NatWest’s artificial-intelligence programme is notable less for one spectacular application than for its spread across the bank. Customer service, wealth-management records, software development, and financial-crime work are all using AI, suggesting the technology is becoming part of the operating model rather than an isolated innovation project.
Chief information officer Scott Marcar described 2025 as the first year these systems operated at scale. The deployment now reaches customer journeys, tens of thousands of employees, and a substantial engineering organisation.
That breadth creates the possibility of shared infrastructure and learning. It also makes governance harder because a conversational assistant, a document summary, generated code, and a fraud alert have different failure modes and consequences.
Cora Moves From Answers Towards Financial Action
NatWest expanded the number of customer journeys supported by generative AI in its Cora assistant from four to 21, reporting faster resolutions and less human intervention. Early in 2026, 25,000 customers are expected to receive an agentic assistant built on OpenAI models.
The new service will answer natural-language questions about transactions and spending patterns in the bank’s app. A later voice-to-voice phase is intended to recognise tone and conversational nuance, including when customers report suspected fraud and manage a case.
These functions require a higher standard than a general information bot. Transaction explanations must be grounded in the correct account, permissions must be explicit, and uncertainty must trigger a human route. Tone detection can support distressed customers, but it should not become an invisible judgment about them.
Internal tools have already saved time. NatWest says call summaries and complaint-drafting systems in retail banking returned more than 70,000 staff hours. The meaningful test is whether complaint quality and customer outcomes improve alongside speed.
Documents and Code Reveal Different Productivity Gains
In private banking and wealth management, AI summarises notes, meetings, and correspondence used by relationship managers. NatWest reports that the systems release 30% more time for direct client contact. Here, the gain comes from reducing administrative review while advisers retain responsibility for understanding the client and providing advice.
The bank’s 12,000 engineers use AI coding tools, and AI now produces more than a third of its code through drafting, review, and testing. NatWest also hired nearly 1,000 graduate software engineers in the UK and India during 2025, indicating that greater code generation can coexist with investment in technical talent.
Trials of agentic engineering in financial crime delivered a reported tenfold productivity increase. Such a figure deserves careful interpretation: teams should define the task, baseline, review burden, and quality outcome before assuming it applies across engineering.
All roughly 60,000 employees have access to Microsoft Copilot Chat and NatWest’s own language model, with more than half completing training beyond the basic provision. Broad access creates value only when staff know which data and decisions belong in each system.
Infrastructure and Governance Connect the Portfolio
NatWest restructured its data estate to create unified customer views, moved workloads to Amazon Web Services, and simplified legacy systems. Those changes support both customer conversations and internal summarisation. AI at this scale depends on data access and computing capacity that individual pilots can avoid confronting.
The bank has also established an AI research office, an AI and Data Ethics Code of Conduct, and participation in the Financial Conduct Authority’s Live AI Testing programme. These structures need to translate into specific controls: evaluation before release, monitoring after it, traceable sources, access boundaries, and named owners.
Success should be measured per workflow rather than with one bank-wide AI number. Resolution quality, customer escalation, adviser time, code defects, fraud outcomes, and employee verification effort all reveal different parts of performance.
NatWest’s approach shows what mature adoption begins to look like: many bounded systems resting on common data, infrastructure, skills, and governance. The challenge is keeping that foundation coherent as capability spreads. Scale creates value when each use case remains accountable, not merely when AI appears everywhere.
