Why signed programs slip at the seams, not in the vendor's product
The contract is signed. The statement of work says 9 to 14 months, and the board has a launch quarter in mind. The schedule rarely breaks inside the vendor's own product. It breaks at the seams, where the new front end meets your core, your card processor, your identity stack, the app stores and your regulator. Those parts of a digital banking platform implementation sit partly outside the vendor's control and partly outside your project team's view, so they get found late.
The nine mistakes below recur in retail and SME banking programs, whether the bank is a regional lender in the Gulf, a mid-size bank in Europe or a credit union in North America. Each is paired with the practice that prevents it. The table is a triage tool: if a row describes your current plan, that's where your schedule risk sits.
| Mistake | When it usually surfaces | Preventive practice |
|---|---|---|
| Core integration scoped as a connector | System integration testing | API gap analysis against your real core version in month one |
| Card processor engaged late | Feature freeze | Processor statement of work signed in parallel with the platform contract |
| Customers forced to re-enroll | Pilot or launch week | Credential and device migration design |
| Only balances migrated, not customer setup | User acceptance testing | Data inventory of payees, mandates, scheduled payments and alerts |
| Authentication rules treated as configuration | Compliance sign-off | Jurisdiction-by-jurisdiction control mapping |
| Regulator and outsourcing notices filed late | Final approval | Notification calendar owned by compliance |
| App-store setup left to the last sprint | Submission | Developer accounts, app transfer and review dry runs early |
| Accessibility tested at the end | External audit | Accessibility criteria in every sprint's definition of done |
| Big-bang cutover with no rehearsal | Cutover weekend | Two full dress rehearsals and a tested rollback |
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
Mistakes one and two: underestimating core and card-processor integration
Mistake one: treating core banking integration as a pre-built connector. Vendors will truthfully say they've integrated with Temenos, Finastra, Oracle FLEXCUBE, Infosys Finacle, FIS, Fiserv, Jack Henry or Mambu. What matters is whether they've integrated with your version, with your customizations, product parameters and batch windows. A core that's several releases behind, or that exposes real-time balances only through an old middleware layer, can add months of adapter work. Watch for an end-of-day batch that locks accounts for hours; your mobile app has to handle that gracefully.
What to do instead: In the first four to six weeks, run an API gap analysis against a working copy of your actual core, not the vendor's reference environment. List every customer journey (onboarding, transfers, standing orders, loan repayments, statements) and mark each call as available, available with changes, or missing. Price the gaps before the plan is baselined. Core integration and middleware work can run from low six figures to well over $1 million in professional services. Treat that as an indicative 2026 market range that varies widely by contract and region, and get firm quotes from your vendor and integrator.
Mistake two: bringing the card processor in after design is done. Card freeze, PIN view, spending controls, virtual cards and push provisioning to Apple Pay, Google Pay or Samsung Pay all depend on your issuer processor, whether that's Fiserv, FIS, Visa DPS, Thredd, Marqeta, Paymentology or an in-house switch. Processors run their own release calendars and change-request queues. Push provisioning also needs approval from the wallet providers and card networks, and those approvals sit outside your project plan.
What to do instead: Sign the processor's statement of work in parallel with the platform contract. Name a processor-side technical owner and start wallet provisioning applications in the first quarter. If wallet approval isn't confirmed by the program midpoint, take it out of launch scope rather than letting it hold the date.
Mistakes three and four: forcing re-enrollment and migrating only half the customer
Mistake three: making every customer re-enroll. This is the most expensive shortcut in digital banking user migration. Asking customers to download a new app, re-verify their identity and set new credentials spikes contact-center volume, drags down app-store ratings and pushes some customers to a competitor. It also tends to delay launch, because once leadership sees the projected call volume, the date moves.
What to do instead: Design credential migration on purpose. The usual options:
- Hash import: If the new identity provider supports the legacy platform's password hashing algorithm, import the hashes directly.
- Just-in-time migration: On first login, the new platform validates credentials against the legacy store, then rehashes and stores them. Legacy auth stays live for a defined window, often a few months.
- Soft device re-binding: Device binding tokens and biometric keys generally can't move between platforms. Plan a one-step re-bind using a one-time passcode or in-app confirmation, not a full re-onboarding.
Just as important, publish the new app as an update to the existing store listing so customers receive it automatically. More on that below.
Mistake four: migrating balances and history but not the customer's setup. Customers judge a migration by whether their saved payees, scheduled transfers, bill payment mandates, alert preferences, nicknamed accounts and e-statement consents survived. Lose those and the launch feels broken even when every ledger reconciles.
What to do instead: Build a data inventory of every customer-configured object in the legacy channel, map each to the new platform and run reconciliation counts per object type in every mock migration. Pay particular attention to future-dated payments falling due during cutover week.
Mistakes five and six: late authentication design and late regulatory notices
Mistake five: treating authentication rules as a configuration task. Strong customer authentication requirements differ by market, and they shape journeys, not just settings. Under PSD2 in the EU and the UK's equivalent rules, payment initiation generally needs two independent factors plus dynamic linking to amount and payee, with defined exemptions. India's central bank requires additional authentication for many digital payments. Singapore's regulator has pushed banks toward anti-scam controls such as cooling-off periods and stricter device binding. Banks in the Gulf, Southeast Asia, Latin America and Africa face their own central bank circulars. A vendor's default flow built for one region won't match another.
What to do instead: Have compliance produce a control map per jurisdiction before journey design starts. It should cover authentication factors, exemptions, transaction limits, cooling-off rules, consent capture and data residency. Treat each control as a testable requirement with an owner. Your legal and compliance advisers should confirm the interpretation for your licence; this guide can't.
Mistake six: filing regulator and outsourcing notifications late. Many regulators expect notice or a risk assessment before a material technology outsourcing goes live. Examples include the EU's Digital Operational Resilience Act, with its register of ICT third-party arrangements, the EBA outsourcing guidelines, India's IT outsourcing directions and Singapore's outsourcing guidelines. Some regimes expect an independent penetration test or a cloud risk assessment first. If a filing lands a month before launch and the regulator asks questions, the date belongs to the regulator.
What to do instead: In month one, build a notification calendar that works back from launch and allows for the regulator's typical response times. Schedule the external penetration test early enough to fix findings and retest, with a comfortable buffer, often six weeks or more, before go-live.
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
Mistakes seven and eight: skipping app-store and accessibility reviews until the end
Mistake seven: leaving app-store setup to the last sprint. Mobile banking go-live depends on two companies that don't sit on your steering committee. Apple expects apps offering banking and financial services to be published by the regulated legal entity, not the vendor. Organization enrollment requires a D-U-N-S number and verification that can take from days to several weeks. Google Play asks for organization verification and, in some markets, financial-services declarations. If your current app sits under a vendor's developer account, you'll need an app transfer to keep the existing listing, ratings and installed base. Transfers have prerequisites that are easy to miss, so check the current store documentation early.
What to do instead: Confirm ownership of both developer accounts in the first month. Keep the same bundle and package identifiers so the new build arrives as an update. Submit an early build for review well before launch to surface rejections on permissions, privacy labels, data-safety forms or demo-account access. Then plan around review time with phased release rather than against it.
Mistake eight: running accessibility testing at the end. Accessibility is a legal requirement in a growing number of markets. The European Accessibility Act now covers consumer banking services in the EU, and the UK Equality Act, Canadian and Australian disability legislation and US ADA litigation all create exposure. WCAG 2.1 or 2.2 AA is the usual benchmark. A late audit that finds unlabeled buttons, broken screen-reader focus order or inaccessible one-time-passcode screens means rework across dozens of screens.
What to do instead: Write WCAG AA criteria into each sprint's definition of done. Test with VoiceOver and TalkBack every sprint, and commission an independent audit at roughly 70 percent feature completion, not at 100.
Mistake nine: a big-bang cutover nobody has rehearsed
The last mistake turns small problems into a lost weekend. A team that moves every customer at once, never having run the full cutover end to end, will learn on Saturday night that the migration scripts take 14 hours, not six. It'll also find that the legacy channel can't be frozen cleanly and the rollback plan has never been tested.
What to do instead: Run at least two full dress rehearsals with production-volume data, timing every step against the cutover window. Define go/no-go criteria in numbers, for example zero reconciliation variance on balances, a payee migration rate above an agreed threshold and login success above an agreed level in pilot. Then phase the rollout:
- Staff and friendly-customer pilot: a few hundred to a couple of thousand users for several weeks.
- Segmented waves: by region, product or customer cohort, holding the riskiest segments (SME users with multiple signatories, customers with large standing-order books) for later waves.
- Legacy decommission: only after the final wave has been stable for an agreed period.
Hypothetically, say a bank with 400,000 digital users plans a single cutover weekend. Splitting it into a pilot and four waves might add about six weeks to the calendar. But it caps any one failure at a quarter of the base and gives the contact center real call-volume data before the biggest wave. For most banks with several hundred thousand digital users, phased waves are the better choice because one failed big-bang weekend costs more time than the waves add.
Brief the contact center and branches before each wave, with scripts for re-binding devices and finding migrated payees. Frontline staff who can't answer those two questions will generate more escalations than any defect.
A 30-day readiness check to run on your own program
If you're a few months into delivery, you don't need a re-plan. You need to know quickly which of the nine risks are live. Within 30 days, ask your program lead for written answers to these questions:
- Has the API gap analysis been run against our actual core version, and are all the gaps priced?
- Is the card processor's statement of work signed, and is wallet provisioning approval on a dated path?
- Which credential migration approach have we chosen, and what share of customers must do anything on day one?
- Do we have an object-by-object inventory of customer setup data with reconciliation targets?
- Is there a signed-off authentication and consent control map for every market we operate in?
- What's the regulator notification and penetration test calendar, worked back from launch?
- Do we own both developer accounts, and has a build already passed store review?
- When is the independent accessibility audit, and what feature completion will it test?
- How many cutover dress rehearsals are scheduled, and what are the numeric go/no-go criteria?
Apply a simple rule. Any question answered "not yet" more than three months before launch is a risk to manage. Any "not yet" inside three months is a reason to rescope or move the date now, while moving it is still your decision rather than the regulator's or the app store's.