AI opens a door of possibilities for banks to modernize their operations. The banking sector is already discussing and implementing several use cases involving AI-driven applications. This is creating a new wave of more efficient, reliable, and automated way of working in banking that is here to stay.
Xomnia's Analytics Translator Sara Yuste Fernandez Alonso explores in the following blog current and potential applications of AI in banking. She concludes with guidelines about what banks need to join the data-driven revolution.
What are the potential opportunities for AI in banking?
The possibilities that AI can bring to banks extend to cover a wide range of applications, from front office to mid- and back office processes.
Examples of potential AI applications in the banking sector include creating tailored customer journeys; automating parts of the loan approval process; generating automatic spending reports; smart portfolio management; automating AML, KYC and risk management processes; supporting data migrations and credit scoring models, and many others.
Below is a breakdown of the potential AI applications in banking:
Front-office AI applications in the banking sector:
- Tailoring customer journeys with automated, AI-driven customer experiences
- Automating (parts of) the loan approval process for natural persons and organizations based on their transaction patterns and customer due diligence (CDD)
- Automatically labeling transactions within banking apps, which enables generating spending reports per category and predictions about periodic expenses
Mid-office AI applications in the banking sector:
- Smart portfolio management through analyzing customer data on the cloud (e.g. finding new products for customers by identifying similarities in features and expenditure patterns between new and existing customers)
- Automating money laundering discovery and KYC procedures
- Supporting data migrations, specifically from decentralized and historical data environments
Back-office AI applications in the banking sector:
- Automating (portfolio) risk management
- Helping the financial services industry adapt to a rapidly changing regulatory landscape (particularly new regulations such as recent sustainability guidelines)
Institution-wide AI applications in the banking sector:
- Automating data lineage and automatically extracting insights from it
- Generating synthetic data, which can be used for data analyses to train models more efficiently, or overcome data privacy concerns, among other use cases
- Supporting in the creation and validation of credit scoring models
How is AI currently being used in the banking sector?
AI is are already implemented in various applications across the banking sector and other financial institutions (FSIs): From front-office, function-agnostic applications, to regulatory compliance and risk management applications, all the way to data management and data governance applications.
Below we explore each of the AI applications in banking in depth:
1) Improving client and customer experience:
Banks and other financial institutions can achieve this by using AI to 1) generate personalized experiences for end-customers, 2) expedite services by reducing and automating routine and repetitive tasks, 3) reduce processing time for different kinds of requests, etc.
2) Improving efficiency or reducing costs:
By providing services such as automation, predictive analytics or anomaly detection, AI and ML help banks strategically prioritize their operations and the use of resources, such as their employees' working hours.
3) Adapting to rapidly changing regulatory environments:
Financial institutions are required to comply with an ever-changing regulatory environment. Many laws, directives, regulations and guidelines are in place to regulate the way the institutions manage their capital (e.g. CRR, CRD or Solvency 2 and the different IFRS standards), their portfolio data (e.g. GDPR), or their risk reporting practices (e.g. BCBS 239), among other aspects. In many cases, financial institutions struggle to get a good grasp of the changes they need to do in time to be compliant with regulation. Institutions might end up receiving fines or capital add-ons from supervisors if they do not act quickly enough. AI can help banks and other financial institutions in identifying and assessing what needs to be done, where to find the relevant data, or even scanning the regulatory horizon to understand which regulations are relevant for their business.
4) Adapting to new sustainability requirements and regulations (EU taxonomy, ESG):
Attention towards sustainability has grown exponentially in recent years. With the release of the EU taxonomy, financial institutions are now facing a great challenge to comply with an entirely new taxonomy. Unlike the long-standing capital or risk regulations that these institutions are already familiar with, sustainability requirements are new and in many cases leave these institutions wondering what to do.
Moreover, navigating this environment is not easy, due to the lack of a well-defined reporting framework, methodology guidelines, and sufficient implementation examples. This has often left it up to institutions to interpret these regulations and decide how to implement them, which further adds to the challenge, both for financial institutions and for the regulator. AI-powered products and solutions can help in understanding the complex sustainability regulatory environment and in enabling its implementation.
5) Better conducting KYC, particularly risk management and fraud detection:
Managing big portfolios implies a considerable risk for financial institutions. Especially in the case of portfolios that extend over several geographies, identifying high risk clients or fraudulent operations can be like finding a needle in a haystack. AI can greatly help in tackling this issue due to its ability to recognize patterns and single out outliers and strange data points; for example, anomaly detection models can be used to point out criminal behavior. Additionally, AI can be used to quickly process huge amounts of data, further strengthening its relevance in KYC efforts. We have explored in more depth how these possibilities fit within the latest updates of the EU AI Act in a separate blog.
There are currently large initiatives in the Netherlands focusing on data sharing particularly to detect criminal behavior. For instance, TMNL (Transactie Monitoring Nederland) is a collective initiative by the biggest banks in the Netherlands to share criminal transactions and use data and AI to counter money laundering.
6) Aiding in data migrations, especially when dealing with historical databases:
Financial institutions might need to perform data migrations for a number of reasons. Sometimes these institutions want to modernize and move towards more robust or manageable environments (like cloud environments). Other times they are required to update their environments due to external factors (for example BCBS 239).
Whatever the reason, these migrations often happen from unstructured data lakes and historical data environments. Relevant data is commonly scattered across several systems, and there is lack of ownership and stewardship. Consequently, complex data remediation efforts are necessary before the migration can take place. Tackling this problem implies a great effort, time and resource investment.
AI can be leveraged to significantly reduce the time and effort required to connect data sources, find relevant data points and business rules in the data, and ultimately reduce the amount of manual labor needed.
7) Improving metadata management:
Financial institutions are becoming increasingly aware of the power of harnessing and managing metadata (or the data about data). Several data catalog tools are adapting their offer to include metadata solutions, like Informatica or Collibra.
When it comes to metadata management, AI offers many powerful possibilities. From seamlessly connecting distant (yet related) metadata points, to creating insights from metadata, or even expanding the understanding of data itself. Leveraging AI to enable proper metadata management can be a determining factor in the success of a financial institution’s data management efforts.
8) Better determining data lineage (ETL):
Following the journey of data from collection to reporting, ETL, and data transformation journey are all well-known concepts for data professionals in the FSI. Understanding how data flows from A to Z, however, can prove quite tricky. This is because data is often not centralized, comes with little to no documentation, and often lacks ownership. Moreover, it’s not rare to find highly unstructured data sources that date way back in time.
Luckily, AI offers many possibilities to reduce the time and effort required to connect dispersed and unstructured data, identify related data points or even uncover transformations done in data. Data professionals in FSI will find a powerful ally in AI when it comes to data lineage, letting the algorithms do the dirty work. This leaves them with the time to focus on using their expertise to provide insights, make decisions and advise their clients.
What do banks need to be able to implement AI?
There are many opportunities for financial institutions to leverage AI in their practices, and many are being already applied in banking sector. These are, however, mostly seen in small, low risk projects, as AI is rarely used to its full potential in the FSI.
Although the potential of integrating AI is relatively clear, it is usually perceived as high risk: there is not enough trust in AI and it’s hard to explain. Moreover, there are implementation barriers that slow down efforts to incorporate AI into institutions’ businesses, since in most cases the data infrastructure is not ready for AI use.
To enable the implementation of AI into their processes, banks and other FSIs institutions need to address some necessities:
- Developing mature data infrastructure
- Securing and managing the necessary resources
- Properly addressing security, privacy and ethics concerns
- Willingness to explore opportunities brought by adopting state-of-the-art AI technologies
Below we explore each on of these fundamental necessities:
1) Developing mature data infrastructure:
Many AI projects in the banking sector require significant data engineering and data remediation pre-work before they can start. This is because banks often struggle to find quality data, which is rarely accessible and ready for AI use.
Banks’ data is usually scattered in historical databases and needs a lot of data remediation before being of value. Moreover, it is very often highly decentralized, without a fast and easy way to access it. In order to implement AI driven capabilities, it’s necessary for banks to begin with improving data management, data storage and data stewardship.
2) Securing and managing the necessary resources:
Most financial institutions lack enough developer workforce to implement AI solutions, as well as engineering experts that can deploy and support of these solutions. Luckily, several consultancies with experience in the financial sector like Xomnia can help those institutions.
3) Properly addressing security, privacy and ethics concerns:
There is a small risk of information leakage and/or misuse when using AI technology. This is especially tricky for banks, considering the nature of data that banks own, which tends to be personal data to a large extent.
This challenge can be mitigated in many ways, such as implementing proper security and data governance measures and policies by enforcing data anonymization as a pre-requirement in any AI project. Another method is using tools like generative AI to generate synthetic data that does not contain any personal information (due to its synthetic nature). Last but not least, by working with qualified AI professionals, banks can ensure the ethical and responsible use of AI from the get-go.
4) Willingness to explore opportunities brought by adopting state-of-the-art AI technologies:
The financial sector is known for its risk-averse and prevalently conservative culture, which springs from several characteristics and regulations unique to this fundamental sector. Stakeholders in FSI institutions are wary of using technology whose decisions are not fully explainable (which most of the state-of-the-art AI/ML technologies will never achieve). They are therefore cautious about supporting decisions made by a machine which they can not interpret. This also means that even after implementation, scaling AI solutions across different banking processes, systems and departments is a challenging and slow process.
Banks and other financial institutions can begin to mitigate and address those concerns by adopting and incorporating responsible and explainable AI approaches from the beginning of an AI/ML project. Several consultancies with experience in the financial sector and regulations in Europe can help banks make strides in this.
This blog was written with the help of Machine Learning Engineer Michael Schoustra, Data Scientist Alexandr Koryachko, CTO Tim Paauw, and Analytics Translators Jasper Küller, Robin Schut, Ichelle van Kleef, and Lisanne Rijnveld.