51% of Singapore Firms Now 'Use' Agentic AI. Look Closer.

Here's a statistic that's going to appear in a lot of pitch decks this quarter: 51% of Singapore enterprises have now adopted agentic AI, up from 22% a year ago. That's from ServiceNow's Enterprise AI Maturity Index, published this week, based on a survey of 4,500 senior leaders across 19 countries (200 of them in Singapore, so treat the local cut as directional rather than gospel).

More than doubled in a year. On the surface, a stampede.

Now the number from the same report that won't make the pitch decks: 10%. That's the share of Singapore enterprises that have actually redesigned a process so AI completes multi-step work end to end. The largest group, 33%, is using AI to help individual employees with their daily tasks. Another 18% report no progress on advanced AI adoption at all.

So half of Singapore's enterprises have "adopted agentic AI", and one in ten has changed how any actual work gets done. CK Tan, ServiceNow's innovation officer for Asia-Pacific and Japan, put it more diplomatically than I will: "adoption and transformation are not the same thing."

I'd go further. Most of what's being counted as adoption is a subscription.

What "we use AI" means when I actually go and look

I run an agency that builds AI into SME operations, so I sit through a lot of first conversations that start with some version of "oh, we're already using AI." I've learnt to ask the follow-up question: show me.

What I usually get shown is ChatGPT open in a browser tab. Sometimes several tabs (each staff member with their own account, on their own personal login, which is a separate article I keep threatening to write). Someone in marketing uses it to draft posts. Someone in ops uses it to rewrite emails to sound less angry. The boss used it once to summarise a contract and was impressed.

None of that is fake. It's genuinely useful, people genuinely like it, and if a survey rang up and asked "has your organisation adopted AI?", the honest answer would be yes. That yes is how you get from 22% to 51% in a year.

But notice what hasn't changed: every process in the company still runs exactly the way it ran before. Same steps, same handoffs, same bottlenecks, same person doing the same job with a slightly faster keyboard. The AI is sanding the edges of individual tasks. The workflow it sits inside has never been touched.

That's the 33%. And I want to be fair to them: individual assist is the right first step, and anyone claiming SMEs should leap straight to autonomous agents is selling something. Starting there is sensible. Stopping there is the trap: assist mode has a low ceiling, most organisations settle in just under it, and the survey lets them tick the same box as the ones who kept going.

Why so few make it to the 10%

The comfortable explanation is that redesigning a process around AI is technically hard. It's mostly not. The models are capable, the tooling has become almost boringly accessible, and the integration work that needed a six-week quote two years ago is often now an afternoon.

The real blocker, in nearly every rollout we've done at Magnified, is that the process to be redesigned doesn't exist anywhere except in someone's head.

I wrote back in February that SOPs had quietly become AI infrastructure, and six months of client work since has only hardened that view. When we take on an automation project, the first deliverable is almost never code. It's a document: what triggers this work, what the steps actually are, what a good outcome looks like, what the exceptions are and who decides them. Writing that down routinely takes longer than wiring up the agent that follows it, and more than once the interviews have surfaced the awkward discovery that the two staff who each held the process in their heads had been running two different processes.

An AI agent given a documented process is a tireless, consistent operator. An AI agent given "just do what Susan does" is a liability with API access.

This is why I don't fully buy the framing that the 33% are "on a journey" to the 10%. Assist mode requires no documentation, no process ownership, and no decisions about what the workflow should be. End-to-end mode requires all three. There's no gradual slope between them; there's a step, and the step is organisational, not technical. Companies don't drift across it by renewing their subscriptions another year. They cross it the month someone senior decides a specific process is going to be rebuilt, writes down how it actually works, and accepts ownership of the result.

For what it's worth, the report contains a genuinely encouraging sign on this front: 23% of Singapore enterprises are replacing legacy systems to support AI, against 16% globally, and 28% have formal AI risk processes versus 20% globally. Some organisations here are doing the unglamorous structural work. Just not most of the ones in the adoption headline.

The number I'd watch instead

The SME picture makes the same point at smaller scale. IMDA's Singapore Digital Economy Report put SME AI adoption at 14.5% in 2024, up from 4.2% the year before. Tripled. And I'd bet the composition is the same: overwhelmingly individual staff using assistants, a thin slice actually restructuring how work flows through the business.

If it's any consolation, this isn't a Singapore failing. Singapore's overall maturity score in the ServiceNow index is 53 out of 100, above the global average of 51. The whole world is at the same awkward stage: budgets more than doubling year on year, adoption curves going vertical, and transformation curves barely off the floor.

But it does mean the adoption statistics have stopped telling us anything useful. When "adopted agentic AI" stretches to cover both a company that rebuilt its quoting workflow around an agent and a company where three people have Copilot licences, the number is measuring enthusiasm, not change.

So here's the question I'd put to any organisation feeling reassured by that 51%, and it's the same one I ask prospects who tell me they're already using AI: name one process that runs differently today because of it. Not faster typing. Differently. Fewer handoffs, a step removed, a decision that no longer waits for a person.

If there's an answer, the subscription money is working. If there isn't, the organisation is in the 33% wearing the 51%'s badge. Call it AI-adjacent: a written-down process and one decision away from the real thing.