In last year’s report, we outlined the precision strategies that could help banks break through. This year, we find that precision strategies are still the order of the day, with a twist: Banks must develop a new and increased velocity of execution to match the speed of AI development. Banks have always been able to follow new technology developments at their own pace, in part because their older customers were also slow to adopt the technology. But customers of all ages are adopting AI faster than they have adopted any technology in history. This time, there is no grace period.
In these excerpts from our new report, we look at some new industry-wide developments, including the emerging imperative for speed; the dividend on offer to banks from AI.
State of the industry
Banking had a stellar year, but change is on the way. Three forces—the rise of fintechs and neobanks, new technologies that dissolve customer “stickiness,” and changing customer behavior—threaten the strength of the relationship between customers and their primary bank. Banks are accustomed to change; what’s different this time is the pace.
In 2025, the global banking industry outdid itself—again. Net income rose to $1.3 trillion, up 7 percent from 2024’s record-setting tally. Once again, banking recorded the most net income of any industry. As rates began to fall in major markets, net interest margins (NIM) declined slightly, globally from 1.65 percent in 2024 to 1.62 percent in 2025. However, some markets improved their margins. Japanese banks’ NIM rose four basis points (bps), UK banks’ three bps, and US banks’ one bp. Emerging markets were not quite as fortunate, as margins slipped slightly, and in Brazil, dramatically (from 3.58 percent to 2.97 percent). Balances (deposits, loans, and assets under management) grew by 6.3 percent in 2025, up from 6.2 percent annually in 2020 to 2024. Investors benefited from higher dividends and share buybacks, and banks socked away another $900 billion in surplus free cashflow to equity which, while somewhat lower than previous years, continued the record-setting trend begun in 2022.
New regional business models
Divergence in performance has deeper roots. From a business model perspective, banking is not global; it’s regional and local, with strong variations. As we analyzed banks worldwide by the degree to which they have shucked off the balance sheet and moved into fee-based services (disintermediation); and by their efficiency (Exhibit 1), we’re seeing an alignment by business model that reflects traditional regional affinities, as well as cultural and regional similarities among far-flung markets.
We see a new pattern emerging—“continental” or transnational banking with digital platforms and new playbooks. Banks are building scale across regions with similar business models, pursuing even platform-driven advantages. DBS, for example, expanded from its Singapore presence with the aim of establishing a pan-Asian network, while Hungary’s OTP Bank is growing fast with its Eastern European and Central Asia expansion model. And BBVA is evolving into a multicountry group embracing Spain, Mexico, South America, and Turkey. Wholesale is the one business line that remains consistently global. But it accounts for only about 11 percent of industry revenues.
Neobanks are a similar story of regional expansion—but their model is very different. See sidebar “Neobanks are breaking out.”
The technology grace period is ending
Banks have fended off many threats in the past, most often by waiting them out. While the rise of the internet and the smartphone upended other consumer industries, banks chose to adopt the technologies more slowly than others. They enjoyed a singular advantage: Banks were able to pace themselves, in part because their more loyal customers were slower to adopt new technology. That advantage won’t work this time: AI adoption is the fastest in history, and young and old people are piling in at nearly the same rate.
The traditional banking strategy for new technologies, call it “smart followership,” worked reasonably well for one reason. Banks’ high-value customer segments (aged 55 and up) were also slow to adopt digital, preserving core revenue pools and allowing banks to maintain their market share (Exhibit 2). The result was a grace period for banks that allowed them and their customers to ease into new technologies together.
Beneath record results, structural challenges are percolating. Customer primacy is increasingly at risk, and the pace of technological change—particularly driven by AI—is accelerating beyond what most banks are equipped to handle. Historically, banks could afford this trade-off. However, now banks must achieve what was previously considered a contradiction in terms: maintaining risk discipline and resilience while operating at significantly higher speed.
Banks need faster cycles, differentiated risk management, and scalable experimentation. One operating model at one speed cannot supply both stability and speed. Core innovations still require care and control; while another part of the portfolio has to run at the business equivalent of the speed of light. To move away from a single operating model, banks should organize technology development and innovation at three speeds (Exhibit 3).
For more on the industry’s performance in 2025 and the forces affecting customer stickiness, see our May 2026 report.
The AI disruption dividend
The economics of banking may shift in quarters rather than years as AI changes what the work costs, how a bank and its customers interact, and how those customers choose. Advantage for some banks is building in several assets: data, context, judgment, trust, and the controls around the technologies, and this could compound with use. The critical variable is the speed at which banks capitalize on these assets to capture the AI dividend. Speed could determine how much value the bank keeps, and unlike in past technology cycles, a late start may not offer chances to catch up, depending on how the profits get reinvested. The race for the dividend is unfolding in four areas.
Industry economics and the customer interface
Start with the most salient development: Money is beginning to move based on advice from software agents. This is not just a question of whether agents can reinvest low-cost deposits; rather, it’s potentially a redesign of the entire ecosystem of payments: agents that help with buying things on websites, agents that advise clients on where to invest their money, and agents that negotiate contracts and issue invoices. The payment networks appear supportive. Visa’s Intelligent Commerce and Mastercard’s Agent Pay let merchants accept purchases from verified agents, PayPal agreed to put its wallet into ChatGPT’s checkout, and Stripe and OpenAI co-developed the protocol behind it. Meta’s Muse agent, launched in September 2026, passed 5 million downloads in the United States in 22 days, faster than ChatGPT, Grok, or Claude reached that mark. It already reads users’ balances and transactions through Plaid.
But the dividend from AI disruption can outweigh a decline in deposit profits. A bank that has anticipated the disruption might have already offered its clients more attractive and suitable investment products, financial advice, and cash management capabilities that rebalance the profitability. While deposit profits fall, the bank may see overall profitability rise.
Large language models (LLMs) have emerged as an entirely new distribution channel in record time. While it took mobile nearly a decade to reach half of bank users, our research shows that more than 50 percent of users with $100,000 in wealth are already using LLMs for financial questions. Historically, US consumers compared an average of 2.5 financial institutions when selecting a product; now LLMs can effortlessly and dispassionately compare a far wider set.
Customers are moving faster than banks
We surveyed 46,000 retail banking customers in major banking markets worldwide. The survey and our work with banks worldwide confirm that consumers are moving rapidly. Awareness of gen AI tools is almost universal, and nearly a quarter of customers are using them for their finances (Exhibit 4). A customer who once asked a branch adviser about a savings rate now asks ChatGPT to compare rates across the market. Seventy-four percent of consumers who use AI for banking tasks go through ChatGPT, Claude, or Gemini, about twice the share who use their bank’s chatbot today (to be sure, not many banks have rolled out fully enabled bots). The transaction still comes to the bank; the conversation that shapes it is increasingly happening elsewhere.
Trust remains the cornerstone, with banks leading the way. Sixty-two percent of AI users would rather get AI services and advice from their primary bank than from anyone else, and privacy is their greatest concern with AI. The pattern we’re seeing: A customer may experiment with a chatbot on a $1,000 balance but a family deciding where to hold $300,000 of savings will meet its human advisor. Trust also unlocks the data that makes answers personal and relevant; third-party tools can access only some of that data today. The advantage is, however, conditional. Roughly half of AI users say they may turn to a third-party financial app if their bank does not offer capable AI, so trust buys a bank the first chance, and half of its AI-using customers say they will not wait indefinitely.
From employee productivity to institutional intelligence
Most banks have used AI so far to help employees work faster. The next step, institutional intelligence, makes the bank itself smarter. When a bank’s best credit officer retires today, her judgment leaves with her. Nor is that the only problem: The bank’s best thinking is usually scattered across the institution. And decision-makers struggle to thread their best ideas through myriad policies and compliance requirements. Institutional intelligence connects the data, policies, past decisions, and expert judgment that currently sits here and there, gathering it such that every agent and employee works from what the whole bank knows, a knowledge that expands on itself. With that, average employees can attain top-quartile performance, in our experience.
Individual productivity benefits are already apparent. In McKinsey’s 2026 State of AI survey, 80 percent of respondents report individual productivity gains from AI. But only 6 percent attribute a significant share of EBIT to it. Individual gains diminish before they reach the P&L. Why? One reason is that the technology today has been designed at the level of the task—but tasks are not easily reallocated. In a benchmark by QuantumBlack, AI by McKinsey, of seven agentic systems running in production at financial and professional services firms, agents cut the time that tasks take by roughly 60 percent. But the bank captures far less because an individual’s savings are fractional, an hour or two of someone’s day. No bank can monetize a fraction of a job.
While tasks have been sped up, processes have not. A credit memo that now takes minutes still waits for a committee that meets on Thursdays. Leading banks have realized they need to redesign the work or restructure roles and teams so the fragments add up. Only one-quarter of companies using gen AI has done that redesign anywhere. The constraint is organizational: Banks need to redesign the end-to-end domains that matter and measure the result in cost, throughput, and risk rather than in hours saved.
An agent is only as good as the knowledge it can reach. Leading banks are discovering that the process of building institutional intelligence takes time, but results get better and better. They’re building three layers of institutional intelligence:
- Data—reliable data products, such as a customer 360 and a credit 360, that agents can consume at scale. Forty-eight percent of the banking executives we surveyed rank data readiness as their biggest AI obstacle, while technology is the smallest problem.
- Context—a map of how a customer’s accounts, obligations, and regulatory status connect, with permissions enforced before anything reaches a model, so that every agent works from one version of the customer rather than its own.
- Judgment, encoded—what the senior credit officer knows that never made it into the policy manual, captured in a format that is reusable by agents.
Since every bank can buy the same models, these layers are the moat, and they are accumulated domain by domain from the institution’s memory and practices.
Controls and coordination at machine speed
Turning to safety, today’s control model was built for people executing steps at human speed: checklists, four-eyes approvals, and audits that sample a few cases at a time. Banks are now running agents that work at machine speed. Control therefore moves from the process to the policy: The bank approves the rules, limits, and escalation paths an agent operates within, monitors the full population of its actions rather than a sample, and keeps a named human accountable for exceptions. The controls themselves become agents, with first-line controls in code and second-line agents monitoring outputs continuously. Confidence in an agent is engineered before release, checked in production, and improved from what goes wrong.
Banking’s new critical path runs from an idea to an agent in production. Most banks still run it the way they ran digital, with two-week sprints, ten-person squads, and approvals in series. That’s a problem, not least because internal resistance to the new technologies is even greater than it was for digital. The banks moving fastest have redesigned the operating model to build and run agentic AI: agile stripped of its ceremonies (structured meetings) for the teams that build; a new chart for how humans and agents are organized; new paths for talent and judgment; and a factory that can build or enable all levels of complexity.
Banks are rapidly adopting AI. Profit pools are shifting slowly, but that will change. What matters are the actions banks take today to set themselves up for an agentic future: finding the best positioning in the industry ecosystem; catching up to customer adoption, and building institutional intelligence and an infrastructure of controls, all at an unprecedented pace.
Banks’ performance in 2025 was one for the books. As 2026 wraps up and the new year awaits, leaders are confronting an environment in constant motion, thanks to AI. We hope that this report unravels much of the complexity that threatens banks’ ways of working and helps you take on the challenges with the necessary precision and speed. Let us know what you think: GBAR@mckinsey.com.