Cheaper data is not the same as better data
Stats NZ publishes more than 270 statistical releases annually covering hundreds of thousands of data series. These numbers flow into Reserve Bank rate decisions, business investment models, insurance pricing, wage negotiations, and government funding formulas. When the agency that produces them absorbs a seven percent budget cut and sheds around 300 staff, the downstream consequences are not theoretical.
Government Statistician Colin Lynch insists the cuts have not affected quality. The evidence says otherwise. Stats NZ bungled the food price index in recent months, part of a cluster of government data failures that included MBIE getting fuel shipment figures wrong and confusion over Covid death counts. Lynch says the food price error had no impact on CPI or GDP. That is reassuring only if you trust a smaller, stretched agency to catch the next one.
The census is gone and nothing like it is coming back
The most consequential change is the abandonment of the traditional five-yearly census from 2030. The cost trajectory made the decision easy to justify: the 2013 census cost $104 million, the 2023 version hit $325 million, and the next was projected at $400 million. Statistics Minister Shane Reti says the replacement model, built on administrative data already held by government agencies plus an annual survey, will “save time and money while delivering more timely insights”.
That is a defensible call on fiscal grounds. But administrative data is collected for administrative purposes, not statistical ones. Tax records exclude informal activity. Health data depends on consistent coding. Benefit records reflect policy design as much as population reality. When these become the backbone of official statistics, the error profile shifts in ways that are harder to detect and harder to explain to data users.
Labour’s statistics spokesperson Lemauga Lydia Sosene has raised concerns about reliability for smaller population groups, and the political stakes are real: electorate boundary redraws will shift to every two terms from 2030, a timeline that could deny Maori voters an additional electorate at the 2029 election.
Three months late is not a rounding error
The push toward AI-assisted data is not purely about cost. There is a genuine timeliness problem. Stats NZ released December 2024 quarter GDP data in March 2025, a three-month lag that is standard internationally but damaging in practice.
Economist Dennis Wesselbaum frames it plainly: “Official statistics such as GDP and inflation have long been delayed, offering a picture of how the economy was, rather than how it is.” The consequences are not abstract. In August 2024, the Reserve Bank cut interest rates a full year earlier than markets expected, relying on private-sector indicators because official data had not caught up.
Former Associate Finance Minister Richard Prebble recalls sitting in Cabinet meetings “trying to interpret an economy using data that were months out of date” and argues AI-driven real-time indicators are now essential. He is right about the problem. The question is whether a depleted agency can execute the solution.
The $16.5 million investment in enhanced monthly CPI data is the clearest upside. Monthly inflation figures would be a genuine improvement for businesses making pricing and wage decisions. But Wesselbaum cautions that not all indicators are equal: “A good nowcast weighs data based on its quality and predictive value.”
When AI reads the data for you, who checks the AI
Stats NZ is rebuilding its website because more users now access official data through AI search tools rather than visiting government portals. Lynch frames this as adaptation: “My focus is ensuring New Zealanders and others can get access to that information in that world.”
Former Government Statistician Len Cook has called the current Stats NZ website “almost impenetrable for an ordinary person to use”, so a rebuild is overdue. But designing for AI-mediated access introduces a new layer of risk. AI tools trained on historical releases may not flag when methodology has changed. Caveats and confidence intervals, already ignored by most human readers, are unlikely to survive the compression of an AI summary.
Treasury’s own analysis identifies New Zealand’s historically slow technology diffusion as a structural barrier to capturing AI productivity gains. It also flags that AI disproportionately affects higher-skilled tasks, meaning a knowledge-intensive agency like Stats NZ faces more disruption, not less.
What business users should actually do
Lynch says trust is at the centre of everything Stats NZ does. That is the right aspiration. But a smaller workforce, a new data collection model, a rebuilt distribution platform, and a recent history of errors add up to a period of elevated transition risk.
For any business that builds forecasts, prices risk, or plans investment around official statistics, the practical implication is straightforward: understand that the numbers are about to change in how they are produced, how they are delivered, and potentially how reliable they are. The new system will probably work. But its first real test will come during a recession, a housing correction, or an election redistribution, when getting the numbers wrong carries consequences no AI tool can fix.
Sources
- Newsroom: AI and cost efficiencies change how beleaguered Stats NZ publishes info (2026-04-07)
- RNZ: Collecting statistics – When the numbers don’t add up
- Newsroom: Govt’s census shake-up advances despite ‘unreliable’ data concerns (2026-03-04)
- Inside Government NZ: Minister announces new direction for Stats NZ
- NZ Herald: NZ needs to embrace AI tools to know what is happening with the economy now – Richard Prebble
- The Conversation: How real-time data can lead to better decisions on everything from NZ’s interest rates to business investment
- New Zealand Treasury: The Impact of Artificial Intelligence – an economic analysis (AN 24/06) (2024-07)