Spark survey: 14% of NZ firms run autonomous AI workflows

By DataForge team17 November 20265 min read

Spark surveyed 400+ NZ business leaders: 69% use AI, 14% run autonomous AI workflows and fewer than a third have a policy on AI use.

What Spark's research found

Spark released research on 16 November called Built to succeed: Resilience in the age of disruption, cyber risk and AI. The research draws on a survey of more than 400 New Zealand business leaders. Most of the media coverage led with the cyber findings. The AI numbers deserve as much attention from anyone planning next year's technology budget.

69% of the businesses surveyed use AI to some extent, mostly through general-purpose chat tools. Only 14% run full or semi-autonomous AI workflows. Spark also asked about data. 21% of respondents have unified data with AI embedded in their strategy and processes, while 26% say their data is mostly siloed.

Governance is thin. Fewer than a third of businesses have a policy governing AI use. The published coverage cites both 28% and 33% for the figure, and the summaries released so far don't explain the difference. Spark separated out a group of "confident performers", about 9% of respondents, who rated both their productivity and their confidence in meeting growth goals at 9 or 10 out of 10. 58% of confident performers have AI use policies, compared with 28% of all businesses. Respondents named privacy breaches, staff skills and resistance, and the reliability of AI outputs as the main barriers to using AI further.

The coverage doesn't give the fieldwork dates, the sampling method or a margin of error. Treat the percentages as a useful signal about mid-market New Zealand. They aren't a census.

Why the gap between 69% and 14% matters

Using a chat tool and running an AI workflow are different kinds of adoption. Staff can sign up for a chat tool themselves, paste in an email and get a draft back. The person stays in control of every step, and the business learns almost nothing from the exchange. A semi-autonomous workflow reads from a business system, makes a decision or prepares an action, and passes the result to a person or another system. Building a workflow needs reliable data, an integration, a named owner and rules about when a human has to approve. The 55-point gap between the two Spark figures shows how many firms have stopped at personal productivity tools.

The data figures go a long way towards explaining the gap. A business whose customer, job and finance records sit in separate systems can't easily give an AI model the context a workflow needs. When 26% of businesses describe their data as mostly siloed, many of the stalled AI projects in the survey are probably data projects nobody has started.

Other recent surveys show the same pattern. Prospa's SME Sentiment Tracker, published on 2 November, found 59% of New Zealand small businesses use AI in some form, up from 50% six months earlier, while only 7% consider AI strategically important across several functions. Among small firms that don't use AI yet, 34% cite uncertainty about return on investment and 32% don't know which tools suit them. AWS research released in September found only 15% of New Zealand businesses have a formal AI strategy, and 34% are confident they can measure AI return on investment. The three surveys define "use" differently, so their headline numbers can't be compared directly. All three still describe broad, shallow use with little structure around it.

The confident-performer comparison needs careful reading. Spark's data shows that the most confident firms are twice as likely to have an AI policy. The data can't show that writing a policy made those firms productive. A business organised enough to set AI rules is probably organised in other ways too. The policy is still a reasonable place to start, because writing one forces decisions about tools, data and accountability that any workflow project needs later.

The cyber findings belong in the same conversation

48% of surveyed businesses had experienced a cybersecurity incident, rising to 58% among organisations with more than 100 staff. Only 44% have a resilience plan, and 16% of those with a plan review or test it regularly.

The cyber figures matter for AI because an AI workflow links systems that used to be separate. An agent with read access to the CRM and write access to the job system widens the reach of a compromised account. Privacy breaches topped the list of AI barriers for good reason. Spark's Sam Thumath put it directly: "Moving quickly without those foundations can create new risks rather than reducing existing ones."

What NZ businesses should take from the survey

For most owners, staff are already using chat tools whether or not anyone approved them. The business has no record of what data has gone into which service. A short AI use policy is cheap to write and closes much of that exposure. Operations and IT leaders need a different conversation. Moving from chat tools into the 14% means choosing processes where the data already sits in one place and the rules are clear enough to write down. Those processes are usually narrower and less exciting than the AI demos vendors show.

Practical next steps

  • Write a one-page AI use policy this month. List the approved tools, the categories of data staff mustn't paste into them, and the person who approves a new tool. Our AI consultancy work usually starts with this step, because the policy also exposes the tools staff already use.
  • Ask teams which chat-tool tasks they repeat every week. Repeated tasks such as summarising inbound enquiries, drafting quotes or classifying supplier invoices are the strongest candidates for a proper workflow.
  • Pick one process whose data already lives in one system. Leave the siloed processes until the data is consolidated, because AI built on fragmented records produces unreliable output.
  • Decide where a person must approve before any build starts. Set a value or risk threshold above which the workflow pauses for sign-off, and record who owns the workflow when the workflow gets something wrong. Our process and automation team designs those approval points into the build itself.
  • Measure the current process before changing it. Record handling time, error rate and volume for a fortnight, so any claimed gain can be checked against real numbers.
  • Add AI tools to the resilience plan and test the plan. Include what staff do when an AI service is unavailable, and which accounts need revoking if an AI-connected system is compromised.

Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.

Get my free assessment How we help
Get new articles by RSS