Enterprises across Asia-Pacific have spent the past few years experimenting with AI, but moving a successful proof of concept into production remains a significant hurdle.
Speaking to iTNews Asia, Rubal Sahni, AVP, India and Emerging Markets at Confluent, said many organisations are discovering that the challenge is not necessarily the AI model itself, but whether the underlying data infrastructure is ready to support it at scale.
“A lot of people have gone ahead and started experimenting with AI as pilots or POCs, but when they tried to scale those successful POCs to production, the results were not that successful, or the projects couldn't scale,” Sahni said.
He added that the lack of ready data infrastructure is a key reason for this gap. If data is not available in real time, at scale and with low latency, AI agents can lose the context required to make reliable decisions.
“If your data is not coming in real time, which is available at scale at a very, very low latency, being shared with all the downstream applications, and the agents are not looking into real-time data, they miss the context,” he said.
That lack of context can ultimately affect the quality of AI outputs, including increasing the risk of hallucinations.
AI agents need data in context
The issue becomes more pressing as enterprises move towards agentic AI, where systems are expected to make decisions and take actions.
Sahni pointed to examples such as an AI agent booking a flight, checking code or shopping for a user. For such decisions, relying on yesterday's or even several hours-old information can make the difference between an appropriate decision and a poor one.
“If the agent is working on one-day-old data, four-hours-old data, or right now the data is up to date and contextualised for that decision, where will the better results be achieved?” Sahni said.
He also stressed that real-time streaming is not a replacement for structured, high-quality historical data as modern AI architectures require both.
Fragmented data is another barrier
The data challenge extends beyond latency. Sahni said IT leaders encounter at least three challenges around real-time data, including infrastructure limitations, uncertainty around data lineage and quality, and fragmented ownership.
Enterprise data often sits across different teams and systems, with separate owners for CRM, SAP, financial and other operational data.
When the ownership of data is not there with one team, we have seen that the projects have been delayed and data is not coming in the fashion at which it was expected.
- Rubal Sahni, AVP, India and Emerging Markets at Confluent.
For AI systems, that fragmentation can make it harder to access the right information at the right time.
Data streaming platforms, Sahni argued, can help address this by making data more trustworthy, contextualised and discoverable. “Data streaming platforms help unblock agentic AI progress by making data more trustworthy, and we can also add context on top of it.”
He said the ability to keep data continuously in motion also allows organisations to work across large volumes of data at low latency.
APAC enterprises are learning from early failures
Sahni said the tendency to invest in AI before fixing the underlying data foundation was more common about a year and a half ago. That dynamic is now changing as enterprises learn from failed or stalled AI initiatives.
“They have started now making the foundation right by having the real-time data infrastructure ready, and then applying agentic AI on top of it,” he explained.
He said this shift can be seen across APAC, from digital natives and fintechs in Australia and New Zealand to recruitment firms in Japan, banks in Singapore, gaming companies in South Korea and quick-commerce, food delivery and financial services businesses in India.
The common thread, according to Sahni, is that organisations that are further along in AI adoption are increasingly treating data infrastructure as part of the AI strategy rather than as a separate technology investment.
Data infrastructure must be an investment priority
The shift is also reflected in enterprise investment priorities. Sahni said APAC leaders consider data streaming a key investment priority, while more than 90 percent rank AI and machine learning highly.
Organisations investing in data streaming reported between two and five times return on their investment in the platform, he said.
For Sahni, however, the value extends beyond the immediate technology investment, as real-time data infrastructure can underpin multiple business and AI workloads.
Start with the use case, not the technology
For enterprises still deciding where to invest, Sahni recommends starting with the business problem rather than AI itself.
Organisations should identify where AI can have the greatest impact, test the hypothesis through a focused proof of concept and build the supporting data and governance layer in parallel.
“I have to know in my business which are the maximum-impact areas, and I should pick a use case there. Just because everybody is using AI and leveraging AI, I shouldn't be starting just like that,” he explained.
The approach is intended to prevent enterprises from proving an AI model only to discover that their infrastructure cannot support it when deployed at scale.
Look at data as an enterprise foundation, and not just a single AI project
The wider opportunity, Sahni argued, is to stop treating data infrastructure as something built for a single AI project.
He pointed to a Confluent customer where a data streaming deployment initially focused on one requirement eventually uncovered opportunities across security and business operations.
The organisation reduced its SIEM costs by 50 percent and improved mean time to respond by 35 percent within weeks, he said. The same platform was subsequently used to improve the timeliness of business data and give management better visibility into customer issues.
“What we started as a one-use-case entry into an enterprise, they saw multiple-fold benefits, and they just reaped benefits by the same platform across the enterprise,” he explained.
For APAC enterprises, the next phase of AI adoption will depend on whether organisations can make their data sufficiently real-time, contextual, trustworthy and governed to support them. The goal is to invest in the foundation once and use it to unlock multiple business outcomes across the enterprise.
