Not all organisations find that the path for translating early AI success into repeatable outcomes is an easy one.

The industry is awash with stories of AI projects that have failed to deliver their value expected, with some estimates putting the failure rate as high as 85 percent.

Gartner has predicted that by 2025, 30 percent of generative AI projects will be abandoned after the proof-of-concept stage, due to poor data quality, inadequate risk controls, escalating costs, or unclear business value.

Despite this challenge, investment in AI is gathering pace, with a United Nations Trade and Development (UNCTAD) report projecting the global AI market will soar from US$189 billion ($287.2 billion) in 2023 to US$4.8 trillion ($7.3 trillion) by 2033 – a 25-fold increase in just a decade.

For executives today, the challenge is to ensure that early enthusiasm and success is translated into ongoing returns. But as many have learned, there is a big difference between running AI in a lab and letting it loose in the real world.

Real world challenges

According to Gartner senior director analyst Tony Zhang, a key challenge comes in the form of the resourcing that is needed to properly support AI at scale. While proof-of-concept implementations can be lightweight in terms of their demands on systems resources, moving AI into production places significant strain on underlying infrastructure.

“Doing a POC is one thing, but if you're moving to production, it's really means another thing,” Zhang said.

“People have realised generative AI requires a lot of effort at the back end. If they don’t have the budget prepared and the processes prepared, they that will probably end up as a failure in the production run.

“And you need to ensure that if anything happens in the production environment in the future that you have a mechanism or process to tackle those issues.”

Another problem arises from the desire to capitalise on early success by rushing proof-of-concept projects into production. Zhang cautioned that the average time needed to translate a proof of concept into a production environment could be around eight to nine months.

Creative disruption at scale

Despite these challenges, the promise of AI and its ability to drive efficiency and innovation remains compelling enough to encourage organisations to seek ways to leap over this operationalisation gap.

One company that is doing so is the international advertising agency TBWA. As a business that markets itself as ‘The Disruption Company’, TBWA has embraced AI as part of its tradition of helping brands break conventions and reframe markets.

According to its chief AI and innovation officer for Australia, Lucio Ribeiro, TBWA is focused on building tools and strategies that allow clients to embed disruptive thinking at speed, without compromising on trust, safety, or creativity.

Ribeiro said the company had moved beyond the proof-of-concept phase with several projects and was seeing solid returns from these investments. One of these was a proprietary product called RISE. With visibility of brands within LLMs now as vital as search engine optimisation once was, Ribeiro said RISE helped ensure brands remain discoverable inside emerging AI interfaces.

“AI in the enterprise isn’t a technology problem—it’s a behavioural one,” Ribeiro said.

“The biggest hurdle has been managing change, bridging the gap between ambition and readiness. Teams want to move fast, but clients need certainty—around IP, data governance, and creative integrity.

“Much of what you see online - AI-generated films, hyper-automated workflows - can’t yet be responsibly deployed in enterprise settings. The tools often lack the security, clarity, or legal footing required. Our job is to balance innovation with protection, which means we move fast, but with guardrails firmly in place.”

Ribeiro said TBWA’s strategy was to only scale those initiatives that truly shifted how the company worked or what it offered.

“ROI in this space is nuanced,” Ribeiro said.

“Sometimes it’s time saved. Other times, its creative velocity, pitch impact, or cultural resonance. We’ve reframed ROI to focus on business acceleration and competitive edge - especially in a creative context where hours saved isn’t the whole story.”

As part of a global network, Ribeiro said his team benefitted from greater capability than it could source locally, and this brought with it clarity around security, extensibility, and cost-efficiency.

“Our rule is simple: if it can’t scale securely and ethically in a commercial environment, we don’t deploy it,” Ribeiro said.

“That’s also how we evaluate tools—through the lens of trust, usability, their ability to amplify human creativity, and real impact.

Growth ambitions

At the Australian National University, the desire to scale its AI projects is tied directly to its strategy to maintain its position as Australia's highest-ranked research university.

According to the university’s director of digital infrastructure and information security Sajid Hassan this goal has fuelled efforts to transition proof-of-concept projects into enterprise scale production systems. This ambition is supported by ANU’s $163 million digital transformation program that is delivering secure, scalable platforms for real-time collaboration, big data processing, and high-impact computational research across all disciplines.

“We're building scalable, future-ready platforms that can adapt to evolving research needs rather than being locked into specific technologies,” Hassan said.

 “A core principle is ensuring that data-driven decision-making becomes embedded across all university operations, from research to teaching, to administration.

“We're also focused on building resilient, secure infrastructure that protects sensitive research data while enabling the collaboration essential to modern research. This balanced approach ensures we're not just investing in technology for its own sake, but creating lasting value that directly supports ANU's strategic objectives.”

A key focus for ANU has been building next-generation research platforms with scalable compute and storage infrastructure that enables research teams to undertake complex modelling and data-driven innovation.

“This success in scaling has been achieved through careful attention to architecture design, ensuring solutions are built to scale from the outset rather than requiring complete rebuilds,” Hassan said.

“When selecting AI tools, scalability is paramount – solutions must be able to grow with increasing research demands without requiring architectural overhauls. Integration capabilities with our existing research infrastructure are critical, as isolated tools provide limited value in our complex ecosystem.

Having established the foundational infrastructure required for AI and machine learning workloads, including high-performance computing, scalable storage, and robust networking capabilities, Hassan said research teams could now undertake complex modelling and data-driven innovation that was previously impossible due to infrastructure limitations.

“We've successfully created scalable platforms that support growing AI research demands across diverse disciplines, from climate modelling to biomedical research,” Hassan said.

“Our data maturity and governance capabilities have improved significantly, providing the trusted foundation necessary for AI applications.

Spatial reasoning

While data and AI capabilities have quickly found themselves embedded into the core of many traditional organisations, it is also giving rise to new forms of business that are specifically enabled by AI capabilities.

For the Australian geospatial and aerial intelligence business Outline Global, its entire business model is based on capturing, processing, and delivering geospatial and aerial intelligence to support critical decision-making for its clients.

According to head of growth, Kevin Kwok, the company has been a long-time user of AI for processing vast volumes of aerial imagery and LiDAR data, combining human intelligence with AI and machine learning to extract valuable insights at scale.

“One of the biggest challenges has been scaling,” said Kwok.

“Our original on-premises infrastructure created scalability and integration bottlenecks, limiting our ability to fully leverage AI analytics and business intelligence tools.”

In 2024 Outline Global began working with Oracle to integrate its existing AI capabilities directly into Oracle Cloud Infrastructure (OCI).

“We’re actively working on expanding the capabilities of the platform, including increasing the automation of geospatial analytics and developing more advanced AI models to support observable compliance and data insights,” Kwok said.

“The next phase of our AI journey is being driven by AI Vector Search, which we’ve integrated through Oracle Database 23ai on OCI.

“This capability is enabling real-time, conceptual searches across complex geospatial datasets using natural language — making the retrieval of 2D, 3D, and spatial data faster, more intuitive, and far more accessible to users.”

Kwok said Outline Global had made significant progress in moving from isolated proof-of-concepts to scalable, production-grade AI solutions, and since implementing Oracle’s AI capabilities had achieved an increase in revenue by 30 percent and grown our lead generation pipeline by 50 percent, directly linked to enhanced speed and accuracy of our geospatial data service.

“The platform has also delivered a 40 percent efficiency gain in data dissemination speed, leveraging Oracle Cloud and database technologies,” Kwok said.

“What’s more, we’re on track to reach our milestone of 20 cloud-enabled clients by 2025.”