Published on the 25/06/2026 | Written by Heather Wright

Why AI looks like a cost, not a growth engine…

A/NZ organisations are struggling to prove AI’s value – and without a clear way to link it to revenue, many are filing it under margin protection, rather than using it as a growth engine.

Amanda Williamson, director of Deloitte’s New Zealand Artificial Intelligence Institute, says most businesses are seeing productivity gains from AI use, but few can link that to revenue gains.

“Experiments do seem to be quite easy to do.”

“AI’s value is hitting the cost line, where it’s easy to see, but not the revenue line. So CFOs are not really clear on how to book it as growth,” she told iStart.

At the same time, research company Omdia says organisations are still approaching AI as a technology investment, rather than defining clear business outcomes – creating a ‘blind spot’ in measuring real returns.

One of the consistent themes across both sets of research is a disconnect between perceived and measurable value. On the ground, teams are using AI tools extensively and seeing benefits, particularly in areas like back office processing and customer interactions. But those gains are rarely tracked in a way that ties back to financial performance.

“We’re really bad at measuring AI right now,” Williamson says. “We’re not tracking value like we would other investments.”

The Omdia research, which included Australia and New Zealand and was conducted for Boomi, found 34 percent of organisations were unable to effectively measure the success of their AI initiatives.

The ROI blind spot

Michael Barnes, chief analyst enterprise IT Asia at Omdia, argues the problem starts with how organisations approach AI in the first place.

“There’s the erroneous assumption, or misguided belief, that somehow technology has value in and of itself,” he told iStart. “That’s simply not the case.”

AI is being treated as an extension of existing IT spend, rather than a strategic business transformation, meaning each new initiative adds complexity, rather than value.

He says organisations need to start with outcomes – something most are not doing.

“Experimentation with no clear end goals isn’t going to cut it. Think in terms of actual business outcomes and then work backwards towards the AI strategy.”

He admits it’s not easy. “It’s hard to engage the business, or have the business lead discussions with a focus on outcomes, when they don’t fully understand the capabilities of the technology.”

Without that outcomes-led approach, AI initiatives remain disconnected from the metrics that matter. And that, in turn, makes it difficult to justify investment and even harder to scale projects and reposition AI as a growth driver.

“Organisations need to have a better sense of outcomes in order to justify budgets for AI within particular business units,” Barnes says.

“The more different business units can clearly understand the value of AI and link that to their measurable outcomes, the more we will see AI budgets linked to those particular business units,” he says.

A/NZ decision makers were the most pragmatic and least enamoured of the potential value of AI and slightly more focused on the challenges that need to be overcome in Omdia’s research, with New Zealand decision makers being even more so than Australian respondents.

In growth markets such as the Philippines and Malaysia, there was a significantly higher expectation that technology would drive business growth or innovation.

That was also clear in the Deloitte research where just 23 percent of Kiwi CFOs saw application of technology, including AI, as a top three growth driver. That’s compared with 30 percent Apac-wide.

Acquiring new customers, increasing sales to existing customers, operational efficiencies, price increases and innovation all ranked ahead of technology for New Zealand CFOs.

Stuck in productivity mode

A key issue remains where organisations are focusing their AI efforts.

Many AI deployments are aimed at improving individual productivity – helping staff generate content, process information more quickly or automate small tasks. While useful, those gains don’t always translate into measurable business impact.

“Focusing on individual level productivity is only going to give us so much in terms of actually getting a return on investment from pilots and AI in general,” Williamson says. “We really need to focus on the whole of the workflow – where is AI really going to make a difference?”

Instead of a scattergun approach, putting efforts into many different ‘hobby projects’, she’s calling on local organisations to focus on where AI might really shift the dial.

“I’m seeing a lot of good effort and good time being spent on projects that you probably can tell upfront would not have been worth the time and effort. So get to the core of value.”

The inability to prove value is also showing up in the gap between experimentation and scale.

While AI pilots are widespread across Australia and New Zealand – covering everything from invoicing to customer-facing tools – many stall before delivering enterprise-wide impact.

“Experiments do seem to be quite easy to do,” Williamson notes. “But getting it into something that scales can be a real challenge because scaling forces you to confront clean data, a motivated team, a real workflow.”

The measurement gap is being reinforced by two practical challenges: Data and cost.

On the data side, both Williamson and Barnes point to long-standing weaknesses in data quality, integration and governance, that were often deprioritised in favour of more visible initiatives. As Barnes notes, AI has made data quality issues ‘very obvious, very quickly… up to board level’.

On the cost front, AI introduces a new level of unpredictability compared to traditional IT investments, with Williamson noting AI is opening up a whole new era of calculating costs. The shift – or tokenomics – makes it harder to build a reliable business case. (You can read more on tokenomics and how to address the issue in our earlier story.)

What needs to change

For CIOs, CTOs and CFOs the message from both Williamson and Barnes is consistent: The issue is not lack of capability, but lack of discipline.

Organisations need to define outcomes clearly and at a function level, with broad concepts like ‘productivity’ or ‘automation’ not sufficient.

“Outcomes for AI need to be context specific – they’re going to be very different by function,” Barnes says. “You can’t talk about process improvement or productivity improvements, that doesn’t necessarily mean anything in business terms. The question is always towards what end?

“Yes, you’ve improved productivity, but what does that mean for your staff, whether it’s your call centre or your fraud detection unit? What are you enabling staff to do that they can’t currently do – that needs to be thought through because in most cases, it isn’t simply a cost saving exercise.”

Measurement needs to be built from the outset, including clear baselines and post implementation tracking. “Measure what is the efficiency is without it and what it is afterwards,” says Williamson.

She says organisations also need to focus on end-to-end workflows, rather than isolated use cases, saying that’s where AI can begin to drive meaningful, measurable impact.

And finally, foundational issues, particularly around data, must be addressed if organisations want to move beyond experimentation.