Why Singapore SMEs Get Stuck in the AI Pilot Phase, According to IPI's Michael Goh

Headline numbers about Singapore's AI adoption tend to flatten a messy reality: an employee using ChatGPT for a report, a pilot that ran for three months and a system embedded into daily operations all count as "adoption." IPI Singapore chief executive Michael Goh, whose agency works directly with small and medium enterprises on innovation, argues the gap between experimenting and scaling is where most Singapore businesses quietly stall.

Based on IPI's work with local SMEs, most companies have moved beyond exploration and now sit at the experimentation and piloting phase. What remains rare, Goh says, is the deliberate integration of AI into actual business processes. He lays out three questions an SME should answer before approving any AI project: what outcome is being improved and how it will be measured, whether the process map reflects what actually happens on the ground rather than what management assumes, and whether the data is available and clean enough to act on. He also urges companies to ask what happens if the AI system fails for 30 days — treating AI as a component that can break, rather than a magical black box.

The obstacles he describes are less about technology and more about organisational readiness. Projects flounder when the core business problem is poorly articulated, when documented workflows diverge from how staff actually operate, and when data is fragmented across silos or still captured on paper. Bringing employees in too late builds no internal buy-in, and staff workarounds become an early warning that a pilot will not survive contact with the real world. IPI walks SMEs through a four-stage journey — Discover, Strengthen, Scale and Expand — and finds most arrive somewhere in the middle.

Goh's framing matters because it reframes the AI adoption story from a technology problem to a process problem. The same logic underlies the wider narrative around Singapore's 51% agentic AI adoption: businesses have doubled the number of AI pilots in a year, yet only a fraction of them have redesigned the underlying processes those agents are meant to run. If Singapore's long tail of SMEs cannot convert trial runs into repeatable operations, the aggregation of those pilots will not show up in productivity or GDP terms.

Why it matters for Singapore: SMEs make up about 99% of Singapore's enterprises and employ around two-thirds of its workforce, so how they adopt AI determines whether the country's AI-driven growth is broadly shared or concentrated in larger firms. IPI's point — that execution, not access to tools, is the dividing line — points to where policy support is heading ahead of events like TechInnovation x AIMX 2026 later this month. The practical takeaway for Singapore founders and operators: the hard part is not picking a tool, but fixing the process and the data around it first.