Public sentiment on AI in banking and high-trust institutions

Research Question

What does current (2024–2025) survey data reveal about customer sentiment toward Artificial Intelligence (AI) in banking and high-trust Financial Services (FS) institutions — in Australia, across Asia-Pacific (APAC), and globally — and would a two-plane architecture (a "production plane" that augments human employees, and an "operational plane" that minimises direct AI involvement in binding customer-financial-data decisions) represent a viable, trust-differentiated approach that can be clearly communicated to customers?

Findings

Executive Summary

• [fact] Australian banking customers show a severe trust-use gap: 50% use Artificial Intelligence (AI) regularly but only 36% trust it, and 96% report reservations about bank AI use (KPMG/University of Melbourne 2025; Publicis Sapient 2024).
• [inference] The contrast between broad reservations and the 96% satisfaction rate among the 21% of customers who have actually used banking AI tools suggests that unfamiliarity and control concerns matter more than observed product failure (EPAM Continuum 2024; Publicis Sapient 2024).
• [inference] A two-plane architecture, with AI augmenting employees in a production plane and tighter constraints on AI in binding financial decisions, fits both customer preference data and current regulatory direction (Bird & Bird; Dentons; ABS Handbook 2026).
• [inference] The architecture itself is approaching a compliance baseline, so the differentiator is clear communication, visible human review, and employee capability to deliver the promised experience (Accenture 2025; KPMG/University of Melbourne 2025).

Key Findings

1. [fact] 96% of Australian banking customers have reservations about AI use by their bank, with the top three concerns being preference for human interaction (58%), job displacement fear (54%), and data privacy worry (49%) per Publicis Sapient 2024.
2. [fact] Australia records the lowest AI trust-to-use ratio in the KPMG/University of Melbourne 2025 global study: only 36% trust AI despite 50% regular use, and only 30% believe AI benefits outweigh risks, which is 18 points below the global average of 48% (KPMG/University of Melbourne 2025).
3. [fact] The "experience gap" is the central paradox in banking AI trust: only 21% of banking customers globally have used AI tools, but 96% of those users report satisfaction (EPAM Continuum 2024).
4. [fact] Asia-Pacific (APAC) regional averages mask extreme country-level variation in AI trust: China (35% increased trust in companies using Generative AI (Gen AI)) and India (29%) are more optimistic, while Australia (19%) and Japan (10% trust per Qualtrics 2025) are much more sceptical (Dataconomy 2024; Qualtrics Experience Management (XM) Institute 2025).
5. [fact] Globally, only 26% of consumers trust organisations to use AI responsibly, and direct experience with AI tools raises trust by roughly 40 to 50 percentage points (Qualtrics Experience Management (XM) Institute 2025; Edelman 2025 summary).
6. [fact] Regulatory frameworks are converging on human-in-the-loop mandates: APRA CPS 230 (effective July 2025), the MAS AI Risk Toolkit, and the Association of Banks in Singapore (ABS) Gen AI Guardrails Handbook all require human oversight for binding financial decisions (Bird & Bird; Singapore Law Watch; ABS Handbook 2026).
7. [fact] Banks with top-20% customer advocacy scores grow revenue 1.7x faster than peers, and 83% of Australians say responsible AI practices and assured accuracy would increase their trust (Accenture 2025; KPMG/University of Melbourne 2025).
8. [inference] The two-plane architecture directly addresses the three dominant customer concerns and aligns with regulatory direction, but it is structurally close to what regulators are beginning to require rather than a unique moat in isolation (Publicis Sapient 2024; Bird & Bird; ABS Handbook 2026).
9. [inference] Competitive differentiation lies in communication clarity and experience design: the first institution to make the architecture observable to customers, through concrete language and visible AI-augmented service quality, is best placed to capture the advocacy premium before the model becomes commoditised (Accenture 2025; EPAM Continuum 2024; Edelman 2025 summary).
10. [inference] Younger customers (18–34) show 41 points more AI trust than older customers (55+) in Edelman 2025 United Kingdom (UK) data, and 37% of 18–34 year-olds are considering switching banks in EPAM 2024, which supports demographic segmentation in how the model is communicated (Edelman 2025 summary; EPAM Continuum 2024).

Assumptions

• Assumption: 2024–2025 survey sentiment remains directionally valid through 2026. Justification: AI adoption is accelerating; structural concerns (human connection, privacy, jobs) are unlikely to reverse in 12 months, though absolute trust percentages may shift.
• Assumption: "Binding financial decisions" can be operationally separated from "augmentation tasks" in a production banking architecture. Justification: APRA CPS 230 and MAS/ABS guidance require risk-tiering of AI use cases, confirming the boundary is definable, though edge cases (AI-ranked mortgage applications reviewed by humans) may blur it.
• Assumption: Customers will perceive meaningful value in an explicit two-plane articulation rather than treating it as marketing language. Justification: 83% of Australians say responsible practices would increase trust (KPMG 2025) and 40–50 point trust lifts from direct AI experience (Edelman 2025) suggest receptivity, though no survey has tested this specific framing.

Analysis

• [inference] Customers are not rejecting all AI uses; they are distinguishing between AI that improves service under human control and AI that appears to act autonomously over their money (Publicis Sapient 2024; F5/Twimbit 2025; Accenture 2025).
• [inference] The production plane is attractive when it appears as better service, continuity of context, and faster human interactions, because those outcomes align with Accenture's advocacy drivers and with EPAM's high satisfaction rate among actual AI users (Accenture 2025; EPAM Continuum 2024).
• [inference] The operational plane is credible because regulators and industry guardrails already push banks toward stronger human oversight, auditability, and risk controls for higher-stakes AI uses (Bird & Bird; Dentons; ABS Handbook 2026).
• [inference] The commercial upside depends on turning safeguards into a better customer experience before these controls are perceived as basic compliance rather than differentiation (Accenture 2025; KPMG/University of Melbourne 2025).

Risks, Gaps, and Uncertainties

• No survey has directly tested two-plane messaging with banking customers, so viability is inferred from convergent indirect evidence.
• The boundary between "augmentation" and "decisioning" is operationally complex; edge cases such as AI that pre-scores loan applications or drafts customer correspondence may not fit cleanly into either plane.
• Australia's low AI training rate (24% versus 39% global, KPMG 2025) means employees may struggle to credibly represent AI-augmented workflows without upskilling investment.
• The "experience gap" thesis assumes positive AI encounters generalise to trust in the institution's broader AI use; that remains unproven at scale.
• APAC country-level variation means a regional strategy based on averages will underperform in both high-trust and low-trust markets.
• The competitive window may be narrow because once regulators fully mandate human oversight on binding decisions, the operational plane may be perceived as undifferentiated compliance.

Open Questions

• What is the cost and timeline to implement a clean production/operational plane separation in a legacy core banking architecture?
• Would an explicit "AI transparency report" (analogous to sustainability reporting) increase customer trust, or draw unwanted scrutiny?
• How do customers react when they learn AI was involved in a decision that benefited them (faster approval) vs. one that disadvantaged them (fraud false positive)?
• What employee training programme is needed to make production plane augmentation credible to front-line bank staff?
• Could a third "innovation plane" (customer-facing AI experiments with explicit opt-in) serve younger demographics while preserving the two-plane trust boundary for risk-averse customers?

sources

• [x] Publicis Sapient Customer Banking Report 2024 — Press Release
• [x] KPMG/University of Melbourne — Trust, Attitudes and Use of AI: Global Study 2025 — Australia Snapshot (Portable Document Format (PDF))
• [x] KPMG/University of Melbourne — Trust in AI Global Report 2025 (PDF)
• [x] 2025 Edelman Trust Barometer — Financial Services Sector
• [x] 2025 Edelman Trust Barometer — Global AI Trust summary
• [x] Accenture Global Banking Consumer Study 2025 (PDF)
• [x] Accenture Global Banking Consumer Study 2025 — Overview
• [x] EPAM Continuum 2024 Consumer Banking Report
• [x] FICO 2024 Bank Customer Experience Survey — Asia Pacific
• [x] F5/Twimbit — The 2025 AI Paradox: Understanding Consumer Perceptions in APAC
• [x] Qualtrics Experience Management (XM) Institute — Consumer Sentiment Toward AI Evolves, 2025 (PDF)
• [x] McKinsey — How AI Will Transform Banking
• [x] APRA CPS 230 — Operational Risk Management
• [x] [ABS Handbook on Generative AI Guardrails in Banking (PDF)](ABS Handbook on Generative AI Guardrails in Banking (PDF) — .pdf)
• [x] MAS AI Risk Toolkit for Financial Institutions
• [x] Dataconomy — APAC Customers Reveal Trust in AI, But Concerns Persist
• [x] FutureCIO — Consumer Sentiment on AI Far More Positive in APAC