Southeast Asia’s AI surge is unmistakable. According to McKinsey’s 2026 findings, nearly half of regional firms have moved beyond pilots. In a separate study, the Google-Temasek-Bain e-Conomy SEA 2026 report claims that 243 million people now use generative AI, accounting for 58% of the region’s working population.
The caveat is that our ambition to adopt AI is outpacing our ability to coordinate its use, exposing our organisations to multiple risks, including runaway costs and compliance risks.
Forrester’s 2027 Asia Pacific predictions warn that one in three large enterprises will uncover at least ten employee-built shadow AI agents for every sanctioned one, spawning unmanaged risk, duplicated tools and undocumented dependencies.
The result is decision debt: the quiet accumulation of fragmented outcomes that erodes enterprise performance.
For CIOs, the task is no longer simply scaling AI but linking initiatives into a coherent decision strategy that balances autonomy with oversight.
The anatomy of a fragmented AI estate
Shadow AI is already widespread as AI experimentation moves into production and AI capabilities become embedded in productivity tools, observes Deepak Ramanathan, vice president and head of Customer Advisory (Asia Pacific) at SAS. Organisations are now seeing duplication of tools and work, as well as rising costs. While businesses recognise AI’s productivity benefits, “the challenge is operationalising AI within existing applications while maintaining human oversight and managing risks such as runaway agents.”
The fragmentation is not random; it follows organisational fault lines. Different departments deploy AI to solve their own priorities—finance focuses on cost optimisation, marketing on customer acquisition, and risk on reducing exposure.
When these systems operate independently, they can produce conflicting recommendations, such as marketing targeting a customer whom risk considers high-risk. Ramanathan notes this creates “inconsistent decisions, duplicated analytical work and potentially poor customer experiences.” The very specialisation that makes departmental AI effective becomes a liability at the enterprise level.
Decision debt defined
Ramanathan defines decision debt as what accumulates “when organisations continue relying on decisions that no longer reflect changing market realities or regulatory requirements.” In financial services, for example, fraud, risk and marketing teams may each make decisions independently.
As regulations and AI governance requirements evolve, organisations must quickly realign these siloed decisions. Without a unified view, outdated or disconnected decisions become a growing organisational burden.
The parallel with technical debt is instructive: just as code accumulates hidden maintenance costs, fragmented AI decision-making accumulates hidden governance and coordination costs that compound over time.
The region’s adoption trajectory amplifies the challenge. The e-Conomy SEA 2026 report projects that as people in Southeast Asia move from simple searches to delegating complex tasks to AI agents, regional AI usage could surge by up to 60-fold by 2030.
Consumer accounts for about 60% of regional tokens, while enterprise use has significant headroom to scale. Without coordination, every new deployment adds to the debt.
Semantic intelligence as the connective tissue
How can CIOs connect individual AI use cases to a single enterprise decision strategy? Ramanathan points to semantic intelligence, which “creates an enterprise knowledge layer across data and applications.”
The foundation is reliable data, followed by data governance, lineage and cataloguing. The next step is connecting information across applications such as HR, finance, marketing and risk. AI can then provide context across these systems, enabling organisations to understand enterprise-wide decisions and build agents with human oversight.
This approach does not require wholesale replacement of existing infrastructure. Organisations need a centralised knowledge repository and the ability to look across existing applications.
Ramanathan identifies emerging protocols such as MCP and agent-to-agent protocols that “can allow applications and agents to communicate with one another,” enabling agents to connect systems such as fraud and risk applications.
This builds greater context across the organisation without necessarily replacing existing systems—an important consideration for Southeast Asian enterprises with substantial legacy investments.
The governance imperative
Forrester’s prediction of ten shadow agents for every sanctioned one underscores the governance gap. Ramanathan recommends that organisations “establish centralised mechanisms to track, monitor and register AI agents, including maintaining metadata about their use.”
AI use cases should also align with organisational AI policies. CIOs need visibility not only into which use cases exist and whether they comply with policy, but also into their costs. “This creates greater control as AI adoption expands across different business divisions.”
The governance challenge extends to rationalising existing deployments. Organisations should conduct an inventory sweep of their AI initiatives to identify projects that have completed their intended purpose but continue operating and incurring costs.
The review should also consider where AI runs and whether data sovereignty requirements are driving workloads on-premises. This helps organisations rationalise their AI estate while addressing cost and data considerations. Forrester’s 2027 predictions echo this concern: one in five large, regulated enterprises in APAC will approve Chinese open-weight models as minimum viable sovereignty expands beyond data residency to model and inference control.
Where CIOs start
The starting point is inventory. Ramanathan advises CIOs to “first inventory the applications running across the organisation and understand what each one does,” then “identify how information and capabilities in one function could benefit other functions, such as marketing, finance and risk.”
These building blocks can then be connected into a semantic layer. Over time, agents can automate understanding applications, their context, and the decisions they support.
CIOs are well positioned to drive this change because application owners typically report into the CIO organisation. They can explain the benefits of connecting applications and building enterprise knowledge.
A unified view can also reduce duplicated decision-making and associated costs, while helping ensure AI use cases are deployed in line with organisational AI policies and governance frameworks. Data platforms provide three critical capabilities: governance, lineage, and cataloguing.
These help organisations understand where data originates and how it is used in decision-making. In regulated environments, this is particularly important because organisations need to explain why they made a decision.
A complete view should show what data and AI agents were involved, how they were governed and how the final decision was executed.
Turning adoption into advantage
The e-Conomy SEA 2026 report notes that Southeast Asia’s AI adoption is scaling four times faster than the mobile internet in under three years. Yet the report also observes that “the next opportunity lies in translating that reach into deeper usage and economic value”. Decision debt represents precisely the gap between adoption and value.
Ramanathan argues that a semantic intelligence layer can provide an enterprise-wide ontology and connect existing applications without requiring extensive infrastructure replacement. It helps organisations build enterprise knowledge while making decision debt, data debt and technical debt more visible.
The focus should be on enabling AI within existing applications, building productive agents and extending the value of legacy systems through generative AI.
The imperative is clear. As the Google-Temasek-Bain report warns, consumers are increasingly willing to let AI agents act on their behalf—75% say they’d let AI agents autonomously compare prices and complete transactions.
Enterprises that cannot coordinate their own AI decisions will find themselves outmatched by those that can. The region’s AI ambition is not in question; what remains to be seen is whether CIOs can build the connective tissue that turns fragmented experimentation into coherent enterprise advantage.
For Ramanathan, the path forward is incremental but deliberate: inventory, connect, govern. The alternative—accumulating decision debt while competitors build semantic intelligence—is a risk that no enterprise can afford to ignore.
Click the PodChats player to hear Ramanathan’s perspective on the hidden risk of fragmented enterprise AI.
- How widespread is shadow AI use among enterprises in Southeast Asia today, and what risks has it already created?
- Where have departmental AI tools produced conflicting recommendations or duplicated work?
- What is decision debt, and what does “decision debt” look like in practice across finance, risk and operations (including IT departments)?
- We are seeing the rapid development of AI use cases, many of which are developed in silos. How should CIOs (and other leaders within the organisation) connect individual AI use cases to a single enterprise decision strategy?
- If I could go a little deeper into this Semantic intelligence you are referring to. Over the years and decades, we have built siloed intelligence. What now needs to happen to create this enterprise-wide semantic intelligence?
- If the owner of intelligence or insight sits with the functional practitioners, where does the CIO sit in this semantic intelligence paradigm?
- What role should data platforms and decision intelligence play in reducing fragmentation?
- We are starting to see calls for the CIO to provide context for how AI has arrived at a specific conclusion or course of action. Is this the new norm in the age of the AI enterprise?
- How should CIOs balance autonomous agents with human accountability as shadow AI rises?
- Coming back to this recognition that shadow AI will continue to proliferate in the enterprise, and with it the risk of unused AI consuming resources and therefore costing companies unnecessary costs, what practical steps can CIOs use to identify and retire fragmented AI initiatives?
- In 2027, how can Southeast Asian CIOs turn rapid adoption into coherent advantage rather than accumulating more decision debt?
- What is your advice for CIOs as their first few steps towards building/layering in semantic intelligence?
