There is no shortage of ambition around artificial intelligence in Singapore. The country has a national AI strategy, significant public investment in AI research and talent, an expanding ecosystem of AI companies and increasingly detailed guidance on how organisations should govern the technology.
Yet the picture inside businesses is considerably less mature than the public conversation might suggest.
Singapore's Ministry of Manpower found that 71.5% of firms had not adopted AI. Among the 28.5% that had, only 3.8% were integrating AI into their core processes. The remainder were still planning, piloting or implementing AI in more limited areas. The study covered private-sector establishments with at least 10 employees.
That gap deserves more attention.
It suggests that the next stage of Singapore's AI journey will not be defined simply by how many organisations experiment with AI. The more consequential question is whether those experiments become embedded in how organisations operate, make decisions and create value.
The language around AI adoption can make the transition sound deceptively straightforward. An organisation identifies a use case, selects a technology, runs a pilot and eventually moves into production.
In practice, the transition is much more complicated.
Singapore's AI for Enterprise Impact Playbook reflects this reality. Developed by IMDA, SkillsFuture Singapore and Workforce Singapore, and informed by engagements with more than 1,000 enterprises, the framework assesses organisations across strategy and leadership, talent and culture, data and governance, technology deployment and integration, and value creation.
The significance of this framework is that it does not treat AI deployment as a standalone technology decision. It places AI within a broader organisational context.
That distinction matters because an organisation can deploy an AI tool without fundamentally changing the way work is done. A chatbot can sit on top of an existing customer service process. A generative AI tool can be provided to employees without changing their responsibilities. A predictive model can produce recommendations that still depend entirely on an existing approval process.
These are useful applications, but they are not necessarily transformation.
The relatively low proportion of Singapore firms integrating AI into core processes suggests that this transition remains difficult.
The reasons are not necessarily technological.
They can involve the organisation's existing processes, skills, data foundations, technology environment and ability to measure value. These are precisely the dimensions reflected in Singapore's enterprise AI framework.
There has been a tendency in recent years to treat generative AI as a technology that can somehow bypass the traditional problems of enterprise data.
The evidence does not support that assumption.
Singapore's refreshed National AI Strategy identifies access to the right datasets for specific AI use cases as an important priority, alongside safeguards for privacy, security and legitimate commercial interests. The strategy also identifies AI testing, assurance and safety as areas requiring greater capability.
Research from Singapore's Ministry of Trade and Industry offers another useful perspective. Its analysis of firm-level AI use found that AI adoption tends to build on firms' existing adoption of foundational and complementary digital technologies. Larger firms and firms in digitally intensive sectors were also more likely to use AI.
This is an important corrective to the idea that AI represents a completely new starting point for organisations.
For many businesses, the ability to use AI effectively is likely to depend in part on what has already been built: the quality of their digital systems, the availability of data, the way information moves across the organisation and the capabilities that already exist within their workforce.
AI does not remove those foundations. It makes them more visible.
The governance challenge becomes even more significant as organisations move from generative AI towards agentic systems.
A generative AI application might help an employee analyse information or draft a response. An agent can potentially interact with systems, use tools, retrieve information and take actions on behalf of a user.
Singapore has been examining these issues directly.
The Singapore Government and Google conducted an AI Agents Sandbox to explore the behaviour of computer-use agents in realistic environments. The work identified challenges involving human oversight, cybersecurity, privacy, identity and authentication, permissions and access controls. It also highlighted the fact that much of today's digital infrastructure was designed around human users rather than autonomous software agents.
Singapore subsequently updated its Model AI Governance Framework for Agentic AI, incorporating feedback from more than 60 organisations. The framework emphasises human accountability and recommends that organisations assess and manage the risks associated with different levels of agent autonomy.
This development has implications well beyond AI ethics.
It touches data architecture and enterprise architecture directly.
Traditional data governance has often focused on ownership, access, quality, lineage and appropriate use. When software can independently retrieve information, invoke tools and execute actions, governance has to account for the behaviour of the system as well as the information it can access.
That does not make existing data governance obsolete. It makes the relationship between data governance, AI governance, identity, security and architecture considerably more important.
There is evidence that businesses are already seeing benefits from AI.
MOM's research found that 70.7% of firms using AI reported improvements in worker productivity. Firms also reported improvements in decision-making and innovation.
At first glance, that sounds encouraging. But the same research shows that deep integration remains limited.
International research also suggests that the productivity effects of AI are more complicated than some of the early enthusiasm implied.
A March 2026 NBER study based on a survey of nearly 750 corporate executives found that executives reported positive labour productivity gains from AI, but also identified a gap between perceived productivity improvements and measured gains. The researchers suggest that some benefits may take time to appear in revenue and other conventional measures of performance.
Another NBER study, published in August 2026, linked AI investment with productivity growth and highlighted the role of organisation-specific knowledge and capabilities in capturing those gains.
Taken together, this points towards a more nuanced view of AI value.
The technology may create productivity gains, but the scale and durability of those gains can depend on how effectively an organisation incorporates the technology into its existing capabilities.
That makes the management question considerably more interesting than simply asking whether AI works.
This is perhaps where Singapore's AI conversation needs to go next.
The country has already made substantial investments in AI capability. Its refreshed National AI Strategy includes sectoral AI transformation, AI research, talent development, data governance and trusted AI adoption. Singapore has also committed more than S$1 billion to public AI research and talent development between 2025 and 2030.
The technology and policy foundations are being built.
The harder part is what happens inside individual organisations.
The evidence suggests that many businesses are still at an early stage of that journey. At the same time, organisations that have adopted AI are reporting measurable benefits, while Singapore's enterprise frameworks increasingly treat leadership, talent, data, governance and value creation as integral components of AI transformation.
That leaves us with a more useful way of thinking about AI maturity.
The question is not simply whether an organisation has adopted AI. It is whether AI has become part of the organisation's operating model.
That distinction matters.
The first generation of enterprise AI was largely about finding useful applications. The next stage will require organisations to connect those applications to their data architecture, decision processes, governance structures and workforce.
For data and AI leaders, this is where the disciplines increasingly converge. Data management provides the foundation. Analytics turns data into insight. AI extends what can be done with that insight. Architecture determines how these capabilities fit into the enterprise. Governance establishes the boundaries within which they can operate. Management determines whether any of it produces lasting organisational value.
Singapore's AI journey therefore presents an interesting paradox. The country has developed considerable momentum around AI, yet the evidence suggests that many organisations are still working through the transition from experimentation to integration.
That may be the more important story to watch over the next few years.
Not how many AI tools organisations deploy, but how deeply AI becomes embedded in the way they work.
Join us at CDAIO, Enterprise AI, and Data & AI Architecture Singapore 2027, 13-14 April to learn more from data analytics and AI leaders. Reach out to Eleen Meleng for more information.