The hardest part of a core banking migration is rarely the new technology. Three respondents explain why freezing features, switching over in stages and planning the reconciliation early matter more.

Core banking replacement is among the riskiest projects a bank or lender can take on. The core holds every account, balance and transaction, so a failed cutover locks customers out of their money. Still, many banks are moving off systems that are decades old, often to the cloud. As a result, the lessons from those who have done it, or tried, are worth more than vendor promises. For how many fintechs avoid running a core at all, see our explainer on banking-as-a-service.

We asked one question: what was the hardest lesson from replacing or modernising your core banking system, and what would you do differently? We wanted CTOs, CIOs and transformation leads at banks, building societies and credit unions, plus core banking vendors. Timelines and costs were welcome, and we asked for first-hand experience over vendor pitches. Three people answered in this round, and we used all three. One led a migration at a regulated lender, so we’ve set all three answers against public cases.

Why core banking migrations go wrong

TSB remains the warning case. Its April 2018 migration to a new platform disrupted branch, phone, online and mobile banking, and services did not return to normal until December 2018. The FCA and PRA fined TSB £48.65 million in December 2022 for failings in governance and in managing its outsourced IT, the FCA said. TSB has also paid £32.7 million in redress to customers.

Banks now tend to replace core banking systems in stages instead. Most are choosing phased replacement, with old and new cores running side by side, and big bang replacements carry higher risk, Oliver Wyman notes. Large moves can still be quick. Commonwealth Bank says it moved its SAP core banking platform, covering 16 million accounts, to the cloud in 18 months, according to its newsroom. It ran five dress rehearsals instead of the three it planned.

Freeze the logic, then switch in stages

Marat Shigapov is founder and CIO of CloudCube IT in Dubai, and previously ran technology at a regulated online lending group.

“My experience is from fintech lending, not a traditional bank. For three years I was responsible for technology at an online lending group under central bank regulation. I think the biggest mistake in replacing a core system isn’t choosing the wrong technology. It’s not fully understanding what the business needs, and I’ve seen both good and bad examples of that.

We had a credit scoring system running inside our analytics platform. As loan applications grew, we started having performance problems, so we moved scoring to a dedicated platform with a workflow engine. Some checks could run at the same time, and if a customer failed an important check, we didn’t need to finish the whole process.

It took six months and around $600,000 in the first year, including licences, implementation and model migration, using our own team and an external contractor. One thing I insisted on was not changing the business logic during the migration. No new features, no new requirements. Just move what we already had and make sure it worked.

We ran both systems in parallel and compared the results. Once they matched more than 97% of the time, we started sending real traffic to the new system: first 1%, then 5%, and more. There was no fixed schedule. Our Chief Risk Officer decided each step based on approval rates and risk indicators. Median scoring time for one stage fell from 15 seconds to 0.8 seconds, and the full automatic decision from 132 seconds to 72. We finished on time and within budget.

I’ve also had the opposite experience. On another project, we built our own system because the business wanted flexibility. We spent money on development, hired people, and after launch found that nobody needed most of that flexibility. A standard product would have been cheaper, maybe several times cheaper, and easier to support.

So now I ask one question first: are we really going to use these custom features, or do we just like having the option? And for the migration itself, my rule is simple. Replace the old system first, then improve it. Trying to do both at once creates problems you don’t need.”

Marat Shigapov, Founder and CIO, CloudCube IT

His answer is the closest to what we asked for: a real migration with a timeline, a budget and results. His method also matches the phased approach Oliver Wyman describes. He froze the logic, ran both systems in parallel and moved traffic in small steps, with a risk officer deciding each one. That is the opposite of the single core banking cutover that failed at TSB.

However, his project moved a credit scoring platform, not a full core banking system, and he says so himself. The lending group is unnamed, so we could not verify his costs or timings. We also found no website for CloudCube IT. Still, his custom-build story is a useful warning for any core banking buyer tempted to pay for flexibility nobody uses.

Cut old processes before you copy them

Abhishek Pareek is founder and director of Coders.dev, a marketplace for outsourced software developers.

“The most important lesson in core banking modernisation is that process debt can matter more than technology debt. On technology projects for billing and banking systems, I’ve seen projects stall because the organisation won’t accept anything less than full feature parity with a 30-year-old legacy system. Teams end up rebuilding hidden, inefficient processes, so the new system is inefficient from day one. You don’t get digital transformation by paying a premium to run outdated, undocumented processes on a more expensive cloud platform. If I were starting a core replacement today, I’d begin with a thorough audit of existing processes, to decide which ones to drop rather than migrate.

On risk, the main mistake is a ‘big bang’ switchover. These projects often run for two to three years, but I’ve seen teams try to squeeze them into one because of quarterly reporting or board pressure. That usually leads to a big budget overrun, as unexpected dependencies surface late. I now favour a gradual, modular rollout, where the new core runs some products while the legacy system keeps handling others. That keeps risk under control and lets the engineering team stay disciplined.”

Abhishek Pareek, Founder and Director, Coders.dev

His point on feature parity fits with Shigapov’s, from the other side. Shigapov froze existing logic to keep the move safe, while Pareek warns against copying every old process into the new core banking system. Together, they suggest a sequence: decide what to drop before the move, then migrate without adding anything new. His modular rollout also matches the dual-core approach most banks now favour.

However, his original answer said these projects normally last 24 to 36 months. We found no source for that range, and Commonwealth Bank’s 18-month move shows timelines vary widely, so we’ve softened it. Coders.dev supplies developers rather than running core banking projects, so his answer reads as advice from the supplier side.

Plan the reconciliation early

Ankit Sarawagi curates CFO Matrix, a finance resource, and focused on the finance side of a migration.

“I haven’t managed a complete core banking replacement, but I can share a finance view. Teams often plan their processes around the new system and forget to plan how they’ll reconcile their data. Every balance needs reconciling, and if they don’t match, you have to find out why. That normally takes longer than expected.

If I were advising on a project, I’d make sure finance is involved early. You’ll want to know which balances need reconciling, who will sign them off and what variance is acceptable. And make sure your data is clean in the old system first.”

Ankit Sarawagi, Curator, CFO Matrix

His point is easy to overlook. A core banking migration can succeed technically and still fail if balances do not match afterwards. Agreeing an acceptable variance and a named sign-off before cutover gives a core banking project a clear test of success. Also, his advice to clean data in the old system first fits Pareek’s audit. He says plainly that he has not run a core replacement, so this is a finance view rather than first-hand experience.

What works for core banking replacement

The three answers and the public cases point to a clear order of work. First, audit existing processes and drop what the business no longer needs, before any data moves. Next, freeze business logic during the migration, so the new system only has to match the old one. Then run both systems in parallel, move traffic in small steps and let a named risk owner approve each one. Finally, involve finance early, agree which balances to reconcile and set an acceptable variance. Rehearse the cutover more than once, as Commonwealth Bank did. Firms seeking new charters face the same build-or-buy questions, as our report on Modern Treasury’s trust bank application shows.

What to watch next

Cloud moves will keep testing core banking teams. Watch for more banks reporting phased go-lives, and whether regulators press harder on operational resilience during cutovers. Also watch for published costs, since few banks share what a core banking replacement really costs.

Above all, core banking projects succeed when teams change one thing at a time. Move the system first, prove it matches, and only then start improving it.

FintechBits covers payments, banking and fintech developments for readers in the US and UK. This article is for information only and is not financial advice. Views expressed are those of the FintechBits editorial team.