Fragmented across systems, inconsistently defined and difficult to operationalize, the vast volumes of data held by banks have largely remained underutilized. Recent advances in artificial intelligence are changing this – in addition to enabling large-scale analysis, AI is now mature enough to power conversational banking and increasingly sophisticated self-service experiences.
Conversational banking – the application of conversational interfaces such as chatbots and voice assistants to help customers engage with their banking services – has the potential to truly transform customer experience in the digital banking age, helping customers engage more naturally, through speech and text to solve problems and access information without complex website and mobile app navigation.
However, unlocking the potential of such services at scale and end-to-end requires more than deploying AI models. It demands a foundational shift in how data is structured and understood. Reliant as it is on pattern recognition and probabilistic outputs, AI alone cannot guarantee the precision and consistency required in a heavily regulated banking environment.
This is where semantic layers and ontologies, such as the Financial Industry Business Ontology (FIBO), play a critical role. By providing a shared, machine-readable understanding of financial concepts, they enable banks to connect data and AI in a way that is both accurate and scalable.
In this article, Capco Poland’s Andrzej Worona, Head of AI & Data and Laura Zusin-Kaczmarek, Data Practice Lead share their perspectives on FIBO and its importance for conversational banking.
Banks and other financial services institutions often assume that with vast amounts of high-quality data, powerful transformer models and scalable platforms, AI will deliver consistent and reliable results. In practice, many solutions still struggle with complex financial concepts. Can you share an example of where AI might fail to correctly interpret even well-structured banking data?
Andrzej Worona: Yes. A useful example comes from corporate banking, where a client is moving funds between subsidiaries across jurisdictions. This typically involves large, repeated cross-border transfers across multiple accounts within a short timeframe. From a purely pattern-based perspective, an AI model may flag this as suspicious, as it resembles common money laundering typologies.
In reality, this is often standard treasury activity (such as cash pooling or liquidity optimization) between legally related and fully disclosed entities. Without an understanding of these relationships and the underlying business context, AI is likely to generate false positives and potentially block legitimate transactions.
Even well-structured data is not sufficient without a shared, precise understanding of financial concepts. This is why consistent definitions and semantic alignment across systems are critical for AI to deliver reliable outcomes.
Laura Zusin-Kaczmarek: A similar challenge arises in conversational banking. A seemingly simple request, such as a customer asking for their balance, can be ambiguous. Customers do not hold balances in isolation. They hold one or more accounts, may have access to corporate accounts, and different parts of the bank may define ‘account’ and ‘balance’ differently – for instance, ledger versus loan accounts.
While traditional digital channels handle this through structured workflows, conversational AI must interpret intent in real time. In this context, ambiguity cannot be resolved through pattern recognition or probabilistic guesses and there is no room for misinterpretation and hallucination.
These examples highlight a broader point: even well-structured data is not sufficient without a shared, precise understanding of financial concepts. This is why consistent definitions and semantic alignment across systems are critical for AI to deliver reliable outcomes.
What is FIBO? And how can banks practically introduce an ontology layer alongside their existing data and AI capabilities?
Laura Zusin-Kaczmarek: The Financial Industry Business Ontology is a global standard developed and maintained by the Enterprise Data Management Council (EDMC). It provides a formal, machine-readable representation of financial concepts, covering everything from instruments and legal entities to transactions and relationships.
Each concept in FIBO is defined in a precise and unambiguous way, making it interpretable by both humans and machines. Importantly, these definitions have been developed and validated over time by EDMC members, including major banks and market infrastructure providers. As a result, FIBO reflects a broad industry consensus and aligns closely with existing data models and messaging standards.
FIBO is freely available to financial institutions, but it is not a plug-and-play solution. Real value is realized when it is integrated into the bank’s existing data and AI landscape. In practice, this means mapping internal data models and applications to FIBO concepts, creating a semantic layer that sits across systems and provides a consistent understanding of key business terms.
Andrzej Worona: A pragmatic approach is to start small and scale incrementally. FIBO is organized into domains, allowing banks to focus on a specific use case (e.g. client data, financial instruments, regulatory reporting) and progressively expand coverage. This avoids large-scale transformation risk while delivering early, tangible benefits.
Many AI strategies focus on better models, more data and stronger infrastructure, often overlooking the semantic layer. In this context, what does true AI maturity look like in banking today? Could you share an example of a seamless user-facing AI interaction?
Laura Zusin-Kaczmarek: Let’s look specifically at conversational banking. A truly mature AI capability would allow a client to interact with the bank in natural language and complete complex tasks like checking balances across multiple accounts, initiating payments or managing liquidity, just as reliably as through traditional digital channels
A pragmatic approach is to start small and scale incrementally. This avoids large-scale transformation risk while delivering early, tangible benefits.
Andrzej Worona: Open banking provides another compelling example. In an open ecosystem, data and services are shared across institutions via APIs, often in real time. While standards define the structure of the data, they do not fully resolve differences in meaning. For instance, the definition of an ‘account’, ‘available balance’ or ‘transaction status’ can vary subtly between institutions.
Laura Zusin-Kaczmarek: An ontology-based approach allows participants to align on meaning, not just format, enabling AI to operate reliably across organizational boundaries. In this context, true AI maturity means moving beyond isolated intelligence towards interconnected, semantically grounded systems, where AI can act with both flexibility and precision, whether within a single bank or across an open banking network.
Looking ahead, what risks do banks face if they continue scaling AI without a shared semantic foundation?
Laura Zusin-Kaczmarek: Without a shared semantic foundation, banks risk scaling inconsistency rather than intelligence. AI systems may appear to perform well in isolated use cases, but as they are deployed more broadly across products, functions and ecosystems, differences in how data is defined and interpreted will begin to surface.
At a technical level, this leads to misinterpretation of data and conflicting outputs between models and systems. The same concept, such as ‘exposure’, ‘balance’ or ‘customer’ may be understood differently across business lines, resulting in inconsistent insights and unreliable decision-making.
Andrzej Worona: As these inconsistencies accumulate, the risks become more material. From a regulatory perspective, banks may struggle to demonstrate how decisions are made, particularly in areas such as credit, financial crime or reporting. Lack of traceability and explainability can quickly translate into compliance issues.
There are also clear reputational implications. In customer-facing scenarios, such as conversational banking, misinterpretation of seemingly simple requests can erode trust.
Cybersecurity risks also increase in a fragmented semantic environment. Ambiguities in how data is classified, accessed or interpreted can create gaps that are harder to detect and govern, particularly as AI systems interact across organizational boundaries in open ecosystems.
Ultimately, the strategic risk is that banks scale AI capabilities without achieving reliability, with models requiring constant oversight and undermining the very efficiencies AI is meant to deliver.
Andrzej Worona and Laura Zusin-Kaczmarek will be discussing this topic in more detail at the Data & AI Warsaw Tech Summit on 21st April 2026 (11:20 – 12:00).
To find out more about the future of conversation banking and FIBO, register to attend: https://dataiwarsaw.tech/agenda/
References
1 FIBO
Related content
Contact us
To find out more about working with Capco and how we can help you overcome any potential challenges, contact our experts via the form below.