Singapore fintechs are adopting artificial intelligence (AI) at a faster rate than their domestic and global counterparts, but widespread deployment has yet to translate into deeper business integration. While 86% of surveyed firms have deployed AI, only 4% identify new revenue generation as its biggest benefit, and just 11% have introduced agentic AI into live environments.
Table of contents
The findings come from The AI Inflection Point: Rewiring Fintech, a report by Kapronasia in collaboration with Airwallex and the Singapore FinTech Association (SFA), released at the Singapore FinTech Festival on 18 November 2026. The study surveyed 100 Singapore fintech leaders and interviewed 10 senior practitioners between July and September 2026.
Functional area of respondents n=100
- Payments & wallets29%
- WealthTech20%
- Digital Assets13%
- Others12%
- Banking Infrastructure & Enterprise software11%
- Lending & Credit8%
- InsurTech7%
Seniority levels of survey respondents n=100
Despite the gaps, 94% of respondents expect their AI budgets to increase in 2027. The findings raise questions about how effectively firms are translating investment into measurable results and whether their operating models can support more advanced AI deployment.
Widespread AI adoption masks uneven integration
Singapore fintechs recorded an 86% AI deployment rate, compared with 65% among Singapore financial institutions and 62% among fintechs globally. The comparison figures come from separate Finastra and NVIDIA studies, rather than the same survey population.
Kapronasia's AI maturity assessment provides a more detailed picture. On a 100-point index, adoption breadth scored 80 and infrastructure and data readiness reached 67. Integration depth was lower at 54, followed by AI autonomy at 42 and workforce adoption at 39. These are maturity scores, rather than percentages of firms.
Nearly six in 10 respondents reported that AI was used by half or less of their workforce, while only 10% said between 71% and 100% of employees used it in daily work. Deployment also differs considerably across business functions. Technology and engineering recorded 99% AI adoption, followed by operations and back-office functions at 98%. Financing, lending and underwriting remained less developed, with approximately six in 10 respondents reporting no AI adoption in that function.
The gap extends to deeper integration. Technology and engineering recorded 94% integrated AI use, compared with 61% in sales and marketing and 29% in financing, lending and underwriting.
Widespread availability of AI has not translated into widespread workforce adoption
% of AI deployed across functions
- AI deployed
- No AI
| Business function | AI deployed / No AI | n= |
|---|---|---|
| Technology & engineering | 99 | |
| Ops & back office | 99 | |
| Workflow automation & ops | 100 | |
| Interface & advisory | 100 | |
| Risk, compliance, & security | 99 | |
| Finance & treasury | 98 | |
| Customer service | 100 | |
| Sales & marketing | 99 | |
| Identity, AML, & trust security | 96 | |
| Transaction movement & settlement | 97 | |
| Human resources | 96 | |
| Financing, lending, & underwriting | 99 |
How widely is AI used in employees’ daily work? n=100
- <50% of workforce
- 51-60% of workforce
- 61-70% of workforce
- 71-100% of workforce
These differences suggest that fintechs have made greater progress in structured technical activities while exercising more caution in functions where AI can influence financial decisions or customer outcomes. Broad deployment therefore offers an incomplete measure of maturity without evidence of consistent employee usage and integration into core processes.
AI investment grows despite weaker reported returns
When asked to identify AI's biggest benefit, 41% of respondents selected operating cost reduction, followed by risk mitigation and compliance at 28% and employee efficiency at 19%. Customer retention and experience accounted for 8%, leaving just 4% who identified net-new revenue creation as the primary benefit.
The 4% figure does not mean that only 4% of firms have generated revenue from AI. Instead, it shows that operational improvements remain more prominent than revenue creation when respondents assess the technology's main contribution.
The difference shows that cost savings from automation can be easier to measure, while revenue growth depends on several factors beyond AI. Investment returns also fell short of expectations for many respondents. While 61% anticipated returns of between 26% and 50%, only 42% reported achieving results within that range. Meanwhile, 45% reported returns of between 1% and 25%, compared with 21% who had expected returns in that lower range.
These findings suggest that some firms may need to reassess how they evaluate AI investments, particularly when benefits such as faster product development and improved customer service are difficult to express through conventional financial measures.
Airwallex, for example, measures AI's contribution through deals closed in sales and service-level agreement throughput in operations. At Endowus, an AI-powered support bot handles approximately 30% of WhatsApp and in-app customer enquiries, allowing advisers to focus on more complex interactions. The report attributes improved client satisfaction primarily to faster resolution times rather than better answers from AI.
Such examples demonstrate how operational value can be measured against specific outcomes. With most surveyed firms planning further investment, establishing those measures will become increasingly important for assessing whether higher AI spending is justified.
Agentic AI remains limited by operational risk
The gap between capability and deployment becomes more pronounced with agentic AI, which can execute multi-step processes and take actions with less direct human involvement. Although 42% of surveyed fintechs considered themselves ready to implement agentic AI, only 11% had deployed it in live environments.
Even in departments with extensive AI adoption, autonomous capabilities remain relatively limited. Agentic and multi-agent systems accounted for 28% of reported AI use in operations and back-office functions and 26% in technology and engineering. The corresponding proportion was just 5% in financing, lending and underwriting.
This pattern reflects the greater consequences of errors when AI moves from assisting employees to executing financial decisions. Endowus and StashAway, both regulated by the Monetary Authority of Singapore (MAS), use secure testing environments to evaluate agentic capabilities rather than delegating core financial advice directly to autonomous systems.
Agentic AI is advancing, but human oversight remains the guardrail to autonomous execution
Controls required for agentic AI deployment, n=100
Nevertheless, six in 10 respondents expect agentic AI to become mainstream within two years, although interviews with senior practitioners found that many lacked a clearly defined deployment roadmap. The operational challenge extends beyond selecting more capable models. Firms must determine which activities can be delegated, when human approval is necessary and how responsibility is maintained when automated systems make mistakes.
As Uthra Parameswaran, Chief Compliance and Risk Officer at StashAway, observed: “Making an existing task faster is automation, not workflow redesign.”
Governance remains the biggest obstacle to scaling AI
Internal data governance was the most frequently cited barrier to scaling AI, identified by 63% of respondents. Cost followed at 59%, while talent shortages and legacy technology integration were each cited by 55%. These challenges persist despite 87% of fintechs reporting infrastructure readiness and 85% reporting data readiness. The findings indicate that technical capability alone is insufficient to support wider AI deployment.
Engagement with Singapore's governance initiatives is substantial. Some 81% of surveyed firms were engaging with IMDA's Model AI Governance Framework and 86% with MAS's Project MindForge. However, only 9% were building controls and processes under each framework, with most remaining at the assessment stage.
This distinction exposes a gap between understanding governance requirements and translating them into operational safeguards. For agentic AI, 74% of respondents identified human checkpoints as a necessary control, followed by logging at 53%. Model monitoring and auditing, and identity and authentication, were each selected by 52%, while 49% cited approval or transaction limits.
These controls become increasingly important when AI systems can access sensitive information or execute transactions. Firms need to define what agents are permitted to do, how their actions are monitored and when employees must intervene.
Cost management introduces another consideration. Coinbase, for example, has developed an automated system that directs routine tasks to lower-cost AI models while reserving more advanced models for complex reasoning, helping manage expenditure as usage increases.
The report recommends matching AI autonomy to the risk and reliability of individual tasks. High-risk activities require stronger human oversight, while lower-risk, repeatable processes offer greater scope for autonomous execution with lighter controls.
For Singapore fintechs, the findings suggest that the next stage of AI adoption will depend on operational execution rather than deployment figures alone. Firms must demonstrate measurable improvements, establish effective governance and determine where autonomous systems can assume responsibility without introducing disproportionate risks. The ability to balance those requirements will be a more meaningful indicator of AI maturity than the number of tools deployed.




