After Moody’s downgrades, regional bank technology had to argue in unit costs

The week

On August 7, 2023, Moody’s Investors Service downgraded the credit ratings of ten small and midsize U.S. banks. Moody’s is one of the major credit rating agencies, whose grades tell lenders and investors how likely a borrower is to repay. It also placed six larger banks on review for possible downgrade, and changed the outlook on others to negative, a warning that a cut could follow.

The names were familiar. The one-notch downgrades included M&T Bank, Pinnacle Financial Partners, Prosperity Bank and BOK Financial. The banks placed on review included Bank of New York Mellon, U.S. Bancorp, State Street and Truist Financial. Eleven lenders, including Capital One, Citizens Financial and Fifth Third, were moved to a negative outlook. These were established franchises, not the fast-growing, concentrated banks that had failed in the spring.

American banking had just had a frightening few months. Silicon Valley Bank and Signature Bank failed in March, and First Republic followed in May. Those failures were fast and dramatic: depositors pulled money in a rush, and the banks ran out of cash to pay them. By summer the panic had faded, and the open question was what would come next. Moody’s gave its answer.

Moody’s reasons described a slow squeeze rather than a sudden crisis. First, higher funding costs, as deposits moved to higher-yielding alternatives. When savers can earn more in a money market fund than in a bank account, banks must pay more to keep their deposits. Second, pressure on profitability, because the gap between what banks earn on loans and what they pay on deposits was narrowing. Third, unrealized losses on securities as rates rose: bonds bought when rates were low lose market value when rates climb, even if the bank never sells them. Fourth, exposure to commercial real estate, particularly offices, whose values and occupancy had come under strain.

Moody’s added two points that sharpened the picture. It expected a mild U.S. recession by early 2024, with asset quality likely to decline. And it warned that banks with large unrealized losses not reflected in their regulatory capital ratios could be vulnerable to a loss of confidence while rates stayed high. Most banks below the largest tier could leave unrealized losses on securities available for sale out of their regulatory capital. Silicon Valley Bank’s reported capital ratios had not reflected its own large unrealized losses either.

The action came five months after the failures of Silicon Valley Bank and Signature Bank and three months after First Republic’s. S&P Global Ratings, another major rating agency, downgraded several regional banks later in August.

The days around it were crowded. On July 26 the Federal Reserve had raised its policy rate to a range of 5.25 to 5.5 percent. On July 27 the federal banking agencies proposed the “Basel III endgame” capital rules for banks with $100 billion or more in assets. On July 31 the Fed’s survey of senior loan officers showed banks tightening credit standards and seeing weaker loan demand. On August 1 Fitch downgraded the U.S. government’s own credit rating to AA+.

Regulators were watching the same risks. In late July, the federal banking agencies had updated their liquidity guidance to stress contingency funding plans, a bank’s written playbook for raising cash quickly if deposits run. The update also stressed readiness to borrow from the Federal Reserve’s discount window, the Fed’s standing lending facility for banks that need short-term cash.

That July 28 update was joint. It came from the Federal Reserve, the FDIC, the Office of the Comptroller of the Currency and the National Credit Union Administration, so it reached credit unions as well as banks. It asked institutions to review and update their contingency funding plans regularly. Institutions that included the discount window in their plans were expected to keep operational readiness to use it, including periodic test transactions. Credit unions were pointed to the NCUA’s Central Liquidity Facility as their counterpart. The agencies presented this as a restatement of existing expectations, not a new rule. Examiners treated it as a list of things they would now check.

An emergency backstop was still open. The Fed’s Bank Term Funding Program, created on March 12 after Silicon Valley Bank failed, offered loans of up to a year against Treasury and agency securities valued at par rather than market value. It removed the immediate pressure to sell underwater bonds. It did nothing for the slower problem Moody’s described, which was that those bonds and many loans still earned less than deposits now cost.

Together, these moves told regional bank leaders that thin margins and close scrutiny of liquidity would last. That message reached every budget in the bank, including technology.

The bank failures of March 2023 were a liquidity crisis: depositors wanted their money faster than the banks could produce it. Moody’s August actions described something slower. Funding costs were rising, net interest margins were narrowing, and loan books carried commercial real estate exposure that would take years to resolve.

The net interest margin needs explaining, because it drives a traditional bank’s earnings. It is the difference between the interest a bank earns on its loans and investments and the interest it pays to depositors and other lenders, expressed as a share of its earning assets. When depositors demand more and loans reprice slowly, that gap shrinks, and so does the money available for everything else, including technology.

The squeeze had a timing problem built in. Deposit costs reset quickly, because savers can move money in an afternoon. Fixed-rate mortgages, auto loans and bonds bought in 2020 and 2021 reset slowly, as they mature or are paid down. A bank holding long, low-yielding assets would see its margin compressed for as long as rates stayed high, which nobody in August 2023 could forecast with confidence.

For regional and community bank technology leaders, that meant a different budget conversation. Projects justified by growth or customer experience alone became harder to fund. Projects that lowered cost per account, or gave treasury faster sight of deposits and liquidity, moved up. The shift had three parts: the cost case, the liquidity data case, and the vendor contracts that sit under both. All three rest on where a regional bank’s technology money actually goes.

Moody’s rates only a small share of U.S. banks, mostly larger ones. Its logic applied well beyond them. Community banks and credit unions faced the same deposit competition and the same securities losses, without a rating agency to announce it. Their boards saw the squeeze in quarterly earnings and in examiners’ questions instead.

Where a bank’s technology money goes

At the center of every bank is the core banking system, often just called “the core.” It is the master ledger: the software that holds every account, posts every deposit and withdrawal, calculates interest and produces statements. Everything else in the bank depends on it. Most regional and community banks do not own their core. They rent it from a small number of specialist companies, called core processors, which run the software in their own data centers and charge the bank per account or per transaction.

Around the core sit other systems. Card processing handles debit and credit card transactions. Digital banking platforms provide the mobile app and website customers use. Loan origination systems take applications and manage approvals. Treasury and risk tools model interest rates and liquidity. Networks, data centers and cybersecurity tools hold it all together. Each connection between these systems is an integration, often built using an API, an application programming interface, which is a defined way for one piece of software to request information from another.

Renting rather than owning makes sense for most smaller banks. Building and maintaining a core system demands deep specialist skills, constant regulatory updates and round-the-clock operations that only large institutions can afford to run themselves. The trade-off is dependence. Once a bank’s accounts, products and integrations are built around one processor’s system, changing course is slow and costly, and the price of staying is set in negotiations the bank enters with limited options.

Size and charter change the mix

The picture differs by institution. A community bank with a few billion dollars in assets often buys almost everything from its core processor in one bundle: core, cards, online banking, even network management. Its technology staff may number a handful, and most of its budget is a set of monthly invoices. Its bargaining power lies in the bundle, which it can price as a whole.

A regional bank between $10 billion and $100 billion in assets usually runs a more varied estate. Acquisitions leave it with duplicate systems, sometimes more than one core for a period. It employs its own engineers and buys more specialist software. It also crosses regulatory lines that change its revenue: at $10 billion, debit interchange fees become capped under the Federal Reserve’s Regulation II, and the CFPB becomes its consumer compliance supervisor. Both raise the cost of serving a checking customer.

Credit unions follow their own path. Many use cores built for credit unions, and many share services through credit union service organizations, cooperatives owned by several credit unions. Their members own them, they pay no federal income tax, and they answer to the NCUA rather than the bank agencies. Their margin pressure in 2023 was similar, but the board conversation runs through member value rather than shareholder return.

Across all three types, the split between run and change is hard to see from the outside. A project to replace online banking may be booked as change in its first year and as run cost for the next seven. Hosting adds another layer. When a bank moves workloads from its own data center to a processor’s or a cloud provider’s, capital spending on hardware turns into monthly operating expense. The total may fall while the line the CFO watches rises. A budget that does not show both views invites the wrong argument.

Two features of this picture matter for what follows. First, most of the spending is on keeping things running, not on building new things. Second, much of it is locked into a few long contracts with vendors who know how hard it is for a bank to leave them.

The cost case

A bank’s technology spending is mostly run cost: core processing, card processing, digital banking platforms, network and data center, and the staff who operate them. Change projects are the smaller share. In a margin squeeze, the CFO asks first about run cost, and asks it in units. What does it cost to maintain a checking account for a year? What does each digital login, each wire, each loan booked cost in technology and operations?

Few regional banks could answer precisely in 2023. Costs sat in general ledger lines by vendor and department, not by product or transaction. Building a unit cost model takes allocation work: mapping contracts and staff to the products they support, then dividing by volumes. It is not glamorous, but it changes the conversation with the board. A technology leader who can show that digital account opening costs a fraction of branch opening per account can defend that investment in a cost-cutting year. One who presents a project list cannot.

Building the model

The work follows a predictable sequence. Finance pulls a year of vendor invoices and the general ledger lines for technology and operations staff. Each cost is assigned a driver: accounts for core fees, card transactions for card processing, active users for digital banking, headcount for workplace software. Shared costs, such as the network and security tools, are spread by a simple, agreed rule. The result is divided by product volumes and checked with business line heads, who will spot allocations that make no sense.

The first version will be rough, and that is acceptable. A model that is right within a reasonable margin and refreshed each quarter is more useful than a precise one built once. The finance team should own it, with technology supplying the vendor and system data. If technology owns it alone, the CFO will treat it as advocacy.

The model connects to measures the board already knows. The efficiency ratio, noninterest expense divided by revenue, is the headline cost measure analysts use for banks. A narrowing margin pushes it up even when costs are flat, because revenue falls. Unit costs explain what sits beneath that ratio and which levers move it.

Suppose a regional bank has a technology budget spread across dozens of vendors. Its finance and technology teams spend a quarter tracing each contract and each team to the products it supports: consumer checking, small business lending, mortgages, treasury services. They divide by the number of accounts and transactions. For the first time, the bank can see that a mortgage booked through its online channel costs far less in technology and operations than one booked through an older paper-heavy process. It can also see that one product line is carrying the cost of three separate systems that do nearly the same job. The board now discusses choices, not line items.

The model also exposes waste. Duplicate applications from past acquisitions, unused software licenses and per-account fees on dormant accounts all show up once costs are tied to volumes. Retiring them funds part of the investment the bank still needs.

Where the savings usually are

Unit cost models in banks tend to point to the same few places. Paper is one: printed statements, mailed notices and check handling carry postage, printing and processing costs that digital delivery avoids. Overlap is another: two loan origination systems, three document archives or several fraud tools inherited through acquisitions. Licenses are a third, since banks often pay for more seats or modules than they use. Manual back-office work is a fourth, such as reconciliations and exception queues that staff clear by hand.

The revenue side of the unit matters too. For banks above $10 billion, debit interchange is capped, and the Federal Reserve proposed in October 2023 to lower that cap further. A lower cap reduces what each active checking account earns, which makes the cost side of the same account more important. A bank that tracks only cost per account, without revenue per account, can draw the wrong conclusion about which products to grow.

Fixed costs behave badly in a downturn

Per-account pricing looks variable, so it should fall when accounts fall. Contracts often blunt that with minimum fees, volume tiers and charges for products the bank no longer sells. A bank that loses accounts during deposit competition can find its cost per account rising. The model should show which costs actually flex with volume and which do not, because that determines how much the bank can save by shrinking.

The model has a trade-off worth naming. Unit costs favor scale and automation, so they can understate the value of relationship services that drive deposits, such as treasury management for business clients. A good model separates cost to serve from value of the relationship, and leaves the final call to the business line rather than the cost spreadsheet. A business client who costs more to serve but brings large, stable deposits may be worth every cent.

A second limit is allocation itself. Any rule for spreading shared costs is a choice, and a different rule can make a product look better or worse. Teams arguing for their own budgets will challenge the rules that hurt them. The defense is to publish the rules, keep them stable from quarter to quarter and change them only by agreement. A model that shifts every time someone objects loses the CFO’s trust.

The liquidity data case

Silicon Valley Bank lost a large share of its deposits in a day. Supervisors and boards drew a technical conclusion: banks needed to see deposit flows, concentrations and available collateral far faster than monthly or even daily reports allowed. The agencies’ July 2023 liquidity guidance pointed in the same direction, stressing tested contingency funding plans and operational readiness to use the discount window.

Several capabilities follow. Treasury needs intraday views of deposit balances and outflows by segment, including uninsured and concentrated relationships. Uninsured deposits are balances above the federal deposit insurance limit; they tend to move fastest in a scare, because their owners have the most to lose. Concentrated relationships are a few large customers, or a cluster of similar customers, who could withdraw at the same time.

Treasury also needs collateral already positioned and documented with the Federal Reserve, so that borrowing is a matter of hours rather than days. Collateral is the loans or securities a bank pledges to secure a loan. Pre-positioning it means the paperwork is done before the cash is needed. And treasury needs scenario tools that model outflows under stress. Each depends on data from core systems that were built for posting, not for real-time analysis. The core is excellent at recording that a withdrawal happened. It was rarely designed to tell the treasurer, at 11 a.m., how much uninsured money has left since breakfast.

Why the core struggles with intraday data

Many cores record transactions in two steps. During the day they apply a provisional “memo post” so balances look current to the customer. Overnight, a batch run makes the postings final and produces the reports most of the bank relies on. Treasury’s morning report is therefore a picture of yesterday. Getting an intraday view usually means reading the memo-post data, the payment systems or the digital banking logs, and reconciling them later.

Insurance status is harder still. Deposit insurance covers $250,000 per depositor, per insured bank, per ownership category. Working out what is uninsured means grouping accounts by owner and ownership type, which cores do not always hold cleanly. Banks that tried to calculate uninsured balances quickly in 2023 often found duplicate customer records and missing ownership codes. Fixing that data is slow, unglamorous work, and it is the foundation of everything else.

Pricing data is liquidity data

Liquidity is also a pricing problem. As policy rates rose, banks had to decide how much of each increase to pass on to depositors, a ratio treasurers call the deposit beta. Pass on too little and money leaves. Pass on too much and the margin shrinks faster. Making that call segment by segment needs data on balances, rates paid, exception pricing and outflows by customer type. Many banks priced exceptions for large depositors by email and spreadsheet, so the true cost of their deposits was not visible in one place.

Some banks also used reciprocal deposit networks, which spread a large customer’s money across several banks so that each piece stays under the insurance limit. The bank keeps the relationship and the funding. The arrangement adds its own data needs: which balances are placed, at what cost and on what terms. A treasurer cannot judge how stable the deposit base is without that information.

Making the discount window usable

Borrowing from the Fed is not instant even for a bank with good collateral. The bank must have signed the lending agreements set out in the Fed’s operating circular, pledged collateral and had it valued, and set up the staff and systems to request funds. Loans pledged as collateral need to be described in files the Reserve Bank can accept. The July guidance’s point about periodic test transactions was aimed at this machinery. A bank that has never borrowed finds its gaps at the worst moment.

Take a community bank on a bad morning. A rumor spreads on social media about a peer bank, and some business clients start moving money. Under the old setup, the treasurer sees yesterday’s balances in a report that arrives at 9 a.m. and must phone branch managers for anything newer. Under the new setup, a dashboard refreshed every hour shows outflows by segment, flags that three large business depositors have each moved funds since opening, and shows how much the bank could borrow against collateral already pledged at the Federal Reserve. The treasurer calls the CEO with numbers, not guesses, and the asset-liability committee meets with a plan in hand. Nothing about that dashboard is exotic. What makes it work is that the data feeds were defined in advance, someone owns each one, and the bank has rehearsed using them.

These projects were easier to fund in 2023 because examiners and rating agencies were asking about them. They still needed a clear scope. A bank does not need a real-time data platform for every purpose. It needs reliable, frequent feeds of a defined set of liquidity data, an owner in treasury, and a test of the contingency funding plan that shows the data arrives when needed.

An end-to-end test of the contingency funding plan has a recognizable shape. The asset-liability committee runs a stress scenario, such as a rapid outflow of uninsured business deposits over two days. Treasury pulls the actual data feeds the plan names, at the times the plan says they will be available. The bank makes a small test borrowing at the discount window, or from a Federal Home Loan Bank, to confirm the operational steps work. Afterward, someone writes down what was late, missing or wrong, and who will fix it by when.

That record is the evidence an examiner will ask for. A plan that has been tested and has an issues log looks very different from one that has only been approved. The test also tells the technology team which feeds matter most, which is a better guide to investment than a vendor’s product roadmap.

Ownership matters here because the data crosses functions. Deposit data comes from the core and digital channels, collateral data from the investment portfolio and loan systems, and the decisions sit with treasury and the asset-liability committee, the senior group that manages the balance between what the bank owns and what it owes. A named data owner for each feed, with a defined refresh frequency and quality checks, keeps the project from becoming an open-ended platform build.

The scale of the task varies. Banks with $100 billion or more in assets already faced detailed liquidity rules and frequent liquidity reporting to the Fed, so their gap was speed and coverage. A $3 billion community bank might start from spreadsheets. A credit union’s backstop is the Central Liquidity Facility, directly or through a corporate credit union, as well as the discount window, so its readiness tests look different. The data principle is the same in each case.

The vendor contracts

Most of a regional bank’s technology cost sits in a small number of long contracts, above all with its core processor. Those contracts typically run five to seven years, price per account or per transaction, and carry significant termination fees. Banks often signed them in a different rate environment, with growth assumptions that no longer held.

The squeeze changed the negotiating stance. Banks had reasons to revisit pricing tiers, minimum volume commitments and charges for data access and integration. They had reason to ask for rights they had previously traded away: access to their own data in usable formats, the ability to connect third-party applications without punitive fees, and service levels with real credits. A service level is the vendor’s promise about availability and speed; a credit is the money back when the promise is broken. Without real credits, a service level is just an aspiration.

Clauses that decide the cost of leaving

Several contract terms matter more than the headline price. Early termination fees are often calculated as a share of the fees remaining on the contract, so leaving early can cost most of what staying would. Deconversion fees are charged to extract the bank’s data when it leaves. Automatic renewal clauses roll the contract forward unless the bank gives notice months in advance. Bundles tie digital banking or card processing to the core, so changing one means renegotiating all.

A bank reviewing these terms should list each one, its cost and its notice date, in a single register the CFO can see. Many banks keep contracts in procurement folders and track only expiry dates. A notice deadline missed by a few weeks can add years to a relationship.

Regulators had just restated their expectations here. In June 2023 the Federal Reserve, the FDIC and the OCC issued final interagency guidance on third-party risk management. It describes a life cycle of planning, due diligence, contract negotiation, ongoing monitoring and termination. It expects contracts to address data rights, performance standards, audit access and orderly exit. Those are the same terms a cost-focused negotiation wants. A bank can present its renegotiation as risk management as well as cost control, which strengthens its hand.

Smaller institutions negotiate differently

A few large processors serve most U.S. banks and credit unions, which limits every buyer’s options. Smaller institutions have found ways to offset their size. Some negotiate through industry groups or advisory firms that benchmark pricing across many contracts. Credit unions can buy through service organizations shared with peers. Community banks sometimes rely on bankers’ banks, which provide services to independent banks. The common thread is that pricing information is the scarcest asset in these negotiations, and pooling it narrows the gap with the vendor.

The practical approach was to start renewal planning two or three years before expiry, benchmark pricing, and keep a credible alternative in view even if switching was unlikely. A vendor that believes a bank could leave negotiates differently from one that knows it cannot.

There is a limit to what renegotiation achieves. Core conversions, the process of moving every account from one core system to another, are expensive and risky, and processors know it. A conversion is a little like moving house while still living in it, with every customer expecting their mail to arrive on time. Many banks gained more by consolidating smaller contracts, retiring overlapping tools and tightening governance of new purchases than by fighting the largest vendor. A rule that every new technology purchase must show its effect on unit cost, or on a named regulatory requirement, is a cheap and effective filter.

Pressing too hard has its own cost. A processor squeezed on price may assign fewer staff to the account, slow its response to change requests or deprioritize the bank’s projects. The better outcome trades something the vendor values, such as a longer term or a broader product commitment. In return the bank gets things it values more, such as data access, lower integration fees and enforceable service levels. The bank should know in advance which of those it would give up.

What else the budget decides

For a regional bank in 2023 and after, the technology budget showed up in earnings, in rating reviews, in examinations, in customer service and in contracts that would bind the bank for years.

Earnings. When the net interest margin narrows, every dollar of run cost weighs more heavily on profit. Technology is often one of a bank’s largest expense lines after staff, so its efficiency shows up directly in earnings, and in the ratios that rating agencies and analysts watch. Earnings are also how a bank rebuilds capital, so a cost saving today becomes loss-absorbing capacity later.

Ratings and funding. Rating agencies look at profitability and liquidity. A downgrade can raise the cost of borrowing and make some depositors and counterparties more cautious. Evidence that a bank can see and manage its liquidity quickly is part of the story it tells to agencies and investors. For banks without a rating, the same role falls to analysts, large depositors and correspondent banks, who make their own judgments.

Supervisors. Examiners were asking pointed questions about contingency funding plans and discount window readiness. A bank that could show working data feeds, pre-positioned collateral and a tested plan had a better conversation than one with a binder and good intentions. Examiners also review vendor management under the third-party guidance, so contract registers and exit plans serve both purposes.

Customers. Squeezing cost is not the same as squeezing service. A bank that cuts the digital channels customers rely on, or slows down the business services that bring in stable deposits, may save money and lose the relationships that pay for everything else. Unit cost models, used well, help leaders see which is which.

Strategic freedom. Contracts signed today shape what a bank can do for the next five to seven years. Access to its own data and reasonable integration terms decide whether a bank can add a new product from a smaller partner, or must wait for its core processor to offer one. They also shape the bank’s options in a merger, since a bank that may buy or be bought needs to know what a core conversion would cost and when its contracts allow it.

Speaking the CFO’s language

Across all three parts, the skill that mattered was translation. Technology leaders who could express their portfolio in the language of margin, liquidity and risk kept their budgets. Those who presented roadmaps lost them to those who did.

A bank doing this well tends to have a unit cost model that the CFO trusts, a short list of liquidity data feeds with named owners and a tested contingency funding plan, a calendar of major contract renewals years ahead, and a simple test applied to every new purchase. Its board sees technology as a set of costs and capabilities tied to products and risks, not as a list of projects.

The presentation matters as much as the analysis. A good board pack fits on a single page. It shows technology spending by product, the trend in unit costs, the status of the liquidity data feeds and the contract renewal calendar, which gives directors something to question. A thick appendix of project status reports gives them nothing to decide. The CFO should present the cost numbers, with the technology leader explaining what drives them.

The case against budgeting by unit cost

There is a serious objection. Some of the most important technology spending lowers no unit cost at all. Cybersecurity, resilience testing and the retirement of unsupported systems mostly prevent losses that have not happened. A budget process that rewards only measurable savings will starve them, and a bank that has cut its security budget to protect its margin has traded a visible cost for an invisible risk.

The objection is right about the danger and wrong about the remedy. Spending of that kind should not compete with efficiency projects on unit cost. It should be argued in its own terms: the regulatory requirement it meets, the loss it reduces and the risk appetite the board has set. The filter suggested above allows for exactly that by accepting a named regulatory requirement as a reason. What the CFO’s language rules out is spending justified by neither cost nor risk.

A second objection is that cost discipline can lock in the bank’s current shape. Banks that only cut may miss the investment that would change their economics, such as a move to digital account opening or a new core. The unit cost model is the best answer to that too. It shows how much the current shape costs, which is the baseline any transformation must beat.

The usual ways to get it wrong:

• Cutting across the board. Trimming every budget by the same percentage protects waste and starves the projects that lower cost or reduce liquidity risk.
• Building a platform instead of answering a question. Liquidity data projects that try to solve every reporting need at once tend to run late and over budget.
• Negotiating only on price. Per-account fees matter, but so do data access, integration charges and service credits, which shape what the bank can do for years.
• Starting renewals too late. A bank that begins talking to its core processor a few months before expiry has no realistic alternative and little bargaining power.
• Letting the spreadsheet decide. Unit costs inform the choice; they do not replace judgment about which customer relationships matter.
• Cutting security to protect the margin. Loss prevention needs its own budget line and its own case, measured against risk appetite rather than unit cost.

What the budget should show about unit cost, liquidity data and contracts

The questions below are for the next budget cycle. Each should be answered with a number, a date or a test result, and each answer should have a named owner.

• What does it cost us, in technology and operations, to run a checking account for a year, and how does that compare across our channels, and which overlapping applications could we retire?
• How quickly can treasury see deposit outflows by segment, including uninsured and concentrated balances, on a stressful day, and how much collateral is already pledged at the Federal Reserve?
• When did we last run our contingency funding plan end to end, and did the data arrive when the plan needed it?
• When does our core contract expire, when will renewal planning start, what is our credible alternative, and do our contracts give us our own data in usable form?
• For each new technology request, can the sponsor show its effect on unit cost or on a named regulatory requirement?

A bank that answers all five has a technology budget its CFO can defend to the board, the examiners and the rating agencies at once.

With hindsight

The wave of failures many feared in mid-2023 did not arrive, though pressure persisted. New York Community Bancorp’s large loss and dividend cut in early 2024, driven largely by commercial real estate, showed that the slow-moving risks Moody’s had flagged were still working their way through bank balance sheets. The failure of Republic First Bank in April 2024 was another reminder that the squeeze had not ended.

The New York Community episode was instructive in its own right. Within weeks of its January 2024 results, the bank needed an equity investment of more than $1 billion from a group of investors. It later renamed itself Flagstar Financial. Republic First, a Philadelphia bank, was closed by regulators and its deposits taken over by Fulton Bank. Both stories ran on the risks Moody’s had named the previous August: funding costs, commercial real estate and thin profitability.

Parts of Moody’s outlook did not hold. The mild recession it expected by early 2024 did not come, and the U.S. economy kept growing through that year. The July 2023 rate increase proved to be the last of that tightening cycle. Rates then stayed on hold for more than a year, which kept deposit costs high and stretched the margin squeeze well beyond what many banks had budgeted for.

The costs of the spring failures also reached surviving banks directly. In November 2023 the FDIC adopted a special assessment to recover the cost of protecting uninsured depositors at Silicon Valley Bank and Signature Bank. It fell on banks with more than $5 billion in assets and was based on their uninsured deposits. For the banks that paid it, uninsured deposit data had become a line on the expense statement as well as a liquidity metric.

The emergency backstop wound down on schedule. The Bank Term Funding Program stopped making new loans in March 2024, which left the discount window as the main standing source of Fed liquidity. That made the operational readiness the July 2023 guidance had stressed matter more.

Relief came gradually. The Federal Reserve began cutting interest rates in September 2024, which eased funding costs over time. Banks did not have to pay quite as much to keep deposits, and margins had room to recover. Easier conditions, though, are exactly when disciplined spending habits tend to slip, which is why the work done under pressure deserves protecting. The Fed held rates through most of 2025 before cutting again that autumn.

The changes in technology spending lasted longer than the crisis. Contingency funding plans and discount window readiness stayed on supervisory agendas, so the liquidity data work that began in 2023 did not become optional when markets calmed. Examiners continued to expect banks to show, rather than say, that they could see their deposits and raise cash quickly.

The contract side moved too. In October 2024 the CFPB finalized its personal financial data rights rule, which would require many banks to give customers and authorized third parties access to account data through standard interfaces. The rule was challenged in court, and in 2025 the Bureau said it would reconsider it. Whatever its final form, it pushed data access, the term this article urged banks to negotiate, onto the regulatory agenda.

Consolidation then raised the stakes. Bank merger announcements picked up markedly in 2025 under a more receptive regulatory approach. Every merger brings a core conversion or a choice between two cores, and the contract terms negotiated in 2023 and 2024 shaped what those deals cost. Banks that knew their termination fees and notice dates could price a merger more accurately than those that did not.

The habit of arguing in units proved durable for a simple reason: it works in good years as well as bad ones. A bank that knows its cost per account can decide where to invest when money is plentiful, as well as where to cut when it is scarce. A bank that has renegotiated data access and integration terms can add new services more cheaply whatever the rate environment.

The three cases described here, unit cost, liquidity data and vendor terms, remained the practical way for a regional bank technology leader to argue for budget while margins were thin. They also give boards a reusable test: ask that every technology proposal be explained in the bank’s own language of margin, liquidity and risk.