McKinsey APAC banking GenAI survey: 70% of APAC banks now have production GenAI

McKinsey's annual APAC financial services AI survey recorded that 70% of surveyed APAC banks and insurers now have at least one generative AI model in production — up from 31% in 2023 and 51% in 2024. The compound growth rate is unusual for technology adoption in heavily regulated sectors, and reflects both competitive pressure and the improving regulatory clarity from HKMA, MAS, APRA, and FSA that has reduced the governance ambiguity that previously stalled production deployment.

**What 'production' means in this context.** McKinsey's survey defines 'production' as a model handling at least 5% of a real workflow's volume without mandatory human review of every output — a deliberately inclusive definition. In practice, the 70% figure includes: banks using AI for call-centre transcript summarisation (common), banks using AI for credit scoring inputs (less common), and banks using AI for customer-facing advice generation (rare, and typically requires additional disclosure). The range within 'production' is as wide as the range between 0% and 70%.

**Where the remaining 30% is stalling.** The survey notes that banks without production deployments cluster around three blockers: model risk governance requirements (board-level approval frameworks that take 6–12 months to establish), data access restrictions (core banking system APIs that don't support real-time AI integration), and talent gaps (insufficient ML engineering capacity to own a model in production). These are structural, not technical, blockers — which means solving them requires organisational change, not a different AI vendor.

**Implication for mid-market financial institutions.** The survey sample skews toward tier-1 and tier-2 banks. Mid-market financial institutions — regional banks, credit unions, mid-sized insurers, independent wealth managers — face the same blockers but with smaller IT teams and less regulatory relations capacity. For these organisations, the risk is falling materially behind tier-1 peers in AI efficiency over the next 12–24 months.

**AIMenta's editorial read.** 70% production adoption in APAC banking is a significant maturity signal. The competitive pressure it creates for the remaining 30% is real. However, the right benchmark is not 'are we in the 70%?' but 'which specific workflows are we running AI on, what are the measurable outcomes, and where are we visibly behind the market?' The answer to that question determines the urgency and shape of the response.