Published on the 23/04/2025 | Written by Heather Wright

And the fear of getting it wrong…

Organisations are struggling to identify which genAI use cases to prioritise with selecting those cases based on objective measures proving hard.

That’s according to a report from Enterprise Strategy Group, commissioned by IT vendor Snowflake, which surveyed nearly 2,000 business and IT leaders globally, including around 150 across Australia and New Zealand.

“Limited resources – even where funding is generous – and a need to focus efforts require difficult choices,”

The survey also found Australian and New Zealand organisations are experiencing more cost blowouts than their global counterparts when it comes to staffing their genAI initiatives.

Radical ROI of Generative AI highlights the struggles faced by organisations as they sift through a sea of opportunities and attempt to identify the most impactful use cases, while grappling with limited resources, with 71 percent of early adopters saying they have more potential use cases than they can possibly fund.

Globally, that’s translating into problems identifying the right use case to pursue, and here in Australia and New Zealand, the challenge is proving even greater, with 71 percent of local respondents reporting issues, versus 54 percent globally saying selecting the right use case based on objective measures, such as cost, business impact and ability to execute, is hard.

Earlier this year a survey by Informatica also highlighted the challenges of identifying business value for GenAI

“Limited resources – even where funding is generous – and a need to focus efforts require difficult choices,” the Radical ROI report notes. “The promises and proven capabilities of large language models and generative AI are changing week by week.

“Enterprises feel like the competitive pressure to keep up with technology advances has been cranked up to 11, and yet there’s so much low-hanging fruit. It’s what a military strategist might call a target-rich environment.”

It’s a high stakes game: 71 percent of global respondents say selecting the wrong use cases will hurt their company’s market position, while 59 percent believe that advocating for the wrong use cases could put their jobs on the line.

Despite that, Australian and New Zealand respondents profess to seeing a 44 percent return on AI investments – ahead of the 41 percent the global respondents who have quantified ROI (1,268 of respondents) have seen. Globally, more than nine in 10 early adopters say their genAI investment is already in the black.

The report suggests local organisations aren’t shying away from the potentially more risky implementations either. They’re more likely than their global counterparts to cite customer satisfaction, personalised experiences and improving customer engagement as key goals for their AI initiatives (53 percent versus the global 43 percent) and less likely to prioritise internal-facing projects (47 percent versus 55 percent).

Theo Hourmouzis, Snowflake ASEAN senior regional vice president Australia and New Zelaand, says many local businesses are already seeing the benefits of a customer experience focused genAI strategy, though he’s light on specifics.

That flies in the face of the global experience, with Snowflake saying 18 percent believe customer-facing projects would deliver the strongest impact, but instead focus on employee-facing initiatives due to infrastructure limitations, security concerns and accuracy issues. Another 13 percent prioritise customer-facing projects despite seeing greater potential in employee applications, often because they’ve identified ready-to-deploy solutions with more predictable returns.

Globally, the leading genAI use cases include IT operations (70 percent), cybersecurity (65 percent), customer service an d support (56 percent) and software development (54 percent), though the report notes that IT professionals consistently identify less use of generative AI than business counterparts.

“This may mean that shadow IT has been joined by shadow AI, possibly in the form of online consumer genAI tools.”

Highlighting that shadow IT is the discrepancies between the number of business users saying they use genAI versus the IT professionals who seem to know about that use: 73 percent of HR professionals. Say they’re using genAI to screen resumes, train employees and so on, but just 44 percent of the IT respondents were aware of that use, while 72 percet of sales organisations report using genAI but only 37 percent of their IT colleagues seem to know about it.

A/NZ investment is also running high with 32 percent of local organisations putting more than a quarter of their tech budget for the next 12 months towards genAI, versus 25 percent globally.

“Local organisations are funding genAI at a rate above the global average, which bodes well for the development and growth of AI in our region,” Hourmouzis says.

But he acknowledges that while there is clear appetite and drive to be ahead of the AI curve, there are also some hurdles to overcome, with the two biggest challenges being talent and data.

The survey found the cost of staffing for generative AI in Australia and New Zealand is often higher than expected, leading to unexpected costs.

The majority (84 percent) of local organisations say half or more of their genAI use cases have cost more than expected to get into production. That’s ahead of the global average of 78 percent.

Late last year, Gartner noted that costs were as big a risk as security or hallucinations when it comes to genAI, with 500 to 1000 percent errors in cost estimates possible.

Unlike conventional IT investments, where a capex investment is made and ongoing costs are relatively known, volatility in genAI operating costs based on usage means some companies can end up victims of their own success, paying for increasing adoption and more sophisticated queries.

Globally, 64 percent cite compute cost overruns, 61 percent report that supporting software including for logging, monitoring and observability and model optimisation, have cost more than anticipated, and 58 percent say data collection, labelling and processing have been more costly than expected.

On the data side, local organisations more often cited a lack of data diversity/range, time consuming data management tasks and data preparation as difficult areas, and 76 percent said it was hard to break down data silos.

For those that are succeeding though, the report says there are big returns.

A/NZ early adopters were more likely than the global average to report their efforts with genAI were enabling their organisations to make better, faster decisions) 91 percent versus 84 percent globally).