Organisations are well aware that data plays a critical role in ensuring their artificial intelligence (AI) adoption is impactful, but they still are struggling to gain a proper grasp of their data.
It is a challenge that has persisted over decades, as each new wave of technology transformation emerged, including business intelligence, cloud, omnichannel communications, and now AI.
“AI is all about data. To have a successful resolution, everything relies on data,” Krešo Žmak, Infobip‘s chief innovation officer, said in a video call with FutureCIO.
Citing the infamous data maxim “garbage in, garbage out”, Žmak noted that with AI, “garbage” is amplified if organisations do not train their AI models with quality data.
Without the right setup, businesses will continue to face the same challenges that come with operating data in silos, where teams each run their respective stack, he said.
Marketing owns the company’s marketing data while sales claims the CRM stack, with IT taking overarching responsibility for the infrastructure including data.
Without coordinated efforts across the organisation, the different teams end up buying and augmenting their own AI tools, he said.
If data ownership remains within the individual departments, each residing in different systems, the same painpoints will persist for enterprises, he added.
Data is scattered across silos and, without cohesive visibility of their customers’ profile, brands end up with disjointed consumer interactions.
Furthermore, organisations have large amounts of unstructured data, which remains a common bottleneck today, Žmak said.
Making sense of the unstructured
Data volume has continued to grow exponentially, concurred Remus Lim, Asia-Pacific Japan senior vice president at Cloudera.
He noted that unstructured data now is commonly collected from a multitude of devices, including sensors and mobile devices.
Organisations have to make sense of these different datasets and establish relationships and links between various components, including users and products.
With the explosion of data, fuelled by AI, this challenge has further intensified, Lim said in a video call with FutureCIO.
As it is, an average of 35.5% of enterprise data now is generated by AI assistants. This figure is projected to hit 42.1% within a year, AvePoint estimated in a report, which surveyed 750 business leaders in Asia-Pacific, EMEA, and the Americas, who were responsible for AI programs, data security, or data management.
Almost half, at 46.9%, of their employees use AI agents on a weekly or daily basis, with work processes leveraging AI agents expected to double over the next 12 months.
Some 84.1% of organisations currently manage in excess of 1 petabyte of data, up from 79.2% last year, the study found.
In addition, 78.1% note that at least half of their data is more than five years old, compared to 70.7% last year.
“When AI systems consume and act on AI-generated content, including redundant, outdated, or low-quality data, governance failures compound at scale,” AvePoint said.
Market studies reveal that companies continue to struggle with data challenges, even as more look to ride the latest technology surge, in particular, agentic AI.
A majority 89% of Singapore organisations believe agentic AI has moderate to very high potential to transform their business, according to SAP’s Value of AI Report 2026, which polled 2,600 business leaders across 13 markets, including 200 in Singapore.
Most in the Asian nation view data as critical, with 75% saying the availability and quality of data is very important or critical to making decisions about AI.
However, a lower 55% feel they are data-ready for AI, a drop from 63% last year who said likewise.
Some 82% point to incomplete data as a challenge, with 81% noting that low-quality AI outputs have led to rework, delays, or backlogs.
Another 79% cite data quality or availability issues as a key reason their organisation cannot drive more value out of their AI deployments.
“Realising real value from AI is not going to be easy because it demands a new approach,” Sean Kask, SAP’s chief AI strategy officer, said in the report. “Businesses, large and small, will need to connect AI to the data and processes that run their organisations, and make sure it has the context and governance to drive trusted results.”
Žmak noted that institutional knowledge of their organisation’s business operations, as employees learn and optimise processes on the job, can be difficult to capture, but data lakes now allow unstructured data to be stored alongside structured data.
There also are platforms and tools to log and extract tacit knowledge, for example, from internal meetings and discussions on Slack channels, he said.
He also pointed to GenAI (generative AI) tools that help document and provide summaries of work an individual employee carries out, creating a “digital brain” of sorts.
Build the right architecture for agentic
So how should organisations reengineer their data architectures for agentic AI?
The first step is to ensure availability of their data, so organisations will need to have the right APIs (application programming interfaces), or in the case of agents, MCPs (Model Context Protocol) in place, said Žmak.
AI agents also will need to be trained, with context, and to understand relationships between different data objects, he said.
Topmost, companies must establish data governance and compliance, including establishing the data AI agents can access and when it can be accessed, he noted.
These are critical since LLMs (large language models) remain prone to hallucinations, Žmak said.
Lim noted: “Data [itself] is no longer the challenge. What is [a challenge] is data access, governance, and security.”
This is particularly tough for large enterprises, such as banks, telcos, and governments, where data access for users is vast, he said.
These companies have to build a data governance and security layer, so they can control and govern data access, he said.
He echoed Žmak’s observation that organisations today still run data in silos and are making efforts to consolidate these to establish a single layer of truth.
Underscoring the importance of governance, Lim noted that, fundamentally, organisations have to realise they need to govern data access.
He touted architectures that enable AI to be pushed to where the data resides, whether it is on-premises, in data centres, on the cloud, or at the edge.
“Organisations often assume the easiest way to run AI is to spin it out of the cloud and have the AI model run on the cloud,” he said.
This, however, will result in different replications of the data on the cloud, which is not an efficient way to run AI, he noted.
Go where the data is
“AI should come to the data,” Lim said. “You should have standardised data in one place and bring the compute to the data.”
This also provides better cost control, compared to running AI completely on the cloud, he said.
The data fabric, which connects to the company’s data sources, should run independently, regardless of whether its AI models sit on the cloud or on-premises, he added.
This also improves data governance and security, he said.
Worldwide, 53% of organisations cite data quality and readiness as the top challenge they face deploying AI, according to a study from Seagate Technology, which polled 2,712 business technology decision-makers across seven markets, including China, Japan, India, and the UK.
The data storage vendor noted that organisations are expanding their AI infrastructure priorities beyond compute to look at data foundations, to support growing storage needs, improve accessibility and governance, and scale more efficiently.
Just 38% feel fully prepared for the long-term data demands of AI, even though 99% expect the technology to drive storage requirements over the next three years.
Organisations need greater visibility into where their data resides, how it is governed, and how readily it can be accessed, Futoshi Niizuma, Seagate’s vice president of Asia-Pacific Japan sales, said in the report.
“Preparing for the next phase of AI is not simply about adding infrastructure capacity,” Niizuma said. “For Singapore-based organisations managing data and workloads across regional and local environments, building a resilient data foundation that balances accessibility, accountability, and scalability will be critical to unlocking long-term value from data.”
Organisations also want greater control over strategic and sensitive data, amidst concerns that AI will bring new risks around security, security, and compliance.
Some 33% of companies see digital sovereignty as a top or high priority, pushing 53% to want to retain control of strategic and sensitive data, revealed a study by Expereo, which polled 800 tech leaders in Asia-Pacific, Europe, and the US.
