Why Your AI Is Only as Smart as Your Data Strategy

The day before a major insurance renewal, Neil Koo walked into a meeting hoping for clarity. What he got was chaos. Three different teams. Three different versions of the same number. All of them were convinced they were right.

“None of us actually could put our hands on our hearts and say that was the right decision made,” said Koo, head of data and reserving actuary at Zurich Insurance Indonesia. That moment — high stakes, bad data, and no ground truth — became his wake-up call.

It’s a story that plays out in boardrooms across Asia Pacific — frankly, across the world. And it was the perfect opening act for the March 31 roundtable, “Navigating the Data Landscape in 2026: Turning Data Chaos into Business Value”, a sharp, no-filler conversation hosted by CDOTrends in partnership with IBM. Five data leaders; three sessions; one uncomfortable truth: the data problem was never really about data.

The numbers don’t lie (but they do embarrass)

IBM’s Institute for Business Value surveyed 1,700 data leaders across 19 industries and 27 countries. Chip Schenck, IBM’s principal strategist for Generative AI and Data, opened with three numbers that reframed the entire conversation.

First: 92% of chief data officers say they must orient around business outcomes to succeed. A near-universal consensus. Yet only 29% have clear measures to determine whether their data work is actually delivering that value. “Nine out of 10 CDOs know what success is supposed to look like, but less than three out of 10 can tell you if they’re actually achieving it,” Schenck said. That’s not a measurement failure. That’s an alignment failure.

Second: 75% of CDOs claim they have platforms enabling cross-silo integration. But only 19% of Chief Operating Officers say their organization has actually achieved enterprise-wide data architecture and scaled integration. The gap between 75% and 19% is not a technology gap. “It’s a coordination gap,” Schenck said. For most companies, enterprise-scale integration is still an illusion.

Third — and this is the number to take into your next board meeting: Organizations running centralized AI operating models deliver 36% higher AI ROI than those running decentralized operations. Not from better algorithms or faster infrastructure, but from better coordination.

Three traps. No escape hatch.

Schenck named the structural traps cleanly because naming them is the first step to working around them.

Trap one: Accountability without authority. The CDO owns the outcome. The data lives in business units, on systems the CIO controls, maintained by teams that report to someone else. When an AI initiative stalls because Finance’s data is inconsistent, nobody calls the vice president of finance into the boardroom. The CDO gets the call. And 81% of CDOs report that data preparation for AI now happens at the functional or project level, not centrally — meaning the work has decentralized, but accountability has not.

“Nine out of 10 CDOs know what success is supposed to look like, but less than three out of 10 can tell you if they’re actually achieving it.” — Chip Schenck @ IBM

Trap two: Governance can’t keep pace with business. A business unit leader needs results this quarter. They connect to a third-party source, spin up an AI agent, and call it innovation. Leadership applauds the initiative. Nobody notices the ungoverned AI program that just went live, quietly accumulating risk. “If it goes wrong,” Schenck observed, “people look to the CDO.”

Trap three: proving value before the revenue shows up. Data infrastructure ROI is measured in lagging indicators, such as AI performance, decision speed and avoided risk. None of those appear in the quarterly report the board actually reads. And only 26% of CDOs are confident their organizations can use unstructured data to deliver business value — yet every high-priority AI use case they're being asked to lead is built on exactly that.

The practitioner view: From swamp to structure

Rakshith Rao, IBM’s watsonx.data sales leader for APAC, echoed what Koo described — fragmentation compounded by cross-border complexity. “Even if you put the best data platforms in place, they cannot deliver on underlying data when it is not of high quality,” he said. And best-in-class architecture won’t save you if the data it’s processing is broken.

So what separates organizations that extract value from those that just accumulate complexity? “Class technology doesn’t create value on itself,” Rao said. “What wins every time is ownership, governance and accountability — because this is where people start seeing data as a product, not a pipeline.”

Koo put it differently. His team built what he calls a "gravity well." When you treat data as a product — with clear ownership, SLAs and continuous improvement — people stop resisting and start pulling toward it. Teams decommission their shadow systems voluntarily. “You don’t even need to ask them,” he said. “They say, I want to be onboarded to your data product.”

The shift matters: from data-as-order-fulfillment to data-as-product. It changes the entire equation.

Intelligence is not a dashboard

The second session pushed deeper. Romain Barraud, head of AI and data science at bolttech — an insurtech connecting insurance carriers, technology platforms, and device ecosystems in real time — laid out what data intelligence actually means in practice. Not just lineage and metadata. Not just governance. “Data is becoming a product. You have to think strategy. How do you sell data products to other departments and to your customers? How do you establish contracts with SLAs?”

Abhishek Ravi, IBM’s field chief technology officer for data fabric across APAC, was clear: “Your AI maturity assessment is, in a way, driven by your data maturity.” Organizations diving into agentic AI frameworks without building the prerequisite data foundations are building on sand. Bad data produces incorrect inferences. Poor outcomes follow. And in production — in front of customers — that’s not a pilot failure but a crisis.

On the question of becoming agentic-ready, Barraud issued a clear warning against the single-super-agent fantasy. “The mistake would be to have one super agent whereby we ship all our data. It cannot work. It can only create issues — hallucinations and accountability challenges.” The path forward is granular: decompose AI agents, give each the necessary data and nothing more, and build the governance structures that let them operate cleanly.

Reframe the problem. Change the conversation.

As the session drew to a close, Schenck offered the most practical reframe of the day. CDOs spend enormous energy explaining “data readiness” to colleagues who have no frame of reference for what that means. Stop. “Reframe data readiness as decision readiness. The Chief Revenue Officer has no idea what data readiness means. But if you ask them what decisions they need to make to impact the business, they’ll rattle off answers 1, 2, 3.”

That reframe is everything. It moves the conversation from infrastructure to outcomes. From technical jargon to shared stakes. From a data problem to a leadership problem — which, as every speaker at this roundtable confirmed, is what it has been all along.

The 36% ROI premium from centralized AI operating models isn’t an architecture story. “It’s a collaboration story,” Schenck said. “It’s what happens when the right people are aligned around the same outcomes, with clear enough roles that no one is waiting for permission and no one is building the same thing twice.”

Data chaos isn’t inevitable. But mistaking it for a technology problem? That one will cost you.

Image credit: iStockphoto/strelss