Why Your Bank Spends More on Workarounds Than on Banking

How much does your bank spend on software you never see?

The average mid-market bank runs its deposit accounts, loan servicing, and general ledger on a core system built decades ago. That core costs a few million a year. On top of it sits a digital banking platform for the customer-facing app, a CRM for the branch staff, a data warehouse for reporting, a loan origination system, a compliance workflow, and an onboarding tool. Each one is a separate vendor contract with its own integration, its own support agreement, and its own annual price increase.

A former COO of a $21 billion bank described the math: $7.5 million a year for the core, $30 million a year for everything layered around it. The core was the cheapest line item on the technology budget, and every other line existed because the core couldn't do what the bank needed.

Retail customers still left for neobanks, not because those neobanks offered something the bank couldn't build, but because the customer wanted the easiest experience and the bank's stack couldn't deliver one.

The three vendors underneath

Fiserv, FIS, and Jack Henry collectively serve over 70% of US banks, according to a Kansas City Federal Reserve survey. Fiserv alone holds roughly 42% share across multiple core platforms, a portfolio assembled through decades of acquisition. Jack Henry sits at 21%, FIS at 9%.

The concentration creates a specific kind of lock-in. Switching cores means moving to a vendor with a smaller install base and less mature integration ecosystem, or moving laterally to another product within the same oligopoly. Contract provisions at the Big Three limit mid-term exits and impose penalties for early termination, which the Kansas City Fed flagged as a structural barrier to competition.

A mid-market bank unhappy with its Fiserv core can migrate to Jack Henry or FIS, which involves the same vendor dynamics and a multi-year conversion. Or it can try one of the newer entrants like Thought Machine, 10x Banking, or Mambu, betting the institution on a platform with a fraction of the install base and integration partners.

Most banks do neither. They keep the core and buy their way around it.

The layering tax

A bank that can't replace its core compensates by adding platforms on top.

Q2 Holdings provides the digital banking interface, the app and web portal that customers actually interact with. nCino, built on Salesforce's infrastructure, handles loan origination and account opening workflows. Snowflake or a similar data platform aggregates reporting across the fragmented systems underneath. Salesforce Financial Services Cloud manages customer relationships for the branch and commercial banking teams. Each integration solves a problem the core created by not solving it first.

A mid-market bank typically runs 50 to 150 of these third-party integrations. Each one adds a vendor relationship, an API surface, a data reconciliation requirement, and an annual renewal negotiation. The bank's technology team spends more time managing the connections between systems than building anything new.

The economics compound in one direction. A bank that integrates Q2 for digital banking can't pull it out without rebuilding the customer-facing experience, and once nCino handles loan origination, the workflows, approval chains, and compliance logic live inside nCino's platform. Each layer becomes its own switching cost, independent of the core switching cost underneath.

It's a house where every room was renovated by a different contractor, and now none of the doors quite close because the framing isn't square. The homeowner can't fix the framing without gutting every room that was built around it.

The $30 million that COO spent wasn't waste. Each vendor solved a real operational gap. But the aggregate cost of solving those gaps individually, through separate contracts and integrations, exceeded what a unified system would cost by a factor of four. The layering tax is the difference between what the bank pays for its technology in pieces and what it would pay if the pieces were designed to work together.

Where the money goes instead

Celent estimates that IT consumes 18-20% of non-interest expenses across global retail banking, with total spend reaching $285 billion in 2025 and projected to hit $307 billion in 2026. That industry average understates what individual mid-market banks face, because it includes the largest institutions that amortize technology costs over tens of millions of accounts.

A bank with $5 billion in assets and 200,000 customers spreads its core and integration costs across a much smaller base. The per-customer technology cost at a mid-market bank can exceed what a top-20 bank spends, while the mid-market bank generates less fee revenue per account to cover it.

Jack Henry's 2026 survey found that 88% of banks and credit unions expect to increase technology budgets, with 45% of community banks planning increases of 40% or more. The spending is accelerating into the same layered architecture, not away from it. Banks are buying more integrations to keep pace with customer expectations while the underlying core remains unchanged.

Technology budgets grow every year while customer experience improves slowly, because most of the new spending goes toward maintaining and connecting the existing layers rather than building new capability.

The customer side of the equation

While banks layer technology to work around their cores, customers are finding it easier to leave.

J.D. Power tracked annual bank switching rates rising from roughly 4% in 2019 to 9% in 2025. Account aggregation through Plaid and similar services reduced the friction of moving direct deposits and automatic payments. A process that once required visiting a branch and manually re-entering account numbers across every biller can now be completed from a phone in an afternoon.

The reasons for switching haven't changed much. Mintel found 44% of switchers cite high fees as the primary motivation. But the bar for what counts as a better alternative has dropped. Chime, which has roughly 25 million customers, offers fee-free checking with earlier access to direct deposits. The feature set is narrower than a full-service bank, but for customers whose primary interaction is depositing a paycheck and paying bills, the comparison favors the simpler product.

A mid-market bank competing with Chime on digital experience is fighting with one hand tied to a core system that processes transactions in overnight batches while the neobank posts deposits in real time. The layers of software on top of the core can mask some of that latency, but they can't eliminate it. A $5,000 check deposited at a mid-market bank still triggers a $500 first-day availability limit and a multi-day hold, not because the technology to verify funds faster doesn't exist, but because the bank's business rules and core workflows haven't been rebuilt to use it.

Over 1,400 financial institutions have joined FedNow, but most participate as receive-only. They can accept instant payments but can't originate them, because their core systems weren't designed to process outbound transactions outside of batch windows.

The branch paradox

US banks added a net 535 branches in 2025, reversing a 15-year streak of annual contraction. Growth concentrated in the Southeast, with Florida and Texas leading new openings, while the Northeast continued losing locations.

Banks are investing in physical infrastructure at the same time their digital stack consumes a growing share of operating expenses. Branches serve a different function than they did a decade ago. Fewer teller transactions, more advisory conversations and complex product sales that customers still prefer to handle face-to-face.

But each new branch adds another node to the technology stack. The branch staff needs access to the CRM, the loan origination system, the account opening workflow, the compliance tools. Every integration that the central technology team manages now extends to another physical location with its own connectivity, hardware, and support requirements.

A bank opening 20 branches while maintaining 80 integrations is multiplying its technology surface without simplifying the architecture underneath. The branch investment pays off through deeper customer relationships, but the technology cost of extending every integration to every new location takes a growing cut of that return.

Why conversions fail

EY estimates core banking conversions take 18 to 36 months, with decommissioning timelines extending 12 to 24 months beyond original estimates in the majority of programs. IBM's research found that 94% of core modernization projects exceed their original timelines.

A bank converting from Fiserv to Thought Machine doesn't just replace the core. It has to reconnect every system that was integrated with the old one: the digital banking platform, the loan system, the compliance workflow, the data warehouse, the card processor, the ATM network. Each reconnection is its own project with its own timeline, testing requirements, and failure risks.

TSB's 2018 migration from Lloyds' core to a new Sabadell platform locked 1.9 million customers out of their accounts and cost £300 million in remediation. AIB's Flexcube migration exceeded its budget by €84 million. These aren't obscure failures. They're the reason every mid-market bank's board asks what happens to their customers if the conversion goes wrong.

The rational response, given the risk, is to keep layering. The layering tax grows predictably, and nobody gets locked out of their account.

The replacement wave arriving late

The companies building core replacements, Thought Machine, 10x Banking, Mambu, and newer entrants, all target the same thesis: a bank shouldn't need 80 integrations to do what a single platform could handle. Thought Machine's revenue crossed $100 million in 2025 with clients including JPMorgan Chase and Standard Chartered, and 10x Banking reached EBITDA-positive with Chase UK and Westpac. Mambu targets the mid-market with an API-first cloud-native core.

A newer category of vendors is pitching a middle path: run a modern core alongside the existing one in production, migrate workloads incrementally, and avoid the all-or-nothing cutover. The approach lowers conversion risk but adds its own complexity — two cores running in parallel means two sets of data, two reconciliation surfaces, and a migration that can stall indefinitely if nobody forces the switchover.

Adoption in the US mid-market has been slow regardless of approach. The Big Three's contract structures, integration dependencies, and the conversion risk described above combine to create inertia that a better product alone can't overcome.

The sales cycle in core banking runs through relationships and industry conferences, not advertisements. Vendors in this market don't run Google or LinkedIn campaigns. Deals happen at the American Bankers Association conference through handshakes with bank executives who have known each other for decades. A core banking startup competing against a vendor that has served a bank for 15 years is selling against a relationship, not a feature set.

The roughly 4,200 FDIC-insured banks in the United States represent the largest banking market in the world by institution count. The mid-market segment, banks with $1 billion to $50 billion in assets, includes roughly 1,000 to 1,500 institutions that are large enough to feel the layering tax acutely but small enough that a failed core conversion could threaten the bank's viability.

The bill keeps growing

The $37.5 million that one COO spent was the cost of operating a $21 billion bank on a technology stack designed around a core that couldn't do what the bank needed. That number will be higher next year. Vendors raise prices, new regulatory requirements add compliance integrations, and competitive pressure from neobanks and fintechs acquiring their own bank charters pushes the mid-market bank to add another layer.

The core stays, the layers accumulate, and the technology budget grows while the customer experience gap with neobanks widens.