Skip to the main content.

The Bank Leader’s Guide to AI Risk, Efficiency, and Survival.

AI is changing banking faster than the traditional strategic-planning cycle was designed to accommodate. What was recently viewed as an emerging technology is becoming the gold standard for how financial institutions compete, operate, manage risk, allocate resources, and serve customers.

As consumers evolve into everyday users of AI, employees are mirroring that behavior in the workplace. At the same time, banks are leveraging AI for efficiency gains to reduce manaul work, accelerate decision-making, and improve the customer experience. AI is fundamentally changing the economics of the financial services industry and bank executives are seeing its strategic relevance firsthand.

For bank leaders, the question to ask is no longer: How should we use AI? but: How does AI change the institution we must become?

In practical terms, AI can address many of the concerns bank boards and executives face daily. From improving the institution's competitive position to the effectiveness of its operating model, bank leaders leveraging AI glean insights into investments, risk exposure, the workforce, and long-term ability to remain relevant. This means confronting "unknowns" that go well beyond selecting the right AI technology. Other key questions to ask in the process:

  • Will digital-only competitors out-price us when their efficiency ratio is one-third of ours?
  • Will traditional competitors deliver a better banking experience by removing substantial backroom friction that still slows our processes?
  • Is our core banking platform holding us back from using AI?
  • Are we prepared for an AI-assisted cyberattack on a scale that our incident-response plan was never designed to handle?
  • Is our team more productive because of AI, or are we paying for tokens on top of headcount?
  • Are missed opportunities for operational gains caused by the limitations of the AI-tools? Or, are they caused by limitations imposed by the institution’s AI policy?
  • Are we locked into vendor contracts that will be technologically obsolete before they expire?
  • Is a certain asset size required to compete in an AI-driven era? Or can community institutions participate in AI-driven innovation without achieving that scale?

For bank executives, AI is no longer a technology roadmap item. It is a lens for every strategic decision: where to invest, where to remove friction, which risks to accept, and which capabilities to build.

What NuBank’s 19.5% Efficiency Ratio Should Tell Your Board

 

What if a competitor could deliver the same services at a fraction of your cost? Nubank already does, with a 19.5% efficiency ratio. Of all the questions AI raises for bank leaders, this one is the most immediate: What happens when competitors operate at a fundamentally different cost structure?

 

Banks across the country are beginning to test that question in real time. AI is rapidly moving from a technology initiative to an operating-model decision with the potential to fundamentally change banking economics.

 

For traditional banks, the opportunity is significant. AI may represent the biggest lever of operational efficiency the industry has seen in a decade, creating new ways to reduce manual work, accelerate decisions, improve service, and increase employee productivity. But the same technology exploiting that opportunity is also introducing a new category of risk, often faster than governance, budgets, and operating models can keep pace.

 

NuBank’s 19.5% efficiency ratio puts the competitive implications into perspective. The point is not that every traditional bank should, or even could, operate like a digital native institution. It is that AI could widen the economic gap between institutions that fundamentally rethink how work gets done versus those that simply add AI tools to their existing operating model.

AI Risk: Cybersecurity and Nation-State Threats

The same capabilities that make AI such a powerful efficiency tool also make it a powerful weapon. As banks use AI to automate work, speed decisions, and operate more efficiently, cybercriminals and nation-state actors use the technology to increase the speed, scale, and sophistication of their attacks.

These two incidents demonstrate how AI can lower the barrier to launching sophisticated attacks.

AI Cyber Risk: GTG-1002 and Project Glasswing

In November 2025, Anthropic disclosed a cyber-espionage campaign it attributed [with confidence] to a Chinese state-sponsored group, designated GTG-1002. The campaign attempted intrusions against approximately 30 organizations, including financial institutions, technology companies, chemical manufacturers, and government agencies. Anthropic estimated that Claude Code performed 80–90% of the tactical work, such as reconnaissance, vulnerability discovery, exploitation, credential harvesting, lateral movement, and analysis of stolen information. Humans still selected targets and stepped in at critical decision points, and Anthropic validated only a handful of successful intrusions. The key takeaway was the sophistication of a cyberattack AI could carry out with limited human involvement.

For bank defenders, this changes the threat in kind, not just degree. AI can allow attackers to automate work that previously required people to perform reconnaissance, test vulnerabilities and analyze information. PwC describes a growing imbalance between attackers that can scale activity through AI and defenders that often still depend on people, processes and separate security tools. For banks, this risk also extends to the third parties connected to their technology environment, including core providers, cloud platforms, payment processors, and fintech partners.

The stakes escalated in April 2026, when Anthropic introduced Claude Mythos Preview. The company reported that the model could autonomously discover previously unknown vulnerabilities in real software and, in many cases, produce working proof-of-concept exploits during authorized testing. Because those same capabilities could be used offensively, Anthropic chose not to release the model broadly. Instead, it launched Project Glasswing, a controlled-access initiative for defensive cybersecurity work. The program began with 12 named partners and more than 40 additional organizations responsible for critical software infrastructure, supported by up to $100 million in model-usage credits. **Note: Anthropic has since reported that Glasswing partners found more than 10,000 'high or critical' severity vulnerabilities in the program's first month. 

The issue also reached banking leadership. Reuters and other outlets reported that Treasury Secretary Scott Bessent, Federal Reserve Chair Jerome Powell and CEOs of major U.S. banks met in Washington to discuss the cyber risks associated with Mythos and similar models. For community and regional banks, the issue is not whether they can match the cybersecurity resources of the country's largest institutions. It is whether their incident-response plans, vendor oversight and security controls are keeping pace with attacks that AI could make faster and easier to scale.

The same agentic capabilities will not remain confined to a handful of sophisticated actors. Nation-state adversaries, state-sponsored groups and organized criminal networks will adopt comparable tools, while AI-enabled tactics such as voice cloning, deepfakes and highly personalized social engineering are already viable today.

If AI capabilities can worry the CEOs of banks spending upward of $19.8 billion a year on technology, a bank with a fraction of that budget and none of that in-house security bench needs a different answer, one built on vendor diligence and layered defense, not on trying to out-invest the largest banks in the country.

AI Reward: Team Efficiency and the New Credit Workflow

The same technology that gives bad actors more leverage can give bank employees more leverage too. That is the other side of the AI equation. While banks need to prepare for the risks AI can introduce, they also need to capture the productivity gains it makes possible. There is more opportunity than simply automating tasks or reducing headcount, AI can redesign workflows so employees spend less time gathering information, moving data between systems and completing repetitive work and more time applying judgement, serving customers, and making decisions. Few areas illustrate that opportunity more clearly than the credit workflow.

 

Consider a commercial bank evaluating a CRE deal. Instead of the analyst manually working through building financials, rent rolls, operating statements, and personal tax returns for a CRE deal, AI can analyze those materials and produce a first draft in minutes instead of hours. That frees up analysts and relationship managers to spend more of their time evaluating the credit, exercising judgement, and having proactive conversations with clients.

 

This example is not an isolated use-case. AI assisted lending workflows are increasingly being used to reduce the manual work involved in underwriting and credit preparation across the industry. Credit memo drafting historically consumed 30-60% of an underwriter’s time. AI assisted workflows are cutting that from hours/days to minutes and delivering 20-60% productivity gains for credit teams.

 

This shift is already happening at scale. In another case, DBS rolled out an agentic AI credit-assessment tool to roughly 1,500 relationship managers and credit risk managers in August 2026, targeting at least a 30% reduction in the 40% of time Relationship Managers spent doing credit memo prep, while keeping the final credit judgement with a human.

 

Every bank who is using AI for credit workflow is explicit that AI handles more of the information gathering, analysis, and first-draft work so experienced bankers can concentrate on decisions that require human judgement.

 

This changes the productivity question bank leaders should be asking. Instead of “How many fewer analysts do we need?” ask “How many more relationships can the same team effectively manage?”

 

But realizing that productivity gain requires more than giving employees access to an AI tool. AI needs timely access to data, the ability to interact with workflows across systems, and an architecture capable of supporting automation without sacrificing control. If critical information remains trapped in disconnected systems or the core cannot easily participate in an AI-enabled operating model, the technology’s potential will hit a ceiling.

 

bank pillar page-1

Core Banking Modernization: The Legacy Bottleneck

The productivity gains AI promises depend on something far less exciting but equally important: whether the technology can actually access the data it needs. For many banks, that is where the limitations of legacy core infrastructure become impossible to ignore. In fact 55% of banks cite legacy core platforms as the single biggest barrier to digital transformation, while only 32% have successfully integrated AI into their core systems at all. 

Most core systems were architected decades ago around batch processing, not real-time data exchange. Ask an AI model “what does this customer’s relationship look like right now?” and the answer is only as good as the data it can reach. If checking, lending, payments, and other relationship data remain fragmented across systems, or are only updated through batch processes, the institution may have the information without having the ability to use it when it matters.   

That is why APIs, data accessibility and interoperability are becoming increasingly important considerations in core banking modernization. APIs provide a standardized way for institutions to access and exchange data between the core and the broader technology ecosystem. As banks evaluate their core strategies, the question is no longer simply whether a platform can process transactions reliably. It is whether the architecture can safely make the bank’s proprietary data available to the systems, workflows and AI capabilities that will increasingly depend on it. 

Core system providers, like FIS, Fiserv and Jack Henry are all moving toward more open, cloud-oriented and API-enabled banking technology. However, the issue is more complicated than saying older core systems simply do not have APIs. Fiserv has multiple modernization paths, including Finxact, a cloud-native, API-first core acquired in 2022, and CoreAdvance, which adds newer capabilities around its Premier foundation. FIS describes its Modern Banking Platform as cloud-native, modular and API-first, allowing banks to add capabilities over time rather than necessarily replacing everything at once. Jack Henry's jXchange is an API and integration environment rather than a replacement core. The broader Jack Henry Platform is the company's cloud-native, componentized modernization strategy designed to work alongside existing core environments and support a gradual transition. 

The challenge is that a bank's core cannot simply be swapped out like a single piece of software. Years of customer data, digital banking, payment systems, lending platforms, risk tools, workflows, custom integrations and third-party relationships are connected to it. McKinsey and Deloitte both describe core modernization as difficult, costly and operationally risky because banks have to untangle these dependencies while continuing to operate normally. Data has to be mapped and validated, integrations may need to be rebuilt, and employees and processes have to adjust to the new environment. This helps explain why a bank may recognize the benefits of newer technology but still hesitate to undertake a full core conversion. 

For AI, then, the decision is not necessarily old core versus new core. Banks need to understand what their existing environment makes possible: what data the core exposes, how quickly that data can be accessed, which APIs and integrations are available, what those connections cost and what contractual restrictions govern how the data can be used. 

That last point is becoming increasingly important. The American Bankers Association's principles for bank-core provider relationships state that banks should have the ability to easily access and manage their data through APIs. A bank may own the data it wants to put to work with AI, but ownership does not automatically translate into accessibility. The cost, speed and contractual rights associated with accessing that data could ultimately determine how quickly a bank can turn AI from an experiment into an operational capability. 

The economics make that distinction important. According to VisBanking, a $1 billion community bank may spend $500,000 to $1.2 million annually on its core, while switching platforms can require $2 million to $5 million upfront and 18 to 24 months of operational disruption. For many institutions, replacing the core solely to become more AI-ready is neither practical nor necessary. 

A more realistic near-term strategy may be to create an integration or middleware layer that exposes the data AI applications need through APIs without requiring a full core replacement. That can give banks greater flexibility while preserving the systems and processes that still work. 

But banks cannot make that decision without first understanding what their current environment actually allows. AI readiness should begin with an honest inventory of what data the core exposes, how that data can be accessed, what it costs to access it and what contractual or technical limitations stand in the way. In that sense, preparing the core for AI is also a vendor-evaluation exercise. 

The goal is not modernization for modernization’s sake. Banks do not necessarily need to rip out the core to take advantage of AI. They do need an architecture, and vendor relationships, that allow them to put their own data to work.

Next Move: Banking

Artificial intelligence is reshaping the future of banking, but its impact extends far beyond automation and efficiency. In Next Move: Banking, Virginia Heyburn, Principal Strategist at Engage fi, explores how declining AI costs could drive greater technology spending as financial institutions apply AI to more decisions across their operations. She also examines why core banking contracts are becoming increasingly important to long-term innovation, particularly when it comes to data access, ownership rights, and the flexibility to adopt emerging technologies.

As banks navigate these changes, understanding how technology investments translate into meaningful business outcomes is critical. Virginia challenges the assumption that more AI-powered experiences are the key to attracting younger generations, encouraging financial institutions to focus on what truly drives customer relationships. Together, these episodes offer insights into the evolving economics of AI, the strategic importance of data rights, and how financial institutions can make smarter technology decisions that support long-term growth.

Newsletter Image
Play button
Why Your Data Contract is Your AI Strategy

7 mins

Play button
Why Cheaper AI Doesn't Mean Lower Costs

8 mins

Play button
More AI Won't Win Gen Z Banking Customers

7 mins

×

Digital Banking Disintermediation: The Nubank and SoFi Warning

Digital-first financial institutions are reshaping competition, not simply by offering a strong mobile experience, but by operating with different cost structures, distribution models and approaches to customer relationships. Built without large branch networks or decades of layered technology, many can serve and acquire customers at a lower cost than traditional institutions.

That cost advantage can create strategic flexibility. Digital competitors may have more room to compete on deposit pricing, invest in customer acquisition, expand product offerings and improve the customer experience while preserving profitability. For boards and executives, AI is part of this broader competitive question, not a standalone technology issue or one addressed solely through an AI-vendor policy.

Many digital-first competitors are now regulated banks operating at meaningful scale. Traditional institutions should continue to apply appropriate safety, soundness and governance standards, but they also need to understand how competitors are using automation, data and technology architecture to reduce the cost to serve and deepen customer relationships.

Nubank and SoFi illustrate two distinct versions of this pressure: one centered on operating efficiency at scale and the other on expanding the breadth of the customer relationship.

Nubank: What a Sub-20% Efficiency Ratio Means

Nubank is an example of how different the cost structure of a large digital-native financial institution can look compared with a traditional bank. Nu Holdings reported a 17.6% efficiency ratio in Q1 2026 and 19.5% in Q2 2026. In Q2, it served 138.9 million customers, generated approximately $1.06 billion in quarterly net income and reported a 33% return on equity. For comparison, the FDIC reported an aggregate 61.87% efficiency ratio for U.S. community banks in Q1 2026.

This is not a direct apples-to-apples comparison. Nubank operates in different markets and has a different business model, regulatory environment, scale and distribution model than a U.S. community bank. However, the difference is useful for showing how significantly the cost of serving customers can vary between operating models.

Nubank's efficiency also should not be attributed to AI alone. The company was built as a digital-first institution without the same branch footprint or decades of accumulated technology infrastructure carried by many traditional banks. It also benefits from significant customer scale, digital distribution and operating leverage. AI is increasingly part of that environment. Nu reported that AI agents now handle more than 60% of customer-support conversations in Brazil at or above human parity, while its monthly cost to serve per active customer was approximately $1.

For traditional banks, the concern is not that they should be expected to reach a 20% efficiency ratio. The issue is the competitive flexibility available to an institution with a much lower cost to serve. A lower operating-cost base can create more room to invest in pricing, customer acquisition, technology, products and customer experience while maintaining profitability. Nubank provides an example of why banks are increasingly looking at efficiency, automation and technology architecture together rather than treating them as separate issues.

SoFi: A Different Digital-First Competitive Model

SoFi provides a different example of how a digital-first financial company can compete. Rather than using SoFi primarily as an efficiency-ratio comparison, the stronger story is the breadth of the customer relationship. SoFi's model is designed to bring a member in for one financial need and then expand that relationship across saving and spending, investing, lending, credit cards, cryptocurrency, financial guidance and other services.

In Q2 2026, SoFi reported 15.8 million members and 24.4 million products. More importantly, 51% of new products opened during the quarter were opened by existing members. This shows that a significant amount of product growth is coming from customers who already have a relationship with SoFi rather than only from acquiring new members.

SoFi has continued expanding the number of services within that relationship. In June 2026, it introduced SoFi Coach, an AI-powered financial guide designed to provide personalized guidance based on a member's finances. It also expanded into small-business lending. The company remained GAAP profitable in Q2 after Q1 marked its tenth consecutive GAAP-profitable quarter.

SoFi and Nubank are useful together because they show two different competitive pressures for traditional banks. Nubank demonstrates what a significantly lower-cost operating model can look like at scale. SoFi demonstrates what can happen when a digital provider gets existing customers to use more products within the same financial relationship. For banks, the competitive issue is therefore broader than whether they offer digital banking. It also includes how efficiently they can serve customers, how much of the customer's financial relationship they can retain and how effectively they can use technology and data to deepen that relationship.

Artificial intelligence is changing how consumers discover, evaluate, and select financial products, creating new challenges and opportunities for bank marketers. As more consumers turn to AI-powered search tools for financial recommendations, traditional search engine optimization (SEO) may no longer be enough to ensure visibility. Financial institutions must also consider how their brands appear in AI-generated answers and what influences whether their products are recommended. 

The next stage of competition may become even more difficult to see. As AI assistants become capable of evaluating providers in real time, a consumer may simply ask, “Where should I put my money to get the best return?” The answer could compare rates, terms, convenience and product fit across the market in seconds.

In that environment, a local savings rate that is 50 basis points higher than another traditional institution may not be enough to matter when a digital alternative offers 300 or 400 basis points more and can be opened with a few taps. The risk is not merely whether a bank has a mobile app. It is whether the institution remains visible, competitive and relevant when AI increasingly guides the customer’s financial decisions.

That is the disintermediation challenge: not losing the account overnight, but losing the moment when the customer decides where the money, loan or next financial need should go.

Call to Action: The Vendor Contract Review

AI is evolving far faster than the typical five- to seven-year vendor contract cycle. A technology agreement signed today reflects today's capabilities, pricing and assumptions. Long before that contract expires, new AI capabilities could change what the technology can do, what it should cost and what banks should expect from their vendors. Without the right flexibility, institutions risk spending years paying for yesterday's capabilities while competitors move forward. 

That makes vendor strategy part of AI strategy. Banks should understand not only what their technology providers offer today, but what their contracts allow them to do tomorrow. Data-access rights, APIs, integration costs, AI functionality, pricing structures, renewal terms and the ability to adopt new capabilities should all be part of the conversation. 

But successful AI strategies do not begin with visions of a science-fiction future. They begin with the very real problems employees, executives and boards encounter every week: manual processes that consume hours, data employees struggle to access, customer interactions slowed by disconnected systems, repetitive work that limits capacity and technology constraints that have simply become accepted as part of doing business. 

That is where an AI assessment should start. Examine the institution's existing technology, workflows, data and vendor relationships, then identify where AI can meaningfully remove friction, increase capacity, improve decision-making or create a better experience for employees and customers. 

At Engage fi, we approach AI differently. The goal is to determine where it can create measurable value, what technology and data are required to make that possible, and whether existing vendor relationships support, or stand in the way of that future. 

The institutions best positioned for AI will not necessarily be the ones chasing the most tools. They will be the ones that understand where AI can change their business, whether their technology can support it and what needs to change when it cannot. 

Guiding Financial Insitutitions Toward a Stronger Future

Position your bank or credit union to thrive as a modern, efficient, and sustainable organization. Focus on supporting your customers and members today while adapting to meet future challenges.