top of page
Immagine del redattoreAndrea Viliotti

GenAI in Banking

Aggiornamento: 6 giorni fa

Generative Artificial Intelligence (GenAI) is emerging as a powerful tool to transform the financial services sector. Recent research conducted by Thomas Kaiser (CEO and Co-Founder of Kodex AI), Boon-Hiong Chan (Industry Applied Innovation Lead and Head APAC Market and Technology Advocacy at Deutsche Bank), and Delane Zahoruiko (Founders Associate at Kodex AI) highlights how GenAI can be used to improve regulatory compliance, optimize customer interaction, and manage risks more efficiently, paving the way for new levels of productivity and innovation.

However, to fully leverage the potential of GenAI, institutions must address several challenges, including ensuring the quality of systems and cybersecurity, while also guaranteeing a gradual adoption of new technologies.

GenAI in Banking
GenAI in Banking

A Gradual Approach to Adopting GenAI in Banking

An effective adoption of GenAI in the banking sector requires an incremental and structured strategy, starting with basic applications and moving towards more complex use cases. The research suggests a three-phase approach to building a GenAI use case portfolio, allowing institutions to progressively gain confidence in the technology, mitigate risks, and achieve tangible benefits at each stage.


  1. Language Analysis Capabilities: The first phase focuses on leveraging GenAI's basic language analysis capabilities to perform tasks such as text synthesis, processing customer emails, and drafting standardized content. These features enable the organization to improve service efficiency and quality by managing large volumes of text and unstructured data. This phase not only lays the foundation for system capability development but also allows for GenAI to be tailored to the specific needs of the financial domain.

  2. Chat-to-Agent: In the second phase, the goal is to transform GenAI into a tool that goes beyond text analysis, enabling specific commands to be executed based on user requests. For example, an executive agent can receive a natural language query, translate it into code (e.g., Python), and use AI models to analyze large datasets and return understandable results. An experiment conducted with the MILA project demonstrated how a chat-to-agent solution allowed non-technical users to obtain detailed analyses of relationships and patterns in data, using visualizations to facilitate understanding. This phase allows for a high degree of autonomy in analysis, while still ensuring human supervision and control for critical results.

  3. Chat-to-Execution: The third phase represents the evolution towards autonomous capabilities, where GenAI not only executes commands but also makes autonomous decisions with contextual awareness. This level of development enables the system to operate with a high degree of independence, managing complex decision-making and operational processes. For example, a chat-to-execution system can autonomously decide which approach to take in responding to a specific request, based on a combination of reinforcement learning and memory of past interactions. This capability not only allows for the execution of repetitive tasks but also for adaptation and improvement over time, offering increasingly targeted solutions.


The transition from simple language processing applications to fully autonomous solutions requires not only advanced technological infrastructures but also a constant commitment in terms of governance, risk management, and continuous training. The creation of controlled testing environments (AI sandboxes), the development of fair use policies, and the active involvement of industry experts are key aspects for successful adoption.


Benefits for the Banking Sector

The adoption of GenAI in banking offers several significant benefits, not only in terms of operational efficiency but also in the ability to address complex challenges such as risk management and regulatory compliance. One of the main advantages is GenAI's ability to improve decision quality through the automation of complex analyses. The technology allows the integration of significant volumes of data from various sources and provides real-time analysis, promoting a deeper understanding of market trends and potential risk areas.

Moreover, the use of models such as Retrieval-Augmented Generation (RAG) helps improve the accuracy of responses generated by GenAI by tapping into external and verified data. This is particularly useful to ensure that responses are always based on up-to-date and relevant information, a crucial aspect in risk management and regulatory compliance, especially in scenarios requiring high precision and reliability.


Another significant advantage relates to the democratization of access to advanced analytics. Tools like those developed in the MILA project have shown how GenAI can enable non-technical users to perform advanced data analyses, reducing the reliance on data science specialists. This capability was demonstrated by experiments where GenAI reduced analysis times by data engineers from several hours to just a few minutes, making the decision-making process faster and more accessible.


The use of techniques like Parameter Efficient Fine Tuning (PEFT) and Low Rank Adaptation (LoRA) also helps reduce training costs and improve model customization, making them more suitable for integration into existing infrastructures without the need for excessive computational resources. This optimization not only supports cost reduction but also improves the adaptability of models to the specific needs of each banking organization.

Additionally, the use of synthetic data makes it possible to train models in the absence of real data, addressing issues related to privacy and data availability. This approach ensures that models can operate on representative and diversified datasets without compromising customer privacy.


Improving customer engagement is another crucial aspect. GenAI enables the development of more personalized and timely interactions, based on a deeper understanding of customer needs and automated request management. This not only increases customer satisfaction but also improves the efficiency of customer service operations, reducing response times and enhancing service quality.


Finally, the adoption of GenAI can increase the scalability of operations. In a constantly evolving context like financial services, the ability to quickly scale processes and infrastructures is essential. GenAI systems, thanks to their flexibility, can be adapted to handle an increasing number of requests and processes without compromising the effectiveness or accuracy of operations. This is particularly advantageous during periods of high demand, where it is essential to maintain high service standards without experiencing slowdowns.


Quality and Benchmarking

To ensure that a GenAI system delivers adequate performance and meets the required standards in the financial sector, it is essential to establish quality measurements through accurate benchmarking. The quality of GenAI depends not only on the model architecture but also on the training data and enhancement tools like RAG and PEFT. The use of benchmarks such as GLUE, SuperGLUE, and MMLU is essential to evaluate the model's ability to understand and process natural language in general contexts. However, the financial sector presents specific challenges that require more targeted measurements.


In the banking sector, GenAI effectiveness is often evaluated through specialized financial benchmarks such as FinanceBench, FinQA, and FNS (Financial Narrative Summarisation). FinanceBench assesses the model's ability to accurately process and interpret financial data for market analysis, risk assessment, and compliance reporting. FinQA, on the other hand, focuses on the system's ability to answer questions based on financial contexts, analyzing structured data such as financial reports and earnings calls. FNS evaluates a model's ability to summarize complex financial narratives from dense datasets, such as earnings reports or annual reviews, thus providing a measure of effectiveness in generating key insights automatically.


Besides benchmarks and optimization methods, other architectural and process factors play a crucial role in determining the quality of a GenAI system. Among these, data management is critical. The use of preprocessing techniques such as chunking and parsing, in addition to content filters, ensures that data is managed appropriately before being processed by the model.


Finally, the issue of explainability is equally fundamental. Implementing transparency systems, such as source attribution in responses and integrating human verification systems (Human-in-the-Loop), helps ensure that the decisions made by the models are traceable and understandable, thereby building the trust needed for GenAI adoption in highly regulated sectors like banking.


Challenges and Risks

The implementation of GenAI involves numerous challenges and risks that must be addressed to ensure the long-term success of the technology within the financial sector. One of the main issues is model drift. This phenomenon occurs when a model's performance begins to degrade due to the difference between the data used for training and the data the model encounters in real-world scenarios. Changes in customer behaviors or regulations can lead to a significant divergence between the operational context and the data originally used to train the model. To mitigate this risk, it is essential to implement continuous model performance monitoring through metrics such as prediction accuracy and error rate, as well as regular retraining on updated datasets to keep the model aligned with reality.


Another significant risk is model hallucination, which refers to the generation of plausible but incorrect or unverified responses. This problem is inherent to the nature of GenAI but can be mitigated with specific techniques. For instance, using Retrieval-Augmented Generation (RAG) techniques, which allow the model to draw from external data sources to verify and confirm information, reduces the likelihood of hallucinations. Moreover, human oversight through the integration of Human-in-the-Loop (HITL) systems allows monitoring of model responses, especially for high-risk decisions, thus ensuring that responses are accurate and relevant.


Feedback loop degradation is another risk that occurs when a GenAI system is overly exposed to user feedback without proper quality filters. In such cases, the system may learn undesirable behaviors, worsening the quality of responses over time. To address this issue, it is essential to implement feedback filtering mechanisms that allow for evaluating the quality of user-provided data before it is used to influence model learning.


In addition to these specific risks, there are also dependency risks, such as reliance on specific infrastructures or external providers. To mitigate such risks, it is important to adopt modular and interoperable architectures that allow easy migration to alternative models or platforms, avoiding technological lock-in situations.


Finally, the financial sector must address cybersecurity risks, especially when using GenAI systems that may interact with sensitive data. Adopting advanced security measures, such as protection against data poisoning or prompt injection attacks, is essential to ensure the resilience and reliability of the system.


Recommendations for the Financial Industry

To foster effective adoption of GenAI in the financial sector, it is essential to develop an implementation strategy that considers regulatory, technological, and ethical aspects to ensure the responsible and secure use of technologies. It is recommended to invest in the creation of controlled testing environments (AI sandboxes) where new applications can be developed and evaluated in a protected context, ensuring that every new feature or use is compliant with existing regulations before any market release.


Another crucial step is continuous training and staff updates. GenAI evolves rapidly, and so do the skills needed to use it effectively. Financial institutions must invest in training programs to ensure that their employees are prepared to face technological changes and make the most of the new opportunities offered by GenAI. At the same time, it is important to encourage collaboration among different departments to foster a complete and shared understanding of the technology's potential and limitations.


Collaboration between the public and private sectors plays a fundamental role. AI regulation is still evolving, and cooperation between companies and regulatory authorities can facilitate the development of guidelines that promote innovation without compromising security or privacy. For example, the introduction of fair data-sharing practices, which allow access to high-quality datasets while respecting intellectual property and confidentiality, could facilitate the development of more performant and secure models.


It is also necessary for financial institutions to adopt open standards and transparency policies, which not only help avoid technological lock-in risks but also improve public trust in the use of AI. Transparency practices should include complete documentation of training processes, the use of mechanisms for model decision explainability, and regular audits to verify compliance with regulations and ethical standards.


Conclusions

The adoption of Generative Artificial Intelligence (GenAI) in the banking sector is not merely a technological choice but a strategic transformation that redefines the very foundations of operational and corporate competitiveness. It is not just about implementing tools to improve efficiency but about rewriting the rules of interaction between financial institutions, their customers, and the regulatory environment. This transformation brings extraordinary opportunities, but also risks that require deeper reflection beyond simple cost-benefit calculations.


One of the profound implications of using GenAI in banking is the redefinition of the concept of trust. Traditionally, trust in banks is based on transparency, solidity, and human reliability in making critical decisions. With GenAI, this trust must be extended to a non-human intelligence, an entity that decides and acts based on complex mathematical models and immense volumes of data. This implies a significant cultural shift for customers and institutions themselves, who must make otherwise opaque decisions understandable and demonstrate that such systems can operate without compromising ethics or security.


The democratization of advanced analytics, one of the main advantages of GenAI, introduces unprecedented dynamics in corporate roles and required skills. If GenAI systems can provide complex insights without the intervention of data science experts, traditional hierarchies within banking organizations are redesigned. This poses a managerial challenge: how to rebalance roles between technical specialists and strategic decision-makers, ensuring that the latter have the skills to interpret and fully exploit the provided analyses?

The ability to rapidly scale operations and processes through GenAI reduces traditional operational limits but also raises questions about long-term sustainability. Automating decisions and processes does not only mean responding to current demand but implies a reflection on managing future complexity. Systems that are too autonomous could create a level of technological dependence that makes effective human intervention difficult in crisis situations—a risk that no bank can afford to ignore.


In terms of innovation, GenAI also redefines the concept of time in the financial sector. It is not only the speed of execution of analyses or responses that changes, but the ability to predict and adapt to market changes in real-time. This acceleration creates a competitive context where leaders will be those who can integrate speed with accuracy and security. However, this same speed can make regulatory interventions more difficult, increasing the risk of a gap between technological innovation and regulatory capacity.


Ethics becomes the critical ground on which GenAI adoption in banking is played out. The management of synthetic data, the use of techniques like Retrieval-Augmented Generation (RAG), and modular fine-tuning, while reducing technical risks, amplify the need for transparent governance. The banks that will stand out will not only be those that successfully implement GenAI but those that do so in a way that makes the technology an element of trust rather than alienation for customers and stakeholders.


Ultimately, the introduction of GenAI in the banking sector is not simply a technical evolution but a systemic change that requires a long-term strategic vision. Sector leaders will need to go beyond the logic of efficiency and innovation to embrace a mindset of continuous adaptability, ethical responsibility, and inclusivity. Only in this way can generative artificial intelligence transform from an operational tool into a pillar of the future of financial institutions.


 

2 visualizzazioni0 commenti

Post recenti

Mostra tutti

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page