The most important development in banking AI this year may not be a more powerful model or another generation of autonomous agents. It may simply be that banks are beginning to attach real numbers to what the technology is delivering. Santander says AI generated €35 million in business value in the first quarter of 2026 and is targeting more than €200 million for the year. Lloyds Banking Group says generative AI produced around £50 million of value in 2025 and expects more than £100 million of additional value in 2026. Elsewhere, banks are measuring hours returned to employees, reductions in processing times, faster fraud investigations and improvements in customer operations. After years in which the banking industry largely discussed AI through pilots, use cases and potential, the conversation is beginning to acquire something it badly needed: an economic measure of whether any of it is actually working.
From AI Potential to AI Economics
The business case for artificial intelligence in banking has always sounded compelling. Banks employ large workforces, operate complex technology estates, process enormous volumes of documents and transactions and maintain operational functions filled with repetitive knowledge work. Even relatively small productivity improvements multiplied across thousands of employees can theoretically create substantial value.
The difficulty has been proving it.
Banks have announced hundreds of AI initiatives over the past several years, but the metrics accompanying those announcements have often focused on deployment rather than outcomes. The number of employees given access to an AI assistant, the number of models placed into production or the number of use cases identified says relatively little about whether the investment ultimately improves the economics of the institution.
That distinction becomes increasingly important as AI budgets grow. Finnoex recently examined why Banks Have Plenty of AI Ideas. Getting Them Into Production Is the Hard Part. The next challenge is becoming equally clear: getting something into production is no longer enough. Banks increasingly need to demonstrate what happens after it gets there.
Santander offers one of the clearest indications of how that measurement is evolving. The bank has established a target of generating more than €1 billion in business value from AI between 2026 and 2028 through additional revenue and cost reductions. It reported €35 million of value during the first quarter of 2026 and expects to exceed €200 million by the end of the year.
That turns AI from an innovation narrative into something much closer to an investment case.
Productivity Starts to Become a Financial Number
Some of the earliest measurable returns are emerging from a deceptively simple source: giving employees back time.
BBVA estimates that employees using AI to automate repetitive tasks save around three hours per week on average. At an individual level, three hours might not appear transformational. Across an organisation employing more than 100,000 people, however, the potential scale changes dramatically.
Not every saved hour becomes a direct cost reduction, nor should productivity statistics automatically be interpreted as cash savings. The value may instead appear through additional customer interactions, faster decision-making, increased capacity, lower reliance on additional hiring or employees spending more time on work requiring human judgement.
This is an important distinction in calculating AI returns. A bank does not necessarily need to remove a job for AI to create economic value. If a relationship manager can serve more customers, an engineer can complete a migration faster or a contact-centre employee can resolve enquiries more efficiently, productivity can flow through the organisation in multiple ways.
Lloyds Banking Group provides another useful example. Its Athena AI-powered knowledge tool has reduced the time employees spend searching for information by an average of 66%. Around 5,000 engineers using AI-supported coding tools have also achieved a 50% improvement in converting code for established systems. These are operational measurements, but at sufficient scale they begin contributing to financial outcomes.
Lloyds says around 50 major live generative AI use cases generated approximately £50 million of value in 2025. For 2026, it is targeting more than £100 million in incremental P&L benefit.
The language matters. AI is beginning to appear not simply as productivity technology, but as something banks expect to influence the profit and loss account.
Fraud, Compliance and Operations Could Produce the Clearest Returns
Customer-facing AI receives much of the attention, but some of the most compelling economics may emerge from areas customers rarely see.
Santander says it now has more than 280 process automation agents in production. At Openbank, its digital banking subsidiary, AI models process around 100,000 anti-money laundering alerts annually. The bank says investigations that previously required hours can in some cases be completed within minutes.
The same principle applies across fraud investigations, transaction monitoring, document verification, complaints, onboarding and other operational processes. These functions combine high transaction volumes with significant manual effort, making relatively modest improvements valuable when repeated hundreds of thousands or millions of times.
This also changes how banks may prioritise AI investments.
The most visually impressive application is not necessarily the one with the strongest return. An intelligent customer assistant might attract more attention than a system that reduces the time required to investigate an AML alert, but the latter can potentially deliver a much clearer operational equation: number of cases multiplied by minutes saved, employee cost, error reduction and improved processing capacity.
As AI investment becomes more disciplined, banks may increasingly favour use cases where this equation can be demonstrated.
The Numbers Also Need to Become Harder to Challenge
There is an important caution behind the emerging AI ROI story. “Business value” is not a universally standardised accounting measure.
One bank may calculate employee hours saved and assign a monetary value to them. Another may count avoided costs, incremental revenue, reduced losses or additional operational capacity. Some benefits may eventually appear directly in the P&L, while others represent theoretical capacity that only becomes financially valuable if the organisation actually uses it.
That means headline AI value figures should not automatically be compared bank against bank.
The more important development is that banks are attempting to measure them at all.
Over time, AI measurement will likely need to become considerably more sophisticated. Institutions will need to distinguish between gross productivity benefits and realised financial benefits, account for the cost of models, cloud infrastructure, integration, governance and specialist employees, and determine whether improvements remain after systems are scaled.
A use case that saves $10 million while requiring $8 million of infrastructure, implementation and oversight has very different economics from one generating the same benefit at a fraction of the cost.
The next generation of banking AI dashboards may therefore look much less like technology dashboards and much more like investment portfolios.
Scale Is Where Small Improvements Become Material
AI economics in banking has another characteristic working in its favour: scale.
Consider a process that takes an employee ten minutes and occurs one million times each year. Reducing it to eight minutes saves more than 33,000 hours annually. Cut the process to five minutes and the saving exceeds 83,000 hours.
The individual interaction barely looks transformational. The aggregate impact does.
Banks contain thousands of processes with exactly this characteristic. Customer enquiries, payment investigations, document reviews, compliance checks, credit preparation, software development, internal searches, account servicing and reporting are repeated continuously across large institutions.
This is why the AI business case may ultimately be built through thousands of incremental improvements rather than one spectacular application.
It also explains why scale matters so much. A successful AI pilot involving 100 employees proves that technology can work. Deploying it across 20,000 employees begins to change the economics of the organisation.
AI Agents Could Push the Equation Further
The emerging shift from copilots to agents could make these calculations considerably more interesting.
Most first-generation generative AI applications improve individual tasks. They summarise a document, retrieve information, draft a response or help an employee analyse something more quickly. The human remains responsible for connecting those tasks into a complete workflow.
Agents potentially remove some of those hand-offs.
An AI agent might retrieve information, evaluate it against policy, interact with another system, prepare documentation and escalate only the cases requiring human judgement. Instead of saving two minutes from one task, AI begins compressing an entire process.
That is why the movement explored in The Next Core Banking Battle Will Be Fought Over AI has implications beyond technology architecture. Connecting governed AI agents to operational and core banking systems potentially gives banks access to a different level of productivity because intelligence can participate directly in workflows rather than merely advising the people performing them.
But it also raises the standard for proving value. Autonomous systems require governance, monitoring, security and human oversight, all of which carry costs. Banks will need to measure the economics of the complete operating model rather than simply the performance of the AI itself.
The Winning Metric May Be Value per AI Deployment
For several years, banks have competed publicly over the scale of their AI ambitions. Thousands of employees trained. Hundreds of use cases identified. Dozens of models deployed. Large technology partnerships announced.
Those numbers are likely to become less impressive on their own.
If one institution operates 500 AI applications producing limited measurable benefit while another operates 100 that materially reduce costs, increase revenue or improve risk outcomes, the second institution arguably has the more mature AI strategy.
This could gradually shift executive attention from AI adoption to AI productivity.
The questions then become more demanding. How much value does each deployed use case generate? How quickly does it recover its implementation cost? What percentage of pilots reach production? How much measured productivity is converted into actual financial benefit? Which AI systems should be expanded, and which should simply be shut down?
Those are not questions about artificial intelligence. They are questions about capital allocation.
And that may ultimately be the clearest sign that AI is becoming embedded in banking. The technology stops being treated as exceptional and begins being subjected to the same economic discipline as every other major investment.
What it means for the industry
- AI ROI is becoming measurable. Banks are beginning to report monetary value alongside traditional measures such as use cases, adoption and productivity, creating greater pressure to demonstrate tangible returns.
- Employee productivity could become one of the largest early sources of value. Small amounts of time saved across very large banking workforces can translate into substantial additional operational capacity.
- Back-office AI may deliver some of the strongest economics. Fraud, compliance, software engineering and operations offer high-volume processes where improvements can be measured and scaled.
- AI value figures will require greater scrutiny. Banks will increasingly need to distinguish between theoretical productivity, avoided costs, incremental revenue and benefits that genuinely reach the P&L.
- Agentic AI could expand the addressable return. Moving from individual task assistance to end-to-end workflow automation could create larger gains, but governance and infrastructure costs must be included in the calculation.
- The competitive benchmark is changing. The number of AI pilots or models in production will matter less than how efficiently banks convert those deployments into measurable business value.
