Inflation’s first casualty in bank IT was the change budget

The week

On November 30, 2021, Federal Reserve Chair Jerome Powell testified before the Senate Banking Committee alongside Treasury Secretary Janet Yellen, at a scheduled hearing on oversight of pandemic relief under the CARES Act. Under questioning, he said it was probably a good time to retire the word “transitory” as a description of inflation.

For most of 2021 the Fed’s message had been that rising prices were a passing effect of the pandemic. Supply chains were snarled, spending had shifted from services to goods, and the mismatch was expected to sort itself out. Dropping the word was a public acknowledgment that the mismatch might not sort itself out, and that the central bank was prepared to act.

The timing was awkward in two ways. A week earlier, on November 22, President Biden had announced that he would nominate Powell for a second term as chair. The testimony came from a chair whose own position still awaited Senate confirmation. And on November 26 the World Health Organization had designated a new coronavirus variant, Omicron, as a variant of concern. Markets fell sharply that day. The first U.S. case was confirmed on December 1. A central bank turning more hawkish into a fresh wave of uncertainty was not what most forecasters had expected that autumn.

Powell also signaled that the Fed would consider ending its asset purchases sooner than planned. Since the start of the pandemic, the Fed had been buying large quantities of Treasury and mortgage-backed bonds, a way of pushing money into the financial system and holding down long-term interest rates. At its November 3 meeting the Federal Open Market Committee, the group that sets interest rate policy, had started reducing those purchases by $15 billion a month. Finishing sooner meant easing off the accelerator earlier than markets expected, and it opened the door to raising rates earlier too.

The data behind the shift was hard to argue with. The consumer price index, which tracks the prices households pay for everyday goods and services, had risen 6.2 percent in the year to October, according to the Bureau of Labor Statistics. That was the fastest pace in about three decades. The November figure, published on December 10, came in at 6.8 percent, the highest since 1982. In October the Social Security Administration had announced a 5.9 percent cost-of-living adjustment for 2022, its largest in about four decades. Any contract tied to a price index was about to feel the same arithmetic.

The labor market was tight as well. Voluntary quits reached record highs in the Bureau of Labor Statistics’ job openings survey that autumn, and employers in almost every sector were raising pay to hire and keep staff. Technology workers were among the most sought after.

Software vendors had moved before the Fed did. In August 2021 Microsoft had announced its first broad increase to commercial list prices for Microsoft 365 and Office 365 in about a decade, effective March 1, 2022. Microsoft 365 E3, a common enterprise license, went from $32 to $36 per user per month. Customers on multi-year enterprise agreements were protected until renewal, then the new prices applied. Other vendors followed with their own increases over the next two years.

Events then moved in the direction Powell had signaled. On December 15 the FOMC doubled the pace of its taper to $30 billion a month, putting the end of purchases in early 2022. The committee’s projections showed three quarter-point rate increases in 2022. The Fed began raising its policy rate, the benchmark short-term rate that ripples through to loans, deposits and mortgages, in March 2022, and kept raising it through 2022 and into 2023. Consumer price inflation peaked at 9.1 percent in June 2022, the highest in roughly four decades.

Most coverage at the time focused on what all this meant for borrowers, savers and markets. Inside banks and credit unions, attention went to deposits, loans and the value of bond holdings, which is where the biggest dollars were.

A less visible effect was already starting in a less glamorous place: the technology budget. Inflation raises the cost of software contracts, cloud bills and engineers’ salaries along with groceries and rent. In most financial institutions, the way those costs were budgeted meant that one kind of spending would absorb the shock without anyone deciding it should. Budgets for 2022 were being finalized in exactly these weeks, mostly on assumptions set before the word “transitory” was retired.

Most commentary on the Fed’s pivot concerned balance sheets: deposit costs, loan demand and the value of securities portfolios. For technology leaders at banks and credit unions, the first effect was inside the IT budget itself.

A typical institution’s technology spending splits into “run,” the cost of keeping existing systems operating, and “change,” the projects that modernize them. Run costs are largely committed and rise on their own. Budgets are usually set as a percentage increase on last year. When inflation jumped, the gap between those two growth rates had to come from somewhere, and in most institutions it came out of change.

That is a claim about mechanics, not about anyone’s judgment. Nobody in a budget meeting in December 2021 proposed slowing modernization. The slowdown was built into the way the numbers were assembled, and it showed up months later as projects that slipped, shrank or were never started.

The sections that follow take this in order. The first explains how run money and change money grow, and why the line between them matters. The next three follow the channels of the squeeze: committed run costs, people costs, and the change portfolio that absorbed both. Then come how the squeeze differed by type of institution, the strongest case against our argument, and the governance that keeps change funded.

Run money and change money

A household budget offers a decent analogy. Some spending is fixed by commitments already made: rent or mortgage, insurance, phone contracts, subscriptions. Some is discretionary: a kitchen renovation, a new car, a course of study. If income rises by a set amount each year and the fixed bills suddenly jump, the renovation is what gets postponed. Nobody decides to cancel it. There is simply less left over.

A bank or credit union works the same way at larger scale. The run side covers the systems that keep the institution open for business every day. At the center is the core banking system, the main ledger that records every account, balance, deposit, loan and transaction. Many institutions, especially smaller ones, do not run their own core. They rent it from a core processor, a specialist company that operates the system on behalf of hundreds of banks and credit unions under long contracts.

Around the core are other services, often also bought from outside. Card processing handles debit and credit card transactions. Digital banking powers the website and mobile app. Networks connect branches and data centers, and a long list of software products covers lending, fraud detection, compliance and more.

The change side covers projects that make things better: a new mobile app, a faster account opening process, a replacement for an aging core, a move of systems to the cloud. Change spending is where an institution improves what it offers customers and reduces future costs. It is also the part of the budget that nobody is contractually obliged to pay.

What matters is how each side grows. Run costs grow according to contracts, prices and usage. Change budgets grow according to whatever percentage the institution decides to add each year. In a period of low inflation, those two rates are close, and the arrangement feels stable. When inflation jumps, run costs can outrun the budget, and the difference has to come out of change.

Where the line between them falls

The boundary between run and change is partly an accounting choice, and that choice shapes what gets squeezed. Under U.S. accounting rules for internal-use software, costs incurred while building an application can usually be capitalized. They go on the balance sheet and are expensed gradually through amortization over the software’s useful life. Costs of planning, training, data conversion and ongoing operation are expensed as they occur.

Since 2020, a related standard has let companies capitalize many implementation costs for cloud computing arrangements and spread them over the term of the hosting contract. Institutions that are not public companies have adopted similar treatment on their own schedules, and the details are a question for their auditors.

The consequence is easy to miss. Every capitalized project becomes a future run cost. A digital banking upgrade finished in 2020 shows up in 2022 as amortization, alongside the new subscription fees it introduced. When an institution measures “run” as everything already on the books, last year’s change is this year’s run. Run growth therefore has two engines: price inflation on existing commitments, and the tail of every project recently finished.

That second engine is the reason decommissioning matters so much. A new system that does not retire an old one adds its cost to the run base permanently. A new system that does retire one can leave run roughly flat. Inflation magnified the difference, because the old system’s contract was escalating too.

Why budgets lag

Most institutions build the next year’s technology budget in the autumn, approve it in November or December, and revisit it in the spring at the earliest. Budgets for 2022 were therefore drafted on cost assumptions from mid-2021, when the official forecast still called inflation transitory. A budget that adds a modest percentage to last year’s total was a reasonable guess in August. By March it was out of date.

Credit union boards, community bank boards and the technology committees of larger banks usually approve the total, not the split. If the total is fixed and the run portion rises, the change portion falls without a vote. The mechanism needs no villain and no error. It needs only a fixed total and costs that move on their own.

Committed run costs

Run budgets are dominated by multi-year agreements with core processors, card processors, digital banking providers, network carriers and software vendors. Contract terms signed in the low-inflation years, including escalators and volume floors, embedded assumptions that later bit.

Escalators and floors

An escalator is a clause that raises the price automatically each year. Some are fixed percentages. Others are tied to an inflation index such as the consumer price index, sometimes with a floor and sometimes with no cap. A volume floor is a minimum the institution pays regardless of how much it actually uses.

Both feel harmless when inflation is low and growth is steady. In late 2021 those terms started to execute. Index-linked increases that had added one or two percent a year began to add far more, and they compounded, because each year’s increase is calculated on top of the last. A clause with a cap at, say, five percent protected the institution. A clause with no cap transferred the whole inflation shock to the buyer.

Volume floors bit differently. A credit union that had signed for a minimum number of accounts or transactions, expecting growth, paid for that minimum even when branch traffic fell or members moved activity to other channels. A floor turns an optimistic forecast into a fixed cost.

The core processor contract

The core contract is usually the largest single item in a smaller institution’s run budget, and the hardest to change. A small number of large processors serve most U.S. banks and credit unions. Contracts commonly run five to seven years and renew automatically unless notice is given well in advance.

Leaving early typically triggers a termination fee, and converting to a new core is itself a project of a year or more. Those terms give the incumbent a strong hand at renewal. They also mean that an institution’s ability to respond to inflation in its largest cost line depends on where it happens to be in its contract cycle.

An institution that renewed in 2020 faced escalators for years with little room to negotiate. One whose contract expired in 2022 could, in principle, run a competitive process. Few did, because a conversion costs change money, which was the very thing being squeezed. The squeeze therefore protected incumbents.

Negotiating with an escalator in place

An institution mid-term still had options, though none was free. It could offer a longer term in exchange for a cap on index-linked increases, accepting less flexibility later for predictability now. It could consolidate several products with one vendor and ask for a blended price, at the cost of deeper concentration. It could prepay part of a contract when it had surplus liquidity, which fewer institutions had once deposit costs rose.

The moment of greatest bargaining power usually comes when the institution wants something new from the vendor: an added module, a new channel, a merger conversion. Buyers who tied those requests to a review of existing escalators did better than those who negotiated each item separately.

Contracts also carry exit terms that matter more in an inflationary year. Deconversion fees, data return formats and the notice period for non-renewal all determine whether a competitive process is a real option or a bluff. Procurement teams that kept a renewal calendar, with notice dates eighteen to twenty-four months ahead of expiry, had room to act. Those that discovered a renewal ninety days out did not.

Consumption and per-seat pricing

Two newer cost lines behaved differently. Cloud infrastructure means renting computing power and storage from a large provider instead of owning servers; the bill rises and falls with usage. Software-as-a-service, often shortened to SaaS, means software accessed over the internet on a subscription, usually charged per user per month.

Vendors facing their own wage and energy costs raised list prices at renewal. Usage-based bills grew with activity whether or not anyone approved a larger budget. Unlike a capital project, there is no single decision point at which someone signs off on the increase. The cost simply arrives on the invoice, and finance sees it a quarter later.

Per-seat pricing hides a specific kind of waste. Licenses are assigned when someone joins and are often not reclaimed when they leave or change role. In a year of high staff turnover, which late 2021 was, the gap between licenses paid for and licenses in use tends to widen. Productivity suites, collaboration tools and security agents are all priced this way.

Unit costs as the diagnostic

The practical response was to make run costs visible as unit costs. That means cost per account served, per transaction processed, per active digital user. A run budget measured in unit costs shows whether a rise came from growth, from price, or from waste such as idle cloud capacity and unused licenses. Only the last can be cut quickly. Institutions that could separate the three negotiated better at renewal and found savings that protected project money.

Consider a community bank whose digital banking bill rises noticeably in a year. Without unit costs, the finance committee sees only a bigger number. With them, it can see that active app users grew, the price per user rose at renewal, and a few hundred licenses were assigned to former employees. The first is good news. The second is a negotiation. The third is a quick fix. Each calls for a different response, and only unit costs make them visible.

Building unit costs takes work. Someone has to map each invoice line to a driver, agree on definitions with finance, and keep the mapping current as contracts change. For a large bank this is a financial operations team. For a credit union it may be one analyst and a spreadsheet updated quarterly. Either version beats a single total, because it turns a budget argument into a set of specific questions about price, volume and waste.

People costs

The second pressure was wages. Demand for engineers, security analysts and data specialists was already high in 2021, and broad wage inflation raised it further. Banks and credit unions competed with technology firms, with fintechs and with each other, and paid more to hire and to retain.

Wages and retention

Pay is usually budgeted once a year, but the market does not wait. When a valued engineer receives an outside offer in March, the institution either matches it from somewhere or loses the person. That somewhere was usually the change budget, because run commitments could not be touched and headcount budgets were already set.

Retention carried a less visible cost. When experienced staff left, institutions lost knowledge of legacy systems that is hard to buy back, especially for core platforms and custom integrations. An integration is the connection that lets two systems exchange information, such as the link between the core and the mobile app. Older integrations are often poorly documented and understood by only a few people.

Each departure slowed projects that depended on that knowledge, and replacement hires took months to become productive. Smaller institutions felt this most. A credit union with a dozen technologists can lose a third of its capacity for core conversion work when two senior people leave. A large bank losing the same two people barely notices in aggregate, but may find that one particular system has lost its only expert.

Contractors and consultants

Contractors and consultants passed their own labor costs through in hourly rates. For institutions that staffed modernization with contractors, every rate increase landed directly on the project budget rather than on the run line. The same inflation that pushed up run costs also made each hour of change work more expensive, squeezing the change budget from both sides.

Fixed-price statements of work shifted some of that risk to the supplier, at a premium. Time-and-materials contracts left it with the institution. Neither was wrong, but many institutions had not chosen deliberately. Contracts written in 2020 on time-and-materials terms carried the full rate increase into 2022 project costs.

Outsourcing and the oversight that stays

Managed services and shared platforms became more attractive where they spread scarce skills across many clients. Credit union service organizations are a good example. A credit union service organization, or CUSO, is a company owned by credit unions to provide shared services, so that many small institutions can pool resources. These arrangements come with their own escalators and service limits.

Outsourcing a function moves the labor cost into a contract. It does not remove the need for someone inside the institution who understands the service well enough to oversee it. Supervisors expect a financial institution to manage the risks of its vendors. In July 2021 the Federal Reserve, FDIC and OCC had proposed joint guidance on third-party risk management, which they finalized in June 2023. Both versions assume a bank has people who can assess a provider, monitor it and plan for leaving it.

That oversight role is itself a people cost, and it rose with wages like any other. An institution that outsourced to save money and then cut the vendor-management analyst to save more had moved risk off the budget and onto the examination report.

The change portfolio

With run costs rising automatically and budgets rising by a set percentage, change spending shrank in real terms. Rising rates then raised the hurdle for long-payback projects.

The cost of capital arithmetic

The cost of capital is, roughly, the return an institution needs to earn to justify tying up money in a project rather than putting it elsewhere. When interest rates rise, that bar rises too. A multi-year modernization program with benefits five years out looked weaker at a higher cost of capital than it had a year earlier. A dollar of savings in five years is worth less today when money earns more in the meantime.

The effect is not linear. Discounting hits distant benefits much harder than near ones. A project that pays back within a year barely notices a two-point rise in the discount rate. A core replacement whose savings arrive in years four through ten can lose a large share of its calculated value. Rising rates therefore did not cut all projects equally. They tilted the portfolio toward quick wins and away from structural change.

For some banks, wider net interest margins in 2022 eased the pressure for a time. The net interest margin is the gap between what a bank earns on loans and investments and what it pays on deposits. When rates rise, that gap can widen for a while, because loan yields reprice faster than deposit rates. Credit unions and community banks with thin margins, fixed-rate mortgage books and concentrated processor contracts had less room.

What got cut first

When change budgets shrink, the visible projects are rarely the first casualties. Executives defend the new mobile app and the branch refresh. The cuts fall on work with no customer-facing launch date: test automation, documentation, data cleanup, security remediation and, above all, decommissioning.

Those are exactly the activities that hold down future run costs. Skipping a decommissioning step leaves an old system running, with its own escalating contract. Skipping test automation makes every later release slower and dearer. The squeeze fed on itself: cuts to unglamorous change raised run costs, which left even less for change the following year.

Why staging survived

The squeeze was not evenly harmful. It forced discipline that many programs needed. Large single-release programs became hard to defend, and programs broken into stages with measurable value at each stage fared better.

A digital account opening rebuild that ships in quarterly increments, each tied to a metric such as completion rate or time to fund, can survive a budget cut. A three-year core replacement with one go-live date cannot easily be trimmed without being stopped. The staged program can lose a phase and keep going; the single release either gets its full funding or stops.

Staging has limits. A core conversion has a fixed cutover weekend, and some regulatory changes have a single compliance date. For those, the protective move is a contingency reserve sized against inflation, set aside at approval and not raided for other work.

Reporting change in real terms

Most committee packs report the change budget in nominal dollars. In 2022 that flattered it. A change budget that rose slightly in dollars while contractor rates and salaries rose faster bought less work than the year before.

A simple correction is to report change capacity alongside change dollars: the number of engineer-months, or the number of planned releases, the budget actually funds. When dollars hold steady and capacity falls, the committee can see the squeeze directly. The measure is crude, and it invites argument about productivity. That argument is useful, because it moves the discussion from the size of the budget to what the budget buys.

How the squeeze differed by institution

Inflation hit every institution, but the channels weighed differently depending on size, charter and how technology was sourced.

Credit unions

Credit unions are member-owned cooperatives. They cannot raise capital by issuing stock, so their net worth is built mostly from retained earnings. Every dollar of unexpected technology cost comes out of the same earnings that build capital, which made boards cautious about discretionary spending.

Many smaller credit unions rely heavily on a core processor and on CUSOs for shared services. That concentrates run costs in a few contracts. It also weakens one lever banks have: the National Credit Union Administration lacks authority to examine credit union vendors directly. The temporary authority it held for the Y2K transition lapsed in 2001, and NCUA chairs have repeatedly asked Congress to restore it. Credit unions are therefore more reliant on their own contract terms and due diligence when a vendor’s prices or service levels change.

Community banks

Community banks share the dependence on a core processor, with similar contract terms. Their federal regulators can examine the services a bank outsources under the Bank Service Company Act, and the FFIEC agencies jointly supervise the largest technology service providers. That supervision addresses risk, not price. It does nothing to restrain an escalator.

Community banks did often benefit from rising rates in 2022, because their deposit costs lagged. That margin tailwind is real, but it was temporary, and in 2023 deposit competition reversed much of it. A community bank that treated the 2022 margin as permanent and approved a large change program on that basis found the funding thinner a year later.

Regional and large banks

Larger banks run more of their own technology, employ thousands of engineers and spend heavily with cloud providers and software vendors. Their exposure ran more through people costs and consumption pricing than through a single core contract. They also had more negotiating power, dedicated procurement teams and the ability to shift work to lower-cost locations.

Their risk was different: scale made waste harder to see. A few percent of idle cloud capacity or unused licenses is a rounding error in one department and a large sum across a bank. Large institutions that built financial operations teams, with engineers and finance staff reviewing usage together, found more savings than those that relied on annual procurement reviews.

The same credit union, two ways

Suppose a credit union’s technology budget grows by a fixed percentage each year, roughly two-thirds run and one-third change. In the year inflation takes hold, its core processor contract escalates on an inflation index and its digital banking provider raises prices at renewal. Its cloud bill climbs as members use the app more, and it gives its small technology team raises to keep them from leaving. Every one of those increases is reasonable on its own. Together they eat most of the budget increase and a slice of last year’s change money as well.

Nobody at the credit union votes to slow modernization. The board approves the same percentage increase it always does. When the projects are funded in the spring, there is noticeably less to go around. A planned improvement to online loan applications slips a year. A cloud migration that would have retired an old server room is paused, so the server room keeps costing money.

Now give the same credit union three habits. Its committee sets the run and change split as a deliberate decision, so the shortfall is visible and debated rather than absorbed. Its finance team tracks unit costs, so it spots the unused licenses and idle cloud capacity and recovers some money quickly. And its loan application project is staged, so the first improvement still ships on time even though later phases are pushed back. It faces the same inflation, but it keeps moving.

Institutions caught mid-conversion

Institutions that were in the middle of a core conversion or a merger integration when inflation took hold had the least room of all. A conversion runs on a vendor’s fixed schedule. It usually requires paying for the old and new systems at once during parallel running. It relies heavily on contractors and specialists whose rates were rising.

Delaying the cutover rarely saved money, because the double running costs continued and the conversion team had to be held together longer. Most such institutions pushed through and absorbed the overrun from the rest of the change portfolio. That was usually the right call. It also meant that other modernization work at those institutions effectively paused for a year or more, a cost that rarely appeared in the conversion’s own business case.

The case against protecting change

The strongest objection to this argument is that run spending is the institution. Payments must clear, cards must work and the core must post transactions every night. In a year of rising costs, funding run first is responsible management, and cutting change is how a prudent institution absorbs a shock.

A second objection is about timing. Inflation raised the price of every hour of project work in 2022. Deferring a project in a year of peak contractor rates and scarce engineers can be cheaper than pushing ahead. On this view, the squeeze on change was a rational pause, not an accident.

A third concerns vendors. Core processors and digital banking providers invest in their own platforms, and some modernization arrives through their upgrades rather than through the institution’s project list. A shrinking change budget may simply mean the institution is buying modernization through its run contracts.

Each objection has force. Funding run first is correct, and pausing some projects in 2022 was wise. The objection fails on one point: a residual is not a decision. An institution that deliberately deferred projects, knowing the cost, made a choice. One that discovered in the spring that projects had shrunk did not. The difference matters most later, when costs stabilize and the deferred work has to be restarted with fewer people who remember why it mattered.

Vendor-led modernization is real but partial. A processor’s upgrade improves its own product. It does not retire the institution’s legacy integrations, clean its data or simplify its estate. Those still need change money.

Governance that keeps change funded

The tools for protecting a change budget are well known, if not always used.

• Set the run and change split explicitly at the board or committee level, rather than leaving change as the residual.
• Stage-gate major projects, releasing funds at each gate on evidence of value.
• Pair every new investment with a decommissioning target, so that retiring old systems reduces future run cost.
• Re-test business cases at more than one cost of capital, so that a rate change does not surprise the committee.
• Read escalators as carefully as headline prices at signing and renewal, and seek caps on index-linked increases.
• Reclaim licenses as part of offboarding, with the identity team and procurement sharing one list.

Each tool has a cost. Stage gates add governance overhead and can slow programs that need sustained momentum, such as a core conversion with a fixed vendor schedule. Decommissioning targets require business owners to give up familiar systems, which they resist. An explicit run and change split exposes trade-offs that executives may prefer to leave implicit. Escalator caps are harder to win from a dominant processor than from a smaller vendor.

Those costs are real, but they are smaller than the alternative, which is a modernization agenda that stalls by default while the run budget grows.

A strategy set by invoices

For directors and senior executives, the issue goes beyond IT housekeeping. Technology modernization is how a financial institution stays competitive on customer experience, keeps pace with fraud and security threats, and reduces long-term operating costs. If the change budget is silently absorbed by run cost growth year after year, the institution falls behind without anyone ever deciding that it should.

There are also risk and regulatory angles. Aging systems that are never replaced become harder to secure, harder to change and more dependent on a shrinking pool of people who understand them. Examiners review technology and vendor risk, and a modernization program that stalls for budget reasons can leave known weaknesses in place longer than intended. A finding that cites an unsupported system is harder to answer when the replacement project was deferred by default.

The split between keeping the lights on and improving the institution is a strategic choice. When it is left as a residual, it is made by contracts and invoices rather than by leaders.

What the technology committee should be shown

• What share of our technology budget is run and what share is change, and how has that changed over the last few years?
• Which of our major contracts have automatic price escalators, what are they tied to, and when do they renew?
• What do we spend per account, per transaction and per active digital user, and why did each number move?
• Which of our modernization programs deliver measurable value in stages, and which have a single go-live date years away?
• For each new system we are building, which old system will it retire, and when?

A committee pack that answers those five questions takes a few pages. Finance owns the split and the unit costs. Procurement owns the escalator schedule. The program office owns staging and decommissioning dates. The chief technology officer owns the narrative that ties them together.

Inflation did not announce itself in the IT budget. It arrived as a series of invoices, renewals and salary negotiations. The institutions that noticed in time were the ones that had already decided how much of their budget was for keeping the lights on, and how much was for building what comes next.

With hindsight

Inflation proved far more persistent than the Fed expected in late 2021, and the tightening cycle that began in March 2022 was the fastest in decades. The Fed raised its target range eleven times, from near zero to 5.25 to 5.50 percent by July 2023, and held it there for more than a year. The word Powell retired turned out to be well retired.

Consumer price inflation peaked at 9.1 percent in June 2022, according to the Bureau of Labor Statistics, and fell back to around 3 percent by mid-2023. The effects on technology budgets ran on long after the headlines moved on.

Technology costs at financial institutions rose through contract escalators, wages and vendor price increases, the same three channels visible in that first winter. None of them required a decision by the institution. They arrived because of commitments already made and markets already moving.

Vendor pricing kept resetting. Microsoft’s increase took effect in March 2022 as announced. After Broadcom completed its acquisition of VMware in November 2023, it stopped selling perpetual licenses and moved customers to subscription bundles. Many enterprises, banks among them, reported higher renewal costs as a result. Virtualization software runs in most bank data centers, so this was a run cost that rose by a vendor’s decision rather than by an index.

The labor channel eased in part. Large technology companies cut staff in late 2022 and 2023, which loosened hiring for some engineering roles. Salaries reset during 2022 stayed reset, however, and specialist skills in security, data and cloud remained expensive.

The rate environment also reshaped deposit competition. In spring 2023 it contributed to the failures of Silicon Valley Bank, Signature Bank and First Republic Bank, which pushed many institutions toward caution on discretionary spending. Discretionary spending is, by definition, where change budgets live. So a second force joined the first: on top of costs squeezing modernization from below, caution pressed on it from above.

Supervision moved in the same direction. In June 2023 the Federal Reserve, FDIC and OCC finalized their joint guidance on third-party relationships, setting expectations for planning, due diligence, contract negotiation, monitoring and exit across the life of a vendor relationship. Contract terms, including pricing mechanics and termination rights, became something an examiner might reasonably ask about. NCUA continued to press Congress for vendor examination authority.

Inflation later eased, and the Fed began cutting rates in September 2024. It made further cuts in late 2024 and three more in 2025, ending that year at 3.50 to 3.75 percent. Contracts and salaries reset in 2022 and 2023 still carried higher cost bases forward. Lower inflation means prices rise more slowly. It does not mean they fall back. An escalator that ran in 2022 stays in the base, and a salary raised to keep a key engineer stays raised.

A new claim on change budgets arrived as well. From 2023 onward, generative AI pilots and the data work they require competed for the same discretionary money that modernization had been waiting for. Institutions whose change budgets had shrunk by default found themselves choosing between finishing old programs and starting new ones.

By this point in 2026, the institutions with the most room for new work are those that measured run costs as unit costs, set the run and change split deliberately and staged their programs. Those that let change absorb every increase are still catching up.

What has held up best from the late-2021 picture is the mechanism, not any particular forecast. Any budget built as last year plus a percentage hides a choice. When costs that rise on their own share a budget with spending that does not have to happen, the second will pay for the first unless someone decides otherwise, explicitly and in advance.