A survey by CRN Fast50 company Mantel Group has revealed challenges preventing large Australian businesses from achieving wide-scale AI adoption.
The 2024 State of Data & AI report was based on a survey of over 20 large Australian organisations engaged in AI projects.
The survey revealed that 86 per cent of organsations are in pilot or have limited AI or ML adoption, while only 10 per cent have achieved wide-scale adoption.
None have integrated AI or ML as a critical component of their business operations.
“More often than not, PoC purgatory is due to a lack of executive understanding, appropriate foundational elements, and appropriate MLOps technology," Mantel's principal consultant of data and AI Catherine Jordan said in the report.
AI projects and business outcomes misaligned
The report found roughly half of all AI projects lack alignment with strategic business outcomes.
“Investors will expect businesses to invest in AI projects that align with core business activities,” Mantel Group's data and AI partner Emma Bromet said in the report.
"For CEOs this means assessing all the ways that AI can transform their organisation and prioritising the biggest prize over quick wins, the loudest voice, or whoever has the ear of the data team" - Mantel Group data and AI partner Emma Bromet
Among the key AI challenges facing organisations, 71 per cent of those surveyed said they do not measure value during or post implementation, while 86 per cent consider the quality of core datasets as acceptable or worse.
Meanwhile, 33 per cent have no specific security standards for sensitive data.
Barriers to AI adoption
Of those surveyed, 62 per cent cited competing organisational priorities as a barrier to realising their AI ambitions, while 48 per cent cited a lack of data literacy and 38 per cent a lack of budget or funding.
Meanwhile, 38 per cent cited a lack of strong data governance and management, and 29 per cent legacy systems and tech debt.
"In the current economic climate, expediting value realisation by aligning the vision of the C-suite with the actual capabilities of the organisation is key. This means resolving competing priorities, budget discussions, and establishing realistic paths to value realisation," the report said.
Data architecture and ownership
Seventy-six per cent of surveyed organisations noted their organisation’s data architecture function is not well-established, lacks full staffing and a mandate for enterprise-wide change.
Meanwhile, 65 per cent have established clear ownership of core datasets. However, only 33 per cent of organisations are elevating their level of ownership above daily operational levels.
“This lack of ownership for non-core data (which often makes up the bulk of data) can create challenges for organisations, namely missed opportunities for better insights and informed decision making, increased security risks, and increased storage expense of low value data,” the report stated.
Data quality challenges
Fourteen per cent of surveyed organisations set, monitor and maintain quality metrics for the majority of their data, with some automated remedial processes.
Of the rest, 52 per cent find the quality of core data acceptable but requiring significant maintenance, while 34 per cent face larger data quality concerns.
"We are still seeing some misalignment between what data and analytics teams are delivering and why, and business outcomes" - Mantel Group data and AI partner Emma Bromet
"Executives are eager to harness the power of LLMs, but often underestimate the importance of fit-for-purpose infrastructure (i.e. ML Ops) and data integrity," Bromet said in the report.
"Competing priorities and a lack of data literacy are now the biggest challenges data leaders face, while having a well-defined data strategy is of much lower priority. Budget constraints are playing a role here, as well as data literacy now being considered a company-wide challenge to solve, in order to support and drive innovation."
Not focussing on culture, literacy and adoption
Mantel's head of data and AI strategy Thomas Maas explained the challenges organisations face in improving data quality and capability.
“...data teams disconnect themselves usually too far from the rest of the organisation. Priding themselves on technical brilliance over solving customer problems, all the way from data graduates to leaders," he said in the report.
"The primary role of data leaders in this day and age should be to focus on culture, literacy and adoption in the rest of the organisation and less on debating which modern cloud platform has the absolute lowest storage and compute."
Key areas of investment
In terms of where investment is needed in the next one to two years, 67 per cent of surveyed organisations identified data engineering, platforms and tools, and data governance and management, while 57 per cent identified ML or AI.
Sixty-two per cent of organisations noted customer service operations as the business function where they are seeing the most value from AI, followed by strategy and finance (43 per cent), and product development, marketing and sales (38 per cent.)
Organisations see the most value in using data and AI to predict consumer behaviour and develop interventions (57 per cent), followed by customer experience segmentation and personalisation (43 per cent), and workplace productivity enhancement (43 per cent).
Mantel helps Woolworths NZ use ML
Mantel’s report cites its work to help Woolworths NZ use ML to improve its data science teams.
The teams operated in silos without standardised processes and lacked rapid prototyping capabilities to iterate quickly, according to the case study.
Mantel helped the company introduce a standardised ML project template and delivery practices applicable to any advanced analytics use case.
The firm also implemented an orchestrated ML pipeline for streamlined development and deployment, and established an ML model deployment framework for prediction services with integrated model monitoring.
The project resulted in the data science teams accelerating development cycles, reducing reliance on manual maintenance, improving data and model quality, and improving the outcomes of data science initiatives.
-1.jpg&w=100&c=1&s=0)
.png&w=100&c=1&s=0)
