At Money20/20 Europe, Nikhil Sengupta, Commercial Director at 10x Banking, sat down with Ohad Kotler, CEO and Co-Founder of Tweezr, to discuss a problem that sits upstream of every migration conversation: before a bank can safely move off a legacy core, it needs to understand what that core actually does.
Summary
-
Legacy migrations tend to fail for epistemic reasons, not technical ones. The knowledge of how legacy systems work is often lost: no documentation, no institutional memory, no map of what was built or why.
-
AI can now extract business logic from legacy code. Not just what a system does, but why it was built that way, and whether those original purposes still apply.
-
Deterministic AI is key in regulated environments. Probabilistic models produce their best guess. In banking, the ground-truth layer of a migration requires certainty, not probability.
-
Tweezr maps the legacy system. 10x Banking is the target modern core. Together, they offer a specific, AI-assisted route from legacy estate to 4th-generation, cloud-native infrastructure.
Watch the interview
Why legacy comprehension is the real barrier to core banking migration
Most analysis of core banking migration focuses on the destination: which platform, which architecture, which migration approach. The harder and less-discussed problem is the starting point.
Ohad Kotler framed it directly:
"Legacy banking systems are incredibly layered – almost like a city built over decades, with streets, buildings, and infrastructure added piece by piece. Trying to move that entire city into a new environment without fully understanding how every component works or why it exists is enormously difficult. That lack of visibility is why so many migration projects have taken years, cost hundreds of millions, or failed entirely."
Three compounding problems sit behind that lack of visibility. The people who built the original systems have retired or left. Documentation was incomplete when it was written and has not been maintained since. The underlying infrastructure is so unfamiliar that very few engineers today can safely modify it.
The result is that scope estimation at the start of a migration project is, in large part, guesswork. Validation during the project is difficult. The risk is not primarily technical; it is epistemic. Banks do not know what they do not know about their own systems.
Nikhil added:
"Migration used to be viewed as inherently high risk. That perceived risk was so significant that many institutions chose to stay with familiar legacy systems rather than attempt a move."
That calculation has changed. Technical debt compounds. Operational costs on legacy infrastructure remain high. Competing with institutions that have already modernised becomes harder each year. As Nikhil noted, standing still is no longer a safe option.
How does AI extract business logic from legacy banking systems?
This is an area where AI is doing something that was not previously possible at speed.
AI applied to legacy codebases can recover not just technical structure (which function calls which, which databases are triggered under which conditions, etc.), but business intent. Why a component was built the way it was. What business problem it was designed to solve. Whether that problem still exists.
Ohad explained:
"AI helps answer critical questions: why was this function built? What business problem was it solving? Do we still need it today? That matters enormously because migration is no longer just about replicating the past."
That last point is significant. A migration that faithfully reproduces all legacy functionality, including rules no one remembers designing and purposes no one has reviewed in years, is not a modernisation. It is a transliteration. AI-assisted comprehension makes it possible to challenge the legacy, not just move it.
This matters even more as banks move toward agentic banking, where AI systems take on more reasoning and decision-making. Validating those AI decisions against historical logic requires first understanding what that historical logic was. Without that baseline, comparison is incredibly difficult.
What is deterministic AI, and why does it matter in a regulated migration?
Large language models are probabilistic: they produce the most probable answer to a question, not a guaranteed one. For many use cases, that is acceptable. For a bank trying to establish ground truth about what a legacy system does, it is not. In regulated banking, people's money, regulatory compliance and operational resilience are all on the line.
The answer is a deterministic layer built beneath the language model: a structured, verifiable map of what the legacy system actually does, derived from direct codebase analysis rather than inference. Ohad Kotler, whose company Tweezr has built exactly this capability, described what it covers:
"Deterministic analysis focuses on facts that can be verified with certainty: which function calls another function, which database is triggered, under what conditions a process executes. This creates a ground truth layer. AI can then explain the system using that verified data rather than making assumptions."
The practical outcome: the LLM's strength – comprehension, explanation, natural language – is applied on top of a layer that has already established certainty about system behaviour. The model is not guessing about what the code does. It is explaining something that has already been verified. In a regulated migration, that is the difference between a defensible process and an assumption-based one.
Phased migration and coexistence
The shift away from big bang migration toward wave-based, phased approaches is now well established. We cover the methodology in detail in why a phased approach is key to a successful core banking migration, and the practical steps to take in how to execute a core banking migration.
The point worth adding here is what AI-assisted deterministic analysis changes about phased migration specifically: it makes wave design precise.
Previously, identifying which parts of a legacy system could move together in a first wave required large consulting teams and a significant amount of informed guesswork. The question – what is both technically feasible and business-meaningful to migrate as a unit – was hard to answer with confidence. With a deterministic map of the legacy system in place, the same question can be answered from data. Each wave also improves the next: learning from one transition feeds directly into planning the following one, so the process becomes faster and more accurate over time.
Coexistence of old and new systems runs through this methodology, as Nikhil described:
"In most transformation programmes, a coexistence phase is a critical component of a safe migration strategy. Rather than switching everything off overnight, old and new platforms run in parallel. It enables gradual migration while continuously validating outcomes – reducing operational risk and allowing banks to realise benefits earlier."
How do 10x Banking and Tweezr work together?
The partnership divides cleanly: Tweezr maps the legacy system; the 10x Banking Platform is the target modern core. Nikhil elaborated:
"Tweezr helps banks move from legacy complexity to clarity, while 10x Banking provides the modern end state: a cloud-native, API-first core designed for future innovation. Together, we help banks move faster and more safely."
In practice, Tweezr builds a deterministic map of the old system and aligns it with the 10x Banking Platform's architecture – identifying which processes, data, and dependencies move in each wave and when. Migration waves can then be designed against two constraints simultaneously: what is technically feasible to move together, and what is business-meaningful to move together. All grounded in precise data.
The value on the other side: real-time processing, product flexibility, AI-native infrastructure, and the removal of the operational cost and technical debt that come with maintaining a legacy estate.
Continue the conversation with Nikhil
FAQ
Why do core banking migrations fail?
The primary cause is the loss of institutional knowledge about legacy systems, compounded by absent or outdated documentation. Without a clear map of the existing system, scope estimation is guesswork and validation is difficult throughout the project. Secondary causes include big bang cutover approaches, underestimated integration complexity, and insufficient testing.
How does AI help banks understand undocumented legacy systems?
AI can analyse legacy codebases and extract both technical structure (which function calls what, which databases are accessed under which conditions) and business logic (what purpose each component was designed to serve). This matters because documentation was often never adequate, or was written and then lost. AI can reconstruct a usable map of the system even where human expertise is no longer available.
What is deterministic AI and why does it matter for regulated banking migrations?
Deterministic AI produces the same output for the same input, every time. Large language models are probabilistic: they produce the most probable answer, not a guaranteed one. In regulated banking, probabilistic variance is not acceptable in the ground-truth layer of a migration, where an incorrect output can affect customers' funds, regulatory standing, or audit integrity. Deterministic AI establishes certainty about what a system does before any language model is asked to explain or act on it.
What is a phased or wave-based migration in core banking?
Phased migration moves specific customer segments, products, or portfolios in defined waves rather than switching from old to new in a single cutover. Each wave realises value progressively, with the ability to pause, validate, and correct before the next begins. Regulators increasingly favour this approach because it reduces systemic risk. See our detailed guide: Why a phased approach is key to a successful core banking migration.
How do 10x Banking and Tweezr work together?
Tweezr builds a deterministic map of a bank's legacy system and aligns it with the 10x Banking Platform's architecture, identifying which processes, data, and dependencies should move in each wave and when. 10x Banking is the target modern core: a 4th-generation, cloud-native, API-first platform. Tweezr enables the journey; 10x is the destination.
What does the 10x Banking Platform offer as a target core?
The 10x Banking Platform is a 4th-generation, cloud-native, API-first core banking system supporting real-time processing, multi-tenancy, and AI-native integration. It was built cloud-native from the ground up. Clients include Chase UK, Westpac, and Old Mutual.
.jpg)
