What it takes to succeed with AI

Enterprises are failing to realise the full potential of artificial intelligence (AI) and risk actively draining employee productivity unless they adopt an organisation-wide platform approach that brings models directly to their data.

That’s the warning from Cloudera’s chief technology officer for Australia and New Zealand, Vini Cardoso, who called for businesses to move from fragmented, ad-hoc AI experiments towards empowering staff and targeting use cases that deliver measurable returns.

In an interview with Computer Weekly, Cardoso said it is also critical to consider the value any AI project can bring to the business. An estimate that serves to secure approval is one thing, but that value must be tracked through the implementation.

“We’ve seen organisations, like one of the banks in the region we work with, achieve A$150m of annual value creation through AI across multiple formats – not just generative AI, but traditional machine learning as well. This was achieved by selecting the right use cases that drive the right value.”

By focusing on the right initiatives – those where they could gain efficiencies, reduce risk and avoid losses – those organisations were able to achieve measurable results.

“In any transformational use case – whether AI today or ERP [enterprise resource planning] in the past – you need business leaders on board. They have to embrace the change, which they will only do if they understand the positive outcomes. Once they do, they should lead by example, establishing a culture that supports the whole organisation towards that goal, giving teams the opportunity to learn and experiment, while keeping focus on the transformation objectives,” observed Cardoso.

This change presents IT and data professionals with an opportunity to redefine their place in the organisation, taking their technology expertise and using it to influence business decisions and outcomes.

For example, a data analytics team might examine quarterly business reviews and provide a recommendation along the lines of: ‘Maybe you should invest in these opportunities, as based on our analysis they are higher value and more likely to close, rather than using a scattergun approach across many prospects.’

“In my experience, this makes a more satisfying job. Employees are no longer just operating software: they see real outcomes and influence critical business decisions,” he suggested.

However, a recent survey by Sapio Research for Foxit Software suggested that the use of AI may temporarily reduce rather than improve productivity.

That survey of 1,400 desk workers and executives in the US and the UK found executives save just 16 minutes per week, while desk workers lose 14 minutes, due to the time needed to validate AI outputs. A likely related finding was that one in four executives, but only one in ten desk workers, are “extremely confident” in AI outputs.

Cardoso noted that technology adoption often starts with scepticism, but trust grows over time as value is demonstrated. “Take self-driving cars: in California, I saw ride-hailing cars operating with no one inside. The level of sophistication and safety implementation is very high, and gradually, people gain confidence and adoption increases.

“The same is happening with AI adoption in business. Organisations take measures to increase confidence in AI outcomes, starting with trusted data: knowing where it comes from, what transformations were applied, and ensuring reliable sources,” he added.

People must be able to trust both the model and the data it is fed. Keeping humans in the loop during the early stages helps develop trust, and that – along with a corresponding increase in risk appetite – accelerates adoption.

“The human side – technical skills plus business knowledge – remains critical to review frameworks and guide decisions. And as society evolves, frameworks will need to adapt. Models must evolve too. The beauty of AI is that it learns continuously, naturally adjusting outcomes over time to reflect new knowledge and changing expectations,” Cardoso said.

Another recurring issue is moving from experiments and pilots to production systems. Those with the potential to deliver real value are more likely to gain the necessary support, according to Cardoso.

He also recommended adopting a platform approach from the start. “Sometimes people design solutions using whatever technology is at hand, but that technology may not be scalable or operationalised. The time required to recode, re-architect, or re-engineer data pipelines can be massive.

“With a platform approach, you have a consistent way to tap into data sources and develop AI for production. Data and workloads become reusable and standardised. You can apply consistent security, governance and controls, making the transition from pilot to production seamless. At the same time, you adhere to governance frameworks and address regulatory and compliance requirements,” he said.

To that end, Cloudera offers trial use of its software so potential customers can test both its capability and their own use cases.

The platform can be deployed on-premises, in the customer’s selected cloud, across multiple clouds, or in a hybrid environment. This means organisations can choose whichever cloud is cheapest at the time while maintaining the same workload, infrastructure and governance standards. That includes the ability to place workloads to take advantage of unused pre-committed cloud credits.

Partly due to the cost of moving data between clouds, “the approach we take is to bring the AI workloads to the data, rather than moving the data to the AI,” Cardoso said. “If the data is on-premises, we help deploy AI capabilities to run models there. If it’s in the cloud, we deploy workloads close to where the data lives.”

However, in appropriate circumstances, for example, where there are low or no egress fees, the Cloudera platform has federation capabilities, allowing access to data across different environments without moving it.

“The platform approach ensures AI runs where the data is, without exposing IP or sensitive information – critical for sovereignty requirements. AI should augment skills, unleash ideas, and generate value, not replace teams.

“High-value use cases are essential. Low-hanging fruit is useful for learning, but targeting real business problems drives investment and board-level buy-in for scaling AI,” Cardoso said.