Loading
Loading
The realistic AI value proposition for most South African organisations in 2026 falls into four categories. First: document intelligence — extracting, classifying, and processing information from documents at a scale and speed that would require large human teams. Insurance claims, contract review, compliance checking, financial reporting. Second: workflow automation — orchestrating multi-step business processes that previously required human coordination at every step. Lead qualification, customer onboarding, supplier management, internal approvals. Third: conversational interfaces — AI-powered customer service, internal helpdesks, and knowledge retrieval systems that reduce the human resource cost of answering routine queries. Fourth: analytical intelligence — pattern recognition in operational data that surfaces insights humans would not identify through manual analysis. Each of these is deployable now, with existing technology, at realistic cost, with measurable operational returns.
The most common failure mode in organisational AI programmes is starting with the technology rather than the problem. Executives who buy an AI platform and then look for use cases to justify it will find the platform underutilised and the returns disappointing. The correct sequence is: identify a specific operational problem that costs the organisation meaningful time or money, assess whether AI is the appropriate solution for that problem (it often is not), implement a targeted solution for that specific problem, measure the returns, and use that learning to identify the next problem. AI programmes that start broad and promise to transform everything typically deliver nothing measurable. AI programmes that start narrow and deliver a specific, measurable operational improvement build the organisational capability and confidence to expand.
AI systems are only as useful as the data they operate on. Most South African organisations I work with have data problems they have not fully diagnosed: data that lives in incompatible systems, data quality that is too poor to support reliable AI outputs, data governance frameworks that do not meet POPIA requirements, or simply data that has never been collected because the processes generating it were paper-based. An honest AI strategy assessment starts with a data audit. What data do we have? Where does it live? What is its quality? Who owns it? Can we access it programmatically? The answers to these questions determine what AI is realistically possible and what foundational work needs to happen first. Many organisations find that the most valuable first AI investment is not an AI system at all — it is the data infrastructure that would make AI systems effective.
South African executives face the same build-versus-buy decision in AI as in other technology domains, with an additional complexity: the AI landscape is evolving rapidly enough that today's best-in-class solution may be significantly inferior to what will be available in eighteen months. For most organisations, the right approach is to use established AI APIs and platforms — OpenAI, Anthropic's Claude, Google's Gemini — rather than building proprietary AI models. The investment in building and training proprietary models is justified only where the organisation has unique data that creates a genuine competitive advantage, and where that advantage is sustainable at the cost of ongoing model development and maintenance. For most South African organisations, that threshold is not met. What is worth building is the infrastructure and workflows that apply these AI capabilities to your specific operational context.
AI deployment in South Africa operates within POPIA's data processing framework, which has specific implications for AI systems that process personal information. Any AI system that makes or influences decisions about individuals — credit decisions, insurance underwriting, employment screening — must be designed with explainability and human oversight built in. The AI systems most exposed to regulatory risk are those that make automated decisions without adequate human review. The mitigation is straightforward: design AI systems to support human decisions rather than replace them, maintain audit trails of AI outputs, and ensure that affected individuals can request human review. For executives in regulated industries — financial services, insurance, healthcare — legal counsel on the regulatory implications of specific AI deployments is not optional.
My recommendation for most South African executives developing their first AI strategy is to start with one specific operational problem that costs the organisation meaningful time and money, has data available to support an AI solution, and has a measurable outcome that will demonstrate returns clearly. Implement a targeted solution for that problem. Measure the outcome. Use that credibility to fund the next initiative. Build organisational capability through practice, not through broad transformation programmes. If you are unsure where that first problem is, the right starting point is a structured operational audit — not a technology selection process.
Independent systems architect and digital strategist. I build digital infrastructure for organisations that cannot afford to get it wrong.