Singapore leads APAC in AI agent deployment, but reliability is becoming the harder test

Singapore enterprises are among the fastest in APAC to move AI agents into customer communications, but Sinch’s latest research suggests that getting these systems live is proving easier than keeping them reliable.

The company’s The AI Production Paradox report found that 72% of Singapore enterprises have AI agents in production, the highest rate in APAC. Yet 82% have already rolled back or shut down a live AI customer communications agent, eight percentage points above the global average.

That gap changes the way enterprise AI adoption should be read. In customer communications, the issue is no longer simply whether companies can move AI projects beyond testing. For markets such as Singapore, the bigger challenge is whether live systems can handle customer interactions consistently, safely and with enough visibility when something goes wrong.

Sinch defines AI agents as autonomous or semi-autonomous systems with capabilities beyond basic chatbots. Its research was based on an independent survey of 2,527 senior decision-makers across 10 countries and six industries, including 586 respondents from APAC and 152 from Singapore.

Deployment has moved ahead of control

Globally, 62% of organisations already have AI agents live across customer communications, while 88% expect to be in production within 12 months. APAC is moving faster, with 67% already in production and 92% expecting to have an AI agent live by the end of the year.

That speed has brought the region into a more difficult phase of adoption. Across APAC, 83% of enterprises have experienced an AI agent failure, above the global rate of 74%. The report also found that APAC organisations are deploying AI across an average of 3.6 customer channels, compared with 3.3 globally.

The breadth of those deployments matters because customer communication is not a contained back-office use case. In APAC, enterprises are integrating AI agents across website and in-app chatbots, email, WhatsApp, social media, SMS or MMS, and voice or interactive voice response. When an AI agent fails across that mix, the result can spill into customer service queues, brand trust and engineering workloads at the same time.

Sinch found that 45% of APAC enterprises cited support team overload as the main consequence of AI failure. In Singapore, the figure was 44%. The risk is sharper for large-scale senders, with the release noting that one in three APAC enterprises sends more than 100 million messages per month.

Singapore’s caution is about reliability, not retreat

Singapore’s deployment lead does not mean companies are pursuing AI expansion at any cost. The report found that 40% of Singapore enterprises plan to increase AI communications investment by more than 25% compared with the previous year, the lowest rate globally in the survey.

That contrasts with India, where 71% plan to increase investment by more than 25%, and Australia, where the figure is 67%. But the Singapore data points less to weak appetite and more to a narrower investment focus.

75% of Singapore enterprises are prioritising trust, security and compliance in AI investment. At the same time, 44% of Singapore respondents identified context as AI’s biggest advantage, compared with 37% globally, while cost reduction ranked last at 3%.

Taken together, those figures suggest that Singapore companies are not treating AI agents mainly as a way to reduce service costs. The priority is whether these systems can understand customer context, maintain trust and avoid failures that create more work for support teams.

Wendy Johnstone, Executive Vice President, APAC at Sinch, said: “APAC is the most advanced region in the world for AI-powered customer communications, but it is also where the gap between deployment and reliability is widest. Our findings reveal that while Singapore is a market that is leading the region on deployment, local business leaders are also approaching AI very deliberately. They are purposefully channelling resources toward reliability rather than rapid expansion, aligning with sustained government-led efforts to promote reliable and responsible AI deployment. This suggests that enterprises here aren’t investing less because they lack ambition, they’re investing selectively because they understand the stakes. The real risk across APAC isn’t moving slowly, it’s scaling on infrastructure that can’t keep up.”

Rollbacks are exposing weak operating foundations

The rollback figures also complicate the assumption that governance maturity alone will solve enterprise AI failures. Globally, Sinch found that 74% of enterprises that deployed AI in customer communications have been forced to roll back, and the rate rises to 81% among organisations with fully mature guardrails.

Singapore’s own governance base remains uneven. Only 27% of Singapore enterprises reported fully mature guardrails, the lowest rate in APAC and eight percentage points below the global average. Across APAC, Sinch found that the link between governance and AI advancement was 48% stronger than the global average.

This means governance is still necessary, but it has to work with the infrastructure that carries customer interactions. Without strong audit trails, reliable routing, context management and data controls, companies may struggle to understand why an AI agent failed or prove that the issue has been fixed.

The provider gap is already visible. In Singapore, 82% of organisations rated high-performance infrastructure as essential or very important, but only 7% said their current communications provider was fully meeting their needs. As a result, 91% of Singapore enterprises are evaluating communications providers, above the global average of 86%.

Across APAC, 88% of organisations rated high-performing infrastructure as essential or very important, while 93% identified at least one shortcoming in their current provider. Sinch’s full report also described communications infrastructure satisfaction as the strongest predictor of AI deployment success across the dataset.

Johnstone added: “When less than a tenth of Singapore enterprises say their communications provider is fully meeting their needs, the issue isn’t with the technology, but with the digital infrastructure that these AI tools and solutions run on. Across APAC, engineering teams are rebuilding safety systems their provider should already offer. That’s the hidden cost holding the region back from effectively scaling AI.”

For Singapore enterprises, the next phase of AI customer communications will be judged less by how many agents are launched and more by how well they can be governed, audited and recovered when they fail. The deployment race has already begun and the reliability test is now harder to avoid.