Trustdeskapp.
Implementation Guide

Ten Core Banking Migration Mistakes That Blow Up Budgets, and What to Do Instead

Most core replacements that overrun were set up to fail in the contract and the plan. Here are ten migration errors, the fix for each, and the clause that makes the vendor share the risk.

Trustdeskapp editorial team · November 2026 · 8 min read

Most overruns are locked in before the first line of configuration

If you're about to sign for a new core, the decisions that will set your budget are already on the table: how you'll cut over, who owns the data, and what the contract says when a rehearsal fails. A core banking migration rarely overruns because the software doesn't work. It overruns because the bank priced a clean, linear project and then paid for a messy one at change-request rates.

The ten mistakes below come up again and again in core replacements, at a 300,000-account regional lender and at a multi-country group alike. For each one you'll get what to do instead and the contract wording that stops the cost from landing only on the bank. All cost figures are indicative 2026 market ranges in USD. They vary widely by contract, region and vendor, so confirm them with vendors and integrators.

Advertisement

Treating a big-bang cutover as a date instead of a rehearsed event

Mistake 1: committing to a big-bang go-live without full dress rehearsals. People argue big bang vs. phased migration as if it were a philosophy. It's really a question of rehearsal. A big bang can work for a bank with a simple product set and one legal entity, but only if the full cutover has been run end to end, inside the real time window, more than once.

FactorBig bangPhased (by product, branch or segment)
Best fitSmaller banks, simple products, one entityMulti-entity groups, complex lending books
Dual-running periodShort, often weeksLong, often many months and sometimes over a year
Interim integration workMinimalSignificant (bridges between old and new core)
RollbackHard after the first business dayLimited to the migrated segment

Instead: fix the number of mock conversions up front and score each one against pass criteria agreed in advance. Clause: make go-live conditional on passing the final rehearsal, not on a calendar date, and require the vendor to support a documented rollback plan for an agreed period after cutover.

Underfunding data migration and blurring who owns it

Mistake 2: budgeting data cleansing as a side task. A legacy core that has run for two decades holds duplicate customers, orphaned collateral records, dormant accounts with odd fee histories, and free-text fields staff used as workarounds. Data work is one of the most common sources of slippage, and programs that treat it as a minor line item usually find it costs far more than planned.

Mistake 3: assuming the vendor or integrator owns the data. They own the tools and the load scripts. The bank owns what the data means, and only its operations staff know why a field is populated the way it is.

Instead: start profiling and cleansing well before the build. Six to twelve months ahead is a sensible target. Put it under a named data owner on the bank side. Then size the data budget from what profiling finds, not from a rule of thumb. Clause: write a RACI for data mapping, transformation and sign-off into the statement of work, along with defined data quality thresholds, so rework caused by unclear ownership doesn't turn into a billable change request by default.

Rationalizing products after the build has started

Mistake 4: migrating every legacy product and deciding later which ones to retire. Each product variant you carry over needs configuration, mapping, test cases and reconciliation rules. Hypothetically, a bank with 400 savings and loan variants that could reasonably shrink to 60 is paying to migrate complexity it already wants gone.

Instead: finish product rationalization before configuration starts. That includes the customer communications and consumer-protection steps needed to move or close legacy products, and those differ by market. Notice periods and fair-treatment rules aren't the same in the UK, the EU, India or Brazil, so bring in local compliance advice early. Clause: tie fixed-price configuration to an agreed product catalog, with a unit price for every product added later. Scope growth then shows up, and its price is known in advance.

Running too few parallel runs and reconciling only balances

Mistake 5: planning one or two parallel runs. Differences in interest accrual, fees and penalties, and month-end and year-end processing only show up across real cycles. One parallel run catches load errors. It won't catch a compounding rounding difference that only appears at quarter-end.

Mistake 6: reconciling only at general-ledger balance level. Totals can match while individual accounts are wrong in both directions.

Instead: plan at least three parallel runs covering a month-end, ideally including a quarter-end. Reconcile account by account on balances, accrued interest, next payment dates and limits. Clause: write the reconciliation tolerance into the acceptance criteria (for example, no unexplained variance above a set amount), and keep the vendor's defect-fix obligation open until that tolerance is met.

Customizing the new core to behave like the old one

Mistake 7: rebuilding legacy quirks as custom code. Business teams will ask for the old screens, the old batch timing and the old exception handling. Each customization adds testing now and makes every later upgrade slower. That throws away much of the reason for buying a modern core from vendors such as Temenos, Thought Machine, Mambu, Finastra or Infosys Finacle.

Instead: adopt the vendor's model by default and route exceptions to a design authority that is allowed to say no. If that body can't refuse requests, custom work will pile up quarter by quarter. Clause: cap customization effort in the contract, and have the vendor confirm in writing whether each extension will survive upgrades.

Leaving integrations and supervisors until late

Mistake 8: discovering the integration estate halfway through. Payments switches, card processors, AML screening, credit bureaus, regulatory reporting and data warehouses add up to dozens of interfaces, and at larger banks often well over a hundred. The undocumented point-to-point feeds are the ones that break at cutover.

Mistake 9: telling the regulator late. Supervisory expectations differ by country. The EU's Digital Operational Resilience Act (DORA), the UK's operational resilience regime, the Reserve Bank of India's directions on IT outsourcing and the Monetary Authority of Singapore's technology risk guidelines all cover material technology change and third-party risk. Many supervisors, there and elsewhere, expect to hear about it early.

Instead: build a complete interface inventory during discovery, and brief supervisors when the program starts, not just before go-live. Clause: require the vendor to support regulatory audits and information requests, and to provide exit and resilience documentation.

Pricing the program without dual running, backfill and decommissioning

Mistake 10: leaving out the hidden run costs. Legacy licenses and hosting keep running through parallel operation. Depending on the incumbent contract, a six-month slip can add anywhere from hundreds of thousands to several million USD. Your best operations experts also get pulled into the program, and someone has to cover their day jobs.

Hypothetical example: a mid-size bank budgets USD 30 million for its replacement. A six-month slip brings extra legacy run costs, longer integrator fees and backfill contractors. Together those can push the total past USD 36 million before a single change request is counted.

Instead: list dual running, backfill and legacy decommissioning as separate budget lines, and hold a contingency in the 15-25% range. A contingency below that is hard to defend once you have an honest view of the data and integration work. Clause: agree flexible extension terms with the incumbent vendor before you sign with the new one, and attach service credits or fee holdbacks to any milestones the new vendor misses.

Five contract clauses to settle before signature

You can negotiate conversion risk, but only before you sign. Work through this list with your legal counsel, who should advise on enforceability in your jurisdiction.

RiskClause to look for
Failed cutoverGo-live conditional on passing rehearsals; supported rollback window
Data reworkData RACI and quality thresholds in the SOW
Scope creepFixed product catalog with unit pricing for additions
Hidden defectsAccount-level reconciliation as an acceptance criterion
Schedule slipMilestone holdbacks on a negotiated share of fees (often discussed around 10-20%); incumbent extension terms

The decision rule: if the vendor won't accept go-live tied to rehearsals and account-level reconciliation as acceptance criteria, assume the overrun risk sits with you, and set your contingency at the top of your range.

Quick answers

Is a phased core banking migration always safer than a big bang?

No. Phasing limits the damage from any single cutover, but it adds interim integration bridges and a long dual-running period, and those can cost more than a well-rehearsed big bang. Smaller banks with simple product sets often do better with a big bang. Multi-entity groups usually gain from phasing.

How many mock conversions should we plan?

Three to five full mock conversions is a common planning range, with the last one or two run in the real cutover window on production-scale data. Agree pass criteria for each one in advance. The exact number matters less than the rule that go-live can't proceed until the final rehearsal passes.

What share of the budget should go to data migration?

No benchmark holds for every bank. The cost depends on how clean the legacy data is and how many products you migrate. Programs often find that profiling, cleansing, mapping and reconciliation cost much more than they first assumed, so profile your data before you commit to a number and treat data as a full workstream in its own right.

Who should own data quality, the bank or the vendor?

The bank owns what the data means and gives final sign-off. The vendor or integrator owns the migration tools and load processes. Write this split into a RACI in the statement of work so disputes over rework don't automatically become change requests.

Do we need to notify our regulator about a core replacement?

It depends on your jurisdiction and how material the change is. Many supervisors in the EU, UK, India, Singapore, the Gulf and elsewhere expect early engagement on major technology change and outsourcing. Requirements and timing differ, so confirm your obligations with your compliance team and legal counsel before the program starts.

When should we negotiate an extension with our incumbent core vendor?

Before you sign with the new vendor. Once the incumbent knows you're committed to leaving and your timeline is slipping, you lose most of your leverage. Aim for month-by-month extension terms at pre-agreed rates covering a realistic slip, which for many programs means at least six to twelve months.

Written by the Trustdeskapp editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.