How algorithms are quietly rewriting the state

Artificial intelligence isn’t coming to the New Zealand public sector – it’s already here.

AI is shaping everything from your tax bill to how quickly police process crime reports. And right now, it’s happening in a way that’s fast, fragmented and largely hidden from public view.

On the latest episode of The Business of Tech, I talk to BusinessDesk journalist Cécile Meier about her multi‑part investigation into the use of AI across government – an investigation built on a trove of Official Information Act responses from major agencies. What she found is both encouraging and unsettling.

Real efficiency gains and cost savings

There are genuine wins. Police have slashed processing times for lower‑severity crime cases from eight to ten minutes per file to as little as one to three minutes using an AI‑powered workflow tool – a saving estimated at 18,000 hours and around $1 million a year. Inland Revenue is using AI in its debt collection models, helping secure tens of millions of dollars in payment arrangements in just weeks. ACC and others report large productivity gains from Microsoft’s Copilot baked into everyday tools.

But scratch the surface, as Meier has, and the story gets a lot more complicated. There is a  public‑service‑wide AI framework governing how these tools should be deployed, but it is not binding. Agencies have guidlines for adoption but are largely left to design their own pilots, measure their own “time saved”, and often rely on the very vendors selling them AI to prove the return on investment. Quality, error rates and real‑world impacts on citizens barely get a look‑in.

The AI deployment divide

Her reporting also exposes a stark divide. Some agencies – IRD, ACC, Police and DOC – are forging ahead, training thousands of staff and embedding AI deep into decision‑support systems. Others, like Oranga Tamariki and Corrections, are so wary of the risks that they’ve blocked external AI tools and confined usage to tightly constrained, low‑stakes tasks. All of this is happening under the same loose, non‑binding guidance.

AI tools are stumbling over te reo Māori and Māori legal concepts, raising obvious concerns about bias and fairness. There's growing reliance on a handful of US tech giants to supply and measure government AI, and a rising wave of “shadow AI”, as public servants quietly use banned tools on personal devices because they’re simply too useful to ignore.

Listen to The Business of Tech for the full interview, and hear how AI is quietly rewiring the state, often without the scrutiny it deserves. Streaming on iHeartRadio, Spotify, Apple, or wherever you get your podcasts.

Show notes

'Slippery slope': Concerns millions spent on AI may not be delivering for public - BusinessDesk

Light rules, high stakes: Agencies push AI into frontline work - BusinessDesk

Government AI spending hits $20.2m with real cost likely far higher - BusinessDesk

See omnystudio.com/listener for privacy information.