 percentage-point ROI difference: 78% of firms with executive backing achieve positive ROI, versus only 43% without.
Business-led AI starts with a defined outcome and a value framework before the first model is deployed. Technology-led AI starts with the tool and hopes the business case follows.
Shadow AI is a POPIA and ROI problem
There is a second problem compounding the first, and most boards do not know it exists. Shadow AI − employees using unapproved tools, personal LLM accounts, and AI features embedded in SaaS products without IT awareness − is already operating at scale inside most organisations.
UpGuard found more than 80% of employees use unapproved AI tools at work; IBM puts the annual cost of insider AI negligence at $10.3 million per organisation.
For South African firms under POPIA, every unapproved tool processing client data is a regulatory exposure that will eventually appear on a breach notification, not an AI budget line. Governance and approved-path enablement are the answer. Banning AI is not.
The pricing shift that will force measurement
The shift to outcome-based pricing makes all of this non-negotiable. The token-based model − pay per query regardless of value produced − is already giving way to something more aligned with business reality. Zendesk, Intercom and Sierra AI now charge only for outcomes delivered.
The AI pilots South African enterprises ran in 2025 are entering CFO renewal cycles right now. “We believe in AI” will not survive that conversation. Demonstrating business value is no longer a governance best practice − it is a procurement prerequisite.
Three things to do now
Set baselines before the next deployment. In every engagement I run, the absence of a pre-AI performance baseline is the single most common reason an organisation cannot defend its investment at renewal. Without a documented before, there is no provable after.
Gain visibility of shadow AI before attempting to govern it. An AI tool registry and tiered risk classification are the starting point. The goal is to make the approved path faster than the workaround − not to write policies that sit unread in SharePoint.
Define what a successful outcome looks like before signing the next vendor contract. Outcome-based contracts are already arriving in renewal conversations. The organisation that cannot define a “resolved” outcome before signing will be in an expensive dispute about it at invoice time. That definition is a business negotiation, not a technical one.
The accountability era is already here
South African technology leaders face a uniquely demanding environment: POPIA compliance, FSCA and SARB oversight, B-BBEE procurement considerations, and boards simultaneously excited about AI and anxious about costs they cannot yet explain.
Our regulatory context makes the governance imperative more urgent, not less. The technology works. But in 20 years of enterprise technology, I have never seen a tool alone change a business outcome.
What separates the organisations pulling ahead is the discipline to define the outcome before deploying the tool, measure relentlessly, and scale only what demonstrably works. That is not a technology strategy — it is a business strategy, and the only AI strategy that survives a CFO renewal cycle.
How many South African enterprises are genuinely ready for that conversation?
Sources:
- Google Cloud ROI of AI Report 2025 (3466 senior leaders globally)
- Larridin State of Enterprise AI 2025 (1000+ enterprise leaders)
- IBM Institute for Business Value: AI and Value Creation 2025
- IBM Cost of Data Breach Report 2025
- S&P Global AI Investment Trends 2025
- UpGuard State of Shadow AI 2025
* Eugene Perumal is a strategy and architecture principal with over 20 years' experience in enterprise technology across telecoms and financial services, including senior roles at Vodacom Group and Absa Group. He holds Master’s degrees and certifications in enterprise architecture, AI governance, cloud and analytics. He writes on enterprise AI strategy, ROI measurement and the shift to agentic AI deployment.
Share
