Machine Learning

South African businesses are deploying artificial intelligence across finance, mining, agriculture and public services, with AI-based systems estimated to contribute about $3-billion to the country's economy yearly and forecast to reach $9-billion by 2030, according to Google South Africa. However, industry experts warn that successful AI adoption depends not only on technical skills but also on domain expertise, robust governance frameworks and readiness to manage unstructured "dark data" that IBM research suggests accounts for more than 60% of enterprise information in many organisations. The Department of Communications and Digital Technologies published a Draft National Artificial Intelligence Policy on April 10, 2026, opening a 60-day public comment period on a framework that proposes six new regulatory bodies and risk-based oversight modelled on the European Union AI Act, though legal analysts caution the document lacks operational specificity on data protection and cross-border flows.

South African organisations are using machine learning models for analytics, AI-enhanced automation, natural language processing chatbots and computer vision systems in mining and agriculture, with the South African Revenue Service deploying AI to enhance tax filing and First National Bank using Microsoft's Copilot system to craft communications in native languages. Software engineering company Dariel says businesses are applying AI to fraud detection, credit analysis, customer support, forecasting and decision support, but success requires blending technical capabilities in machine learning and data engineering with domain expertise, process design and risk management, as organisations that realise value are those that make the technology work within day-to-day operations.

Effective AI governance demands addressing "dark data", the unmanaged and underutilised information that exists across emails, contracts, collaboration platforms and archives, says Datacentrix general manager Shakeel Jhazbhay, noting that AI amplifies existing weaknesses when information is fragmented, duplicated or poorly governed. Organisations must audit existing information assets, classify data types, cleanse redundant or obsolete content, establish retention and destruction policies, and only then accelerate AI deployment with tools to parse unstructured data, as businesses that create visibility and trust across information environments will derive the greatest value from AI rather than those deploying technology fastest.

The Draft National Artificial Intelligence Policy commits to harmonising AI privacy controls with the Protection of Personal Information Act and proposes a National AI Commission, AI Ethics Board, AI Regulatory Authority, AI Ombudsperson Office, National AI Safety Institute and AI Insurance Superfund modelled on the Road Accident Fund. Legal analysts at Werksmans say the policy signals regulatory direction but leaves organisations exposed by failing to specify operational infrastructure, noting gaps around data minimisation versus AI's data-hungry training requirements, automated decision-making rights under POPIA Section 71, and cross-border data flows, while warning that seven new bodies risk overlap and diluted accountability without clear governance lines.

Industrial AI has moved from experimentation to active deployment, with 61% of organisations using AI in live operations and 20% reporting scaled, mature deployments, according to Cisco Research's survey of 1 000 operational technology decision-makers across 21 industrial sectors in 19 countries. The survey shows 83% of organisations plan to increase AI spending and 87% expect meaningful outcomes within two years, but readiness gaps are emerging in networking infrastructure, cybersecurity and IT/operational technology collaboration, with 40% citing cybersecurity as the biggest obstacle to scaling AI and only 57% reporting IT/OT collaboration despite its proven criticality to operationalising AI at scale.