Cameron Thomson, group vice president for the EMEA theatre at Avaya, says banks should treat AI as an extension of existing trust and resilience requirements. The key questions are where AI can act safely, where a human should remain involved, and how control can be preserved. In a demonstrated scenario, AI identifies a suspicious transaction, blocks the card and begins resolving the issue digitally. When a replacement-card request changes the risk profile, a human specialist receives the full context.
In highly regulated sectors, an incorrect decision involving a financial transaction, customer identity or fraud alert can create severe regulatory, financial and reputational consequences. Financial institutions need visibility into data location, access rights, decision governance and service availability. This is driving continued demand for hybrid and on-premises deployment alongside public cloud, particularly in Europe, where sovereignty is a major consideration. Flexibility can make responsible innovation possible.
Thomson highlights the Model Context Protocol (MCP), which provides a standardised and governed way for AI to access CRM platforms, operational applications, knowledge sources and APIs. A language model can generate an answer, but it does not inherently know what is happening in a customer’s account or have permission to act. Secure connections to fraud information, customer records and approved workflows can reduce the need for custom integrations. An AI-agnostic orchestration layer also avoids a permanent commitment to one model provider.
The approach brings customer, operational and risk information together while an interaction is taking place. A card used abroad can trigger an appropriate response, while a later request that changes the risk profile can route the conversation to a specialist with the relevant context. The objective is to move from fraud detection to resolution without fragmenting the customer experience. Banks can make service and risk decisions from the same current picture rather than separate silos.
Regulated organisations are not rejecting modernisation, but they are resisting the idea of replacing everything at once and at scale. Gradual modernisation can analyse existing workflows with AI, translate them into understandable business logic and move them into a modern environment. As institutions add AI agents , applications and data sources, an open architecture and resilient core can support innovation without redesigning the entire customer journey each time.