Why this decision is harder than it looks
AI spend is rising. The returns aren’t following.
Datacom’s 2026 State of AI Index found that 91% of New Zealand organisations now use AI, and 79% increased their AI investment in the past year. Globally, Gartner forecasts AI spending will reach US$2.7 trillion in 2026, up almost 50% in a year.
Yet nearly half of enterprises (46.9%) are already reporting AI spend over budget, according to Futurum. And in New Zealand, only a third of organisations say the returns from AI have exceeded the costs.
91%of New Zealand organisations now use AI Datacom, 2026
46.9%of enterprises report AI spend over budget Futurum, 2026
78%of New Zealand CEOs say AI has had little or no impact on their revenue PwC NZ CEO Survey 2026
55%of NZ CEOs and business owners say their AI activity is still limited to early pilots or consultant-led initiatives The CEO Institute, 2026
4%of NZ organisations have used AI to transform their core operations, down from 8% a year earlier Datacom, 2026
More often than not, the gap between the investment and the return comes down to how the decision was made. The money gets committed before anyone has worked out what the process was worth to begin with, what AI would change in it, and what might be lost.
This guide looks at that decision: how most organisations make it, why it so often fails to pay back, and a better way to make it.
How the AI roadmap usually gets built
Most AI programmes start the same way. The method is easy to follow, the governance feels familiar, and what comes out the other end looks like a strategy.
- 1Form a working group · People from across the business come together
- 2List the processes · Catalogue what the organisation does
- 3Flag the manual steps · Find where people are doing repetitive work
- 4Score the candidates · Rank them by volume and repetition
- ?Missing step: what is each step worth?
- 5Build the roadmap · Organise it around where AI can be deployed
None of this is sloppy work. It served organisations well when the question was which tasks a machine could take over. But AI changes more than tasks, and this method has no way of seeing what each step is worth.