AI in banking has moved from hype to reality
AI in banking has moved from hype to reality
AI in banking is no longer experimental theatre.
Santander reports €35m of AI-driven business value in Q1 2026 with a path to > €200m this year and > €1bn longer term. Lloyds reports ~ £50m delivered in 2025 and targets > £100m in 2026.
The signal: measurable, P&L-linked outcomes are emerging, mostly from productivity and operations, not pure revenue (yet).
Both banks have scaled AI into real workflows.
→ Santander automated fraud claims (95% faster, ~90% automated), accelerated AML investigations, and rolled AI out to 185,000 employees.
→ Lloyds deployed 50+ GenAI use cases: a knowledge assistant cutting search time by 66%, HR AI resolving ~90% of queries first time, and engineering tools boosting legacy code conversion by ~50%.
This is not a lab. It’s the operating model.
For years, AI promised transformation but has struggled to show its value. For these banks the value shows up: high-volume, repeatable work. Not strategy decks, but the queues, calls, claims, and codes.
Both banks are “rewiring” operations around AI, not just layering tools on top.
Their confidence has shifted from does it work, to how do we scale it.
WHY IT MATTERS
The real shift is behavioural. Employees are no longer doing the work, they are supervising, prompting, and judging AI outputs.
Customers are nudged into self-service and algorithmic guidance.
Organisations are redesigning decisions: who acts, who approves, who overrides.
In behaviour and transformation, your work has become decision architecture, trust design, and human-AI interaction loops.
WHAT TO WATCH FOR
→ % of decisions touched by AI, override rates, time-to-decision compression, error vs trust trade-offs, and where humans still intervene.
→ Where value migrates to. From cost-saving (today) to revenue and behaviour-shaping (next). Payments, onboarding, and financial guidance are early signals.
LIMITATION
The numbers are real, but incomplete. “Value” is rarely broken down into its reasons. Is it cost avoided, time saved, or revenue created? Few customer outcomes are disclosed (trust, complaints, vulnerability) and most big numbers are still projections.
There is a risk that scaled activity is mistaken for scaled impact.
It is always good to look beyond the idea to the real delivery. Here are some of the outcomes claimed:
SANTANDER
Real benefit, and commercially focused.
→ In Brazil, card-fraud claims are around 95% faster, with up to 90% automation and error rate below 1%.
→ Target of around 240,000 calls, or 40% of annual volume, resolved through self-service; projected savings of 26,000 customer hours and 45,000 service-team hours.
→ Approximately 100,000 AML alerts per year; investigations that used to take hours can be completed in minutes.
→ More than 17,000 people using agentic AI in software in May 2026; 40% of all code in June was developed by AI.
LLOYDS
More in the process, less tangible and feels more 'back office' ...
→ Enhanced customer interactions, accelerated query resolution, and better frontline colleague support. (No numbers given)
→ Around 5,000 engineers using AI-supported coding tools; 50% improvement in converting code for established systems.
→ Resolves around 90% of HR queries correctly on first contact.
SOURCE
