Your AI runs on Stats SA. Its minister says the budget is not enough.
Credit models, demand forecasts and economic nowcasts are calibrated against official figures that a strained agency must keep producing.

Sign up for my Substack daily AI newsletter: https://johanosteyn.substack.com
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Every year on 18 November, Africa marks African Statistics Day, a reminder of the role official numbers play in economic and social development. In the boardroom, official statistics rarely feel urgent. Artificial intelligence changes that. The models that price credit, forecast demand and estimate where the economy is heading lean, somewhere along the line, on inflation, growth, employment and population figures. In South Africa, one agency produces those figures, and it is under strain.
CONTEXT AND BACKGROUND
In May, Khumbudzo Ntshavheni, the Minister in the Presidency, told Parliament that Stats SA’s allocation of R2.98 billion for 2026/27 “is not enough”. She asked for more money so that household expenditure surveys could be run more often, and called official statistics “our strongest defence against uncertainty and misinformation”.
Her deputy, Nonceba Mhlauli, said the agency’s financial pressures must be addressed to protect “the quality and sustainability” of national statistics. Stats SA plans 299 statistical products for 2026/27 and is redesigning its household surveys around a continuous population survey. The Statistics Amendment Act, in force since October 2025, strengthens the Statistician-General’s authority to coordinate the national statistics system and to share data with bodies such as SARS and the Reserve Bank.
South Africa is not alone. The UK statistics regulator reported in June “continuing concerns about survey response and representativeness”, and found that “AI use in official statistics remains limited”. The UK’s Office for National Statistics has “narrowed the scope” of what it produces in the short term while it works to stabilise its critical statistics.
INSIGHT AND ANALYSIS
The link to AI is direct, even if it is rarely drawn. Official figures are seldom the raw training data for these models. They are the baseline that adjusts, benchmarks and stress-tests them. A bank’s affordability model uses inflation to adjust incomes. A retailer’s forecast leans on population and household estimates. An economic nowcast is judged by how closely it tracks the official growth figure. When the underlying numbers come from smaller samples, lower response rates or less frequent surveys, the models do not fail loudly. They drift, and the decisions built on them become quietly worse. The same applies inside government, where AI tools for planning, social grants or service delivery will inherit whatever weaknesses sit in the population and household data.
I have previously written about this, using the Quarterly Labour Force Survey to show youth unemployment rising as AI reached entry-level work. That argument, like every serious board discussion of AI and jobs, depends on Stats SA’s figures being right.
There are two strong counter-arguments. The first is that private data, from card transactions to satellite images, can replace official figures. It cannot replace them entirely, because private data still needs an official benchmark to correct its blind spots. The second is that AI will make statistics cheaper to produce. That is partly true. The Office for National Statistics already uses generative AI to code labour market data and wants its statistics to be “AI-ready”. Technology cannot create data that was never collected, however, and a survey that is not funded or answered cannot be automated.
IMPLICATIONS
Business has more at stake in a properly funded Stats SA than it usually admits. The data-sharing provisions of the amended Act offer a route for administrative data to strengthen official statistics, and organisations that hold useful data of their own have an interest in partnerships of that kind working. Seen from a board’s perspective, the cost of good statistics is small next to the cost of decisions made on poor ones.
Boards can start with three questions. First, which of our models and forecasts depend on official statistics, and for which figures? Second, would we notice if those figures were delayed, revised or based on thinner samples, and what would we do? Third, are we prepared to support the national statistics system, through data partnerships or public advocacy, as part of protecting our own decisions?
CLOSING TAKEAWAY
AI is often described as the age of private data. Underneath it sits a public good that most organisations take for granted: the national figures against which their models are checked. When a minister tells Parliament that the budget for those figures is not enough, every board using AI should hear it as a risk to its own decisions. On African Statistics Day, the case for funding good numbers is no longer only a matter for government. It belongs on the risk register.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net




Comments