That is the clear message in 2 surveys we’ve seen in the last few weeks. On 26 October, small business lender Prospa published it’s latest SME Sentiment Tracker. A week later, data platform company Cloudera released The State of Enterprise AI 2026, a global survey with an Australian breakout.
One is aimed at businesses with fewer than 50 staff, and the other at organisations spending hundreds of thousands of dollars a year on AI.
Small firms
Prospa’s tracker surveyed 501 New Zealand SME decision-makers through research firm Clarity Insight. 59% now use AI in some form, up from 50% in March. Most of that use is still partial. A third of respondents use AI in specific business functions, 19% are experimenting with it, and only 7% describe it as integrated across several functions and strategically important.
What was interesting was looking at what was stopping businesses. It isn’t cost. The main adoption barrier, at 34%, was uncertainty about the return on investment, followed by not knowing which tools would suit the business, at 32%. “When a third of the businesses still outside AI say they can’t see the return, that’s not scepticism. That’s a gap in the advice they’re getting,” said Adrienne Begbie, Prospa’s New Zealand managing director.
Larger organisations
Cloudera’s report is based on a Wakefield Research survey of 1,500 IT leaders and decision-makers, run in September.
In Australia, 76% of respondents invest at least US$500,000 a year in AI projects, yet only 21% said AI currently generates more value than it costs. Just 43% said their AI spend had grown on the year before.
89% had limited or redesigned AI use cases to mitigate rising token costs, and 98% had pulled back a deployment because of failures, hallucinations or unexpected behaviour. Asked what they would do differently if they started again, 49 per cent said they would put governance in place from the beginning and 46 per cent would have invested in data quality earlier.
New Zealand’s own figures for larger organisations, from Datacom’s 2026 State of AI Index, are more positive but follow the same pattern. 91% of organisations use some form of AI. A third say returns have exceeded costs, and another 30% said benefits and costs are roughly even.
Why is the return is hard to estimate?
Part of the problem is how most AI gets used. A Copilot or ChatGPT licence saves a small amount of time on an email, a meeting summary or a first draft. Across a team those minutes add up, but nobody recorded how long the work took beforehand, so there is nothing to compare against. The time spent checking and correcting the output rarely gets counted either, and subscriptions end up on company credit cards or buried in a software renewal.
Keir Garrett, Cloudera’s managing director for Australia and New Zealand, said success now comes down to “defining clear metrics from day one”. That holds at any size, and it doesn’t need a data team.
Start with a number
For a smaller business, a workable approach looks like this:
- Pick one process you already measure, or could measure within a week. Time from enquiry to quote, hours spent on month-end reconciliations, or how long customer emails wait for a reply.
- Record a few weeks of baseline before changing anything.
- Count the full cost, including licences, time spent reviewing the output and any setup work.
- Run the trial for a fixed period, then compare against the baseline.
A trial that shows a tool costs more than it returns is still a useful result. It tells you where not to spend next quarter.
