By Nibras Adambawa, Founder and Principal Advisor, Limiere
For more than a decade, one question has followed me through conversations with bank CEOs, boards and executive teams across different markets and different phases of structural change: when an institution commits itself to a new strategic direction, what does that change eventually do to the economics of the bank?
The question has taken different forms over the years. During the first major wave of digital transformation, it concerned operating models, distribution economics, cost structures and the changing value of physical infrastructure. As regulatory expectations shifted, it concerned the relationship between resilience, capital, risk and long-term institutional value. As new competitive models emerged, it concerned where future profit pools would migrate and what incumbent banks would have to change before those economics became visible in reported performance. The common thread was that structural change only becomes strategically meaningful when its consequences can be traced into the financial architecture of the institution.
Artificial intelligence has brought that question into sharper focus than anything I have encountered before.
The banking industry has already moved well beyond the stage where AI can be discussed primarily as an emerging technology. Significant institutions are deploying models across credit, fraud, collections, customer service, operations and internal productivity, while simultaneously making substantial commitments to data infrastructure, cloud capacity, specialist talent, platforms and organisational capability. What was once an experimental portfolio is becoming a meaningful category of institutional investment.
Yet the executive conversation has not matured at the same pace.
Banks can often describe what they are building, how many initiatives are in production, how rapidly employees are adopting new tools and where technical performance has improved. What remains much harder to state, with the same confidence applied elsewhere in banking, is what those investments are economically worth.
That is the conversation this publication exists to advance.
The conversation is moving from capability to economics
The first phase of AI adoption was understandably dominated by capability. Banks needed to understand what the technology could do, which applications were viable, how models behaved in regulated environments, what infrastructure was required and whether individual use cases could survive operational deployment. Experimentation was not a weakness during that period. It was a necessary form of institutional learning.
The next phase is different because the financial commitments are becoming more consequential.
Once AI moves into operating budgets, multi-year technology programmes and enterprise-scale deployment, the standard of inquiry must change with it. The relevant executive question is no longer simply whether a model performs better or whether a process can be automated. Management has to determine whether those improvements are large enough, durable enough and sufficiently attributable to justify the capital required to create and sustain them.
That distinction matters because banking already knows how to separate activity from economic consequence in almost every other important area. A credit programme is not judged by the volume of lending alone, but by the risk-adjusted return it produces. A cost programme is not considered successful because a workflow has been redesigned, but because the expected financial improvement is eventually visible in the cost base. A capital initiative is not judged by strategic enthusiasm, but by what it does to return, resilience and the opportunity cost of scarce capital.
Artificial intelligence should mature into the same discipline.
Banking AI is moving from a technology conversation into a capital allocation conversation.
The difficult part is not finding benefit, but proving value
The economic problem with AI is more demanding than conventional return-on-investment analysis because the financial effect is usually indirect. Artificial intelligence changes the quality or speed of a decision, alters the way work is performed, improves prediction, changes a customer interaction or modifies the use of operating capacity. The financial value only appears if those operating effects subsequently move revenue, cost, credit loss, pricing, risk-weighted assets, capital consumption or another economically meaningful variable.
That creates several layers of uncertainty between technical success and financial value.
A contact-centre application may reduce average handling time materially, but the resulting financial benefit depends on whether released capacity is converted into lower cost, additional throughput, higher service levels or some other economically useful outcome. A credit model may improve predictive accuracy, but the valuation depends on whether that improvement changes portfolio selection, pricing, default formation, provisions or capital in a way that can reasonably be separated from changes in the economic environment or credit policy. An employee productivity tool may generate substantial time savings while leaving the institution’s cost base largely unchanged because the capacity is absorbed elsewhere.
None of these outcomes makes the technology unsuccessful. They simply demonstrate that operating improvement and financial value are not identical.
The harder task is attribution. When a financial metric moves after AI has been deployed, management still has to establish how much of that movement belongs to the intervention itself. Changes in customer mix, market conditions, pricing decisions, management action and broader operating improvements can all affect the same financial outcome. A credible valuation therefore requires a counterfactual, explicit assumptions and an evidentiary standard that distinguishes what is expected, what has been observed and what can reasonably be attributed.
That is why the question of AI value cannot be answered by simply multiplying a productivity improvement across the scale of the bank. The number has to be earned through the causal pathway that connects technical performance to financial consequence.
This publication will examine AI through the economics of the bank
There is already more than enough commentary explaining that artificial intelligence is advancing rapidly, that models are becoming more capable and that banks need to respond. Those developments matter, but they are not where this publication intends to spend most of its attention.
The focus here is narrower and, I believe, more useful to senior banking leadership.
We are interested in artificial intelligence primarily through the financial architecture of the institution: how AI affects margin, revenue, operating cost, credit quality, risk-adjusted return, risk-weighted assets, capital efficiency, ROE and ultimately enterprise value. We will examine where value is emerging across banking functions, how that value should be attributed, how the economics of AI change once infrastructure and operating costs are recognised, and how institutions should compare AI opportunities against competing uses of capital.
The publication will also examine the parts of the value equation that are frequently overlooked. Productivity is only financially meaningful when released capacity can be converted into something the institution values. A reduction in losses has to be separated from cyclical improvement before it can be attributed to a model. Capital relief must survive regulatory and methodological scrutiny before it becomes part of an investment case. The cost of waiting also matters, because failing to invest can carry an economic consequence just as surely as investing badly.
These are not technology questions in the narrow sense. They sit at the intersection of strategy, finance, risk and capital allocation, which is exactly where the AI conversation increasingly belongs.
Every article ultimately returns to the same question: where does AI enter the economics of the institution?
The standard has to match the question
If the subject is economic value, the standard of analysis has to be higher than the conventions of ordinary technology commentary.
This publication will not treat vendor claims as established financial outcomes simply because a percentage has been attached to them. It will not use adoption, pilot counts or model performance as substitutes for economic evidence. It will not manufacture urgency through competitive anxiety, nor will it imply certainty where the underlying data support only a range or a directional conclusion.
Where a material number is used, the source should be clear. Where a value claim is modelled, the assumptions should be visible. Where causality is uncertain, that uncertainty should be acknowledged. Where the available evidence is insufficient to support a strong conclusion, the language should reflect the weakness of the evidence rather than conceal it behind precision.
The original purpose of this publication was to operate closer to the standard of a serious board paper, central bank working paper or rigorous institutional case study than a conventional corporate blog. That ambition remains intact. What has changed is the question to which that analytical standard is now being applied.
The discipline we expect from AI valuation should also govern the way we write about it: evidence before assertion, assumptions made visible, attribution examined and uncertainty treated as part of the analysis rather than as an inconvenience to be edited away.
The publication applies the same evidentiary discipline to its thinking that Limiere applies to AI valuation.
From future banking economics to AI valuation
The intellectual lineage behind Limiere began well before artificial intelligence became the dominant strategic subject it is today.
Through Bancly, my work since 2012 has increasingly centred on the future economics of banking: how structural forces alter profitability, capital, operating models, competitive advantage and long-term institutional value. The important question was never simply what the banking industry might look like in the future, but how those changes would eventually manifest themselves in the financial performance of individual institutions.
Artificial intelligence now represents one of the most consequential expressions of that broader question.
Limiere narrows the lens. Rather than examining every force reshaping banking, it focuses specifically on the financial value of AI and on the institutional disciplines required to determine what that value is worth.
That is why this publication will remain deliberately banking-specific. The economics of a bank cannot be understood by taking a general enterprise AI framework and adding financial-services terminology to it. Margin, provisioning, capital adequacy, RWA, liquidity, credit quality and risk-adjusted return fundamentally shape how value is created and how investment decisions should be assessed.
The analysis has to begin from those economics rather than arrive at them as an afterthought.
The conversation banking now needs
The industry no longer needs to be persuaded that artificial intelligence will matter. That argument has largely been won.
The more important task is to develop a mature financial conversation about what AI is producing, where that value appears, how confidently it can be measured and whether the resulting economics justify the amount of capital being committed.
Some AI investments will prove extraordinarily valuable. Others will improve operations without materially changing financial performance. Some will create value that institutions fail to recognise because they have not built the measurement architecture required to see it, while others will appear successful until their operating achievements are subjected to financial scrutiny.
A serious banking institution needs to be able to distinguish among those outcomes.
That requires better questions, better evidence and a common language connecting artificial intelligence to the financial measures through which banks are already governed.
This publication exists to contribute to that language.
Limiere takes its name from the French word lumière, meaning light. The intention behind the name was always clarity. Today, the area of banking AI that most requires that clarity is not the technology itself, which receives more attention than perhaps any technology before it, but the economic value beneath it.
That is the conversation we intend to have here.
