EXCLUSIVE: "Raising a Bank’s IQ" - Adrian Congiu, Mambu in 'The Fintech Magazine'

Where and how a bank’s data is stored has a big impact on how smart its AI can ultimately become. Mambu Insights and its underlying Mambu Data Lake work together to create intelligence at the core

In recent years, the banking industry’s artificial intelligence (AI) conversation has focussed on large language models, copilots and, more recently, agentic AI. Yet, beneath the excitement lies a more fundamental challenge. Before banks can rely on AI to transform customer experience or automate decision-making, they must first solve a problem that predates generative AI by decades: making sense of their own data.

Financial institutions generate vast quantities of customer, transaction and operational information every day, but much of it remains trapped inside systems designed to process transactions rather than generate insight. AI may promise personalised financial advice, faster lending decisions and autonomous back-office operations, but none of those ambitions can be realised if the underlying data is fragmented, inconsistent or difficult to access. That is why the discussion is beginning to shift.

 Instead of asking what AI models are capable of, banks are increasingly asking whether their core banking platforms are capable of supporting them. The result is the emergence of what vendors are calling the intelligent core – a new generation of Cloud-native banking platforms designed not simply to process transactions but to surface trusted operational data that can power analytics, automation and AI at scale.

Among the companies championing that vision is Mambu. Earlier this year, it launched Mambu Insights, which draws information from the Mambu Data Lake to create a unified reporting, analytics and an AI foundation for its Cloud-native core banking platform.

“Our vision is that banks move from static, backwards-looking core banking systems to something that’s much more predictive, more personalised and able to react in real time to events happening on a customer’s account,” says Adrian Congiu, VP, Head of Product Management at Mambu.

“Banks should be able to leverage the data they already have about their customers to deliver more personalised services in the moments that matter.”
It is a subtle but significant distinction. AI may be grabbing the headlines, but data is becoming the competitive battleground, because intelligent agents are expected to complete tasks, orchestrate workflows and make recommendations, based on live banking data. Whether helping a customer complete a mortgage application, monitoring a loan portfolio for emerging risk or supporting product teams designing new financial products, AI agents are only as effective as the information they can access.

This is where data architecture suddenly becomes strategic. For a long time, enterprise data warehouses formed the backbone of banking analytics, storing cleansed and structured information optimised for reporting. They remain invaluable for regulatory reporting and business intelligence, but AI has changed the equation.

Machine learning models increasingly need access to real-time operational data alongside unstructured documents, communications and behavioural information that traditional warehouse architectures were never designed to accommodate. Data lakes emerged to address those limitations by storing information in its native format, using scalable object storage rather than forcing it into predefined schema.

That flexibility makes them ideally suited to AI workloads, although poorly governed implementations have often earned the less flattering label of ‘data swamps’. More recently, hybrid ‘lakehouse’ architectures have sought to combine the openness and scalability of data lakes with the governance, performance and query capabilities traditionally associated with enterprise data warehouses.

“The lakehouse model combines the best of both worlds,” says Congiu. “It delivers the storage economics of a data lake with the structure and queryability of a warehouse.” Importantly, Mambu does not position its lake offering as a replacement for a bank’s existing enterprise warehouse or business intelligence environment. Instead, it becomes a curated operational layer that feeds downstream analytics platforms, AI models and reporting tools, allowing institutions to preserve existing investments while improving the quality and timeliness of the data flowing into them.

The Mambu Data Lake sits at the heart of Mambu Insights. It captures information from the core banking platform using change data capture (CDC), with minimal impact on the core database’s performance. Instead of exposing raw operational tables, the platform organises information into three semantic layers. A bronze layer captures raw banking events in near-real time. A silver layer cleanses, enriches and standardises that information. Finally, a gold layer presents curated, analytics-ready business objects aligned with Mambu’s APIs, allowing reporting tools, AI models and enterprise data platforms to consume trusted banking concepts rather than complex database structures.

That architectural approach reduces one of the largest hidden costs of AI adoption. 

“Access to clean, normalised data is one of the biggest prerequisites for delivering an AI strategy,” says Congiu. “That layer of normalised banking data – transaction data, balances and customer information – is super-powerful for the agentic era. It means there’s a lot less data pipeline work that banks need to build because many of the data products are already available and ready to consume.”

The platform also reflects a broader shift in enterprise architecture. Rather than tightly coupling storage, analytics and user access, Congiu argues that banks should treat them as separate layers that are capable of evolving independently. Open data formats and API-aligned data products allow institutions to change analytics platforms, adopt new AI models or introduce new retrieval technologies without continually rebuilding the underlying infrastructure. As AI evolves, he argues, the organisations that avoid proprietary lock-in today will be best positioned to adapt tomorrow.

The practical implications extend well beyond reporting. The Mambu Insights gold layer brings together transactional, behavioural and product data into a unified view of each customer’s relationship with the bank. Consider a retail customer with a current account receiving a monthly salary, a savings account and a personal loan. The platform can surface repayment history, savings behaviour, spending patterns, income regularity and product tenure as a single business view. An AI copilot could use that context to recommend that a customer service advisor discusses a repayment holiday or temporary overdraft before financial stress becomes visible.

A more autonomous AI agent might trigger an early pre-arrears intervention automatically, while specialist third-party analytics platforms apply their own credit risk or customer lifetime value models directly to the same dataset. That flexibility reflects Mambu’s composable philosophy. Rather than prescribing a single analytics stack, the platform integrates with various tools such as Databricks and Snowflake, while supporting other popular data stacks, allowing banks to combine trusted operational data with their preferred business intelligence and AI tools.

The longer-term ambition, however, extends beyond insight dashboards.

“We’re seeing less of a query-based interaction with data and much more natural-language interaction,” says Congiu. Instead of writing SQL (structured query language) queries or navigating increasingly complex reporting interfaces, users will interrogate operational banking data conversationally, asking questions about customers, products or portfolio performance in everyday language.

Congiu describes the shift as ‘less UI, more AI-led interactions’. Product teams will be able to configure banking products through natural-language instructions rather than navigating complex administration screens. Tasks that once required specialist platform expertise become accessible to product managers or business analysts just by describing what they want the platform to build. Voice interfaces could be a natural next step. 

Congiu says the underlying technology already exists, but cautions that regulated environments require robust confirmation, audit and approval mechanisms before voice-driven configuration can become mainstream. The organisational implications may ultimately prove as significant as the technological ones. As AI becomes the primary interface to increasingly complex banking platforms, routine configuration work shifts from specialist IT teams towards business users.

Rather than executing every change themselves, technology teams become architects and governors – setting guardrails, validating AI-generated outputs, maintaining data quality and ensuring compliance while product teams work more directly with intelligent systems. Governance, however, remains non-negotiable. While Mambu sees intelligent agents taking on an increasing share of operational activity, Congiu emphasises that every automated action must remain observable, traceable and auditable, with a human in the loop for decisions that carry regulatory or customer impact.

That may define banking’s next AI chapter. The winners are unlikely to be those deploying the most AI, but those building the most trustworthy foundations beneath it. The first wave of banking AI demonstrated what large language models could do.

The next will be defined by something altogether more fundamental: whether institutions have built the intelligent data platforms capable of turning AI’s promise into practical, transparent and trusted outcomes.