For Asia-Pacific CIOs, the path to AI ROI lies not in the next model, but in the governance and orchestration layers that make it enterprise-ready. A Gartner survey of 782 infrastructure and operations leaders delivered a stark reality check: only 28% of AI use cases fully meet ROI expectations, while a further 20% fail outright. The bottleneck, however, is not the model itself. The challenge lies in the governance and orchestration layers beneath it.
For CIOs across Southeast Asia and Hong Kong, the conversation has shifted decisively from experimentation to what Gartner calls “industrialisation of AI.” The mandates for 2026–2027 are clear: tame agent sprawl, mitigate shadow AI risk, and build a robust, platform-agnostic control plane.
The ROI mandate: From experiments to measurable outcomes
The pressure to demonstrate tangible value from AI investments has never been more acute. “When we talk to a lot of CIOs, heads of IT, there’s a lot of concern around measuring ROI, measuring value of AI,” observed Celine Siow, VP of sales and GM for APAC at Workato.
Her advice grounds the conversation in fundamentals: “It goes back to what a lot of CIOs know. It is data. It is process. It is people.”
This echoes Gartner’s findings that successful projects are those integrated into existing workflows and backed by strong executive support. Yet the path to ROI is fraught with structural challenges. A key issue identified by Gartner is organisations trying to “force AI into processes where it simply doesn’t add value,” leading to overly ambitious projects and “disappointing returns”. The failure is often one of scope and preparation, not technology.
For APAC businesses, this is a particularly acute problem. While Singapore has been deliberate in positioning AI as a long-term economic capability through its National AI Strategy 2.0, the operational reality is sobering.
The Ministry of Manpower’s inaugural 2026 report found that 71.5% of firms in Singapore have yet to adopt AI at all. Even among adopters, only 3.8% have integrated AI into core business processes, with many stuck in planning or pilot stages. The opportunity is clear, but the execution gap is vast.
Taming agent sprawl and shadow AI
As enterprises move beyond simple chatbots to autonomous agents, a new “technical debt” emerges: agent sprawl. This uncontrolled proliferation of AI agents across an organisation without centralised inventory, ownership, or governance is becoming a board-level concern.
Siow uses a vivid culinary metaphor to illustrate the chaos:
Celine Siow
“Your personal chef is your agent, a very capable one. And then as you progress, what happens is your wife decides to bring in another personal chef. Your guest decides to bring in their own personal chef and maybe some line cooks without you knowing. What’s going to happen to your kitchen? It’s going to be a messy one.”
Shadow AI exacerbates this chaos. An Adaptive Security report reveals that 80% of employees use unapproved generative AI tools at work, yet only 12% of companies have a formal AI governance policy in place.
Workato’s Siow views this as a “demand signal” and a “ticking clock of liability.” Rather than blocking adoption, which she notes is futile, CIOs must empower workers “safely with governance, with control and with proper management and permissions.”
The need for a vendor-neutral control plane
The critical question for CIOs is how to architect a system to govern this growing fleet of autonomous agents. Siow defines a “neutral control plane” as being “platform agnostic… a unified way of looking down and seeing what exactly is happening to all your AI and IT architecture.” This avoids the fragmentation created by a vendor-locked control plane, which Siow warns “just recreates fragmentation.”
This requires an orchestration layer that serves as the “executive head chef,” ensuring all agents act in line with business goals and security protocols. This layer must address the governance gaps found in protocols like the Model Context Protocol (MCP), which, despite being a valuable open standard, is “plumbing” according to Siow.
The Cloud Security Alliance’s research corroborates this concern, highlighting that MCP’s structural governance gaps—including “tool poisoning” and “rug-pull” attack vectors—create “accountability voids at the infrastructure layer of agentic deployments”.
More than 30 MCP-related CVEs were filed in early 2026 alone, underscoring the urgency of moving beyond mere connectivity to robust governance. Siow’s vision of Workato’s enterprise MCP is clear: “Essentially, we’re your kitchen ledger. We are your kitchen video auditor. We are your kitchen bouncer.”
Building the governance and FinOps framework
To move from theory to practice, Siow outlines a multi-step governance framework. It begins with investigating the root causes of shadow AI, then building an internal “app store” for approved tools and creating a risk-based approval process.
Crucially, she advocates for an AI Centre of Excellence (CoE) that includes business line managers and even AI-literate interns to foster a culture of innovation, not just a top-down mandate.
A high-performing AI CoE is not an administrative committee but a “centralised function that owns the standards, infrastructure, governance, and institutional knowledge that allow AI to be built and deployed reliably across the organisation,” and it must be chartered with clear decision rights and measurable KPIs.
Furthermore, AI FinOps is emerging as a critical discipline. With token consumption set to grow exponentially and costs predicted to run 30% higher than planned, traditional cost models are obsolete. Siow points to “interoperability” and “token consumption” as key cost drivers.
For CIOs designing their AI architecture for 2026 and beyond, the advice is to think beyond technology. “Adoption of technology can only happen with people’s acceptance,” Siow argues. She stresses that culture and change management are frequently overlooked. “It is about how we can drive this towards what we’re trying to achieve and align with the company’s goals.”
To succeed, CIOs must treat governance as a day-one architectural principle, not a retrospective compliance exercise. As Siow concludes:
“We have to start today. Don’t hold back. I understand that culture can be painful and people can be difficult to change. It’s human nature. But start now and iterate later as you go along.” Celine Siow
Click on the PodChats player to listen to Siow elaborate on the key trends that are shaping the CIO agenda for 2026-2027.
With only 28% of AI projects meeting ROI targets, how can we build a formal “value playbook” to ensure our AI investments deliver tangible, measurable business outcomes?
As “agent sprawl” becomes the new technical debt, how can we architect a neutral control plane to govern, orchestrate, and observe our growing fleet of autonomous agents?
Given that 80% of workers now use unapproved AI tools, how can we move from simply blocking “shadow AI” to enabling governed, safe adoption across the entire workforce?
What does an effective governance framework look like when AI moves beyond content generation to action management, requiring real-time control over permissions, escalation, and audit trails?
Given that only 3.8% of Singaporean firms have integrated AI into core processes, what is the most effective strategy to upskill our workforce and bridge the critical “AI Bilingual” talent gap?
What is your advice for CIOs? They’re looking to design their AI architecture to be integrated with their current environment, and they must orchestrate the running of AI and the adoption of AI across the enterprise. What is the one or two pieces of advice that you have for these executives?
