Artificial intelligence in the banking sector: how to make data work

Artificial intelligence will play a central role in the economy of the future and serve as a growth driver. Application of artificial intelligence ranges from predictive analytics and chatbots to fraud prevention and regulatory compliance. Banks ‘ investments in machine learning technologies grow by an average of 46.2% annually, but many projects remain incomplete. So the question is how to make such projects effective.

Two effective approaches to implementing AI in banking

For the past three years the banking industry has been gripped by a real “gold rush” in terms of artificial intelligence. Banks have created a lot of data lakes, launched a lot of AI initiatives, and spared no expense in attracting the best young specialists in data processing and analysis.

And what do we see as a result of these 3 years? There was a lot of sand and very, very little gold. According to Gartner, 80% of AI projects have failed or have not progressed beyond prototype development.

What is the reason for this? It seems that all the necessary prerequisites were available for the success of the projects:

• a large number of open source machine learning algorithms have been developed;
• there are significantly more data sets accumulated than one can imagine;
• computing resources are widely available.

Evaluating the results, one can assume that banks are very actively engaged in business, without having thought through a strategy for working with data. In many large projects that have been launched, responsibility has been assigned to its departments, rather than to business units.

For the AI projects to bring results, it makes sense to follow two strategic approaches.

The first is that machine learning capabilities are adapted, implemented and integrated into all banking applications for the front office and back office, including for digital interactions with customers, compliance with regulations, financial management and HR management. As a result, customers can use AI capabilities directly at the time of the transaction.

The second approach is to focus on specific cases with a clear understanding of business goals. Being specific is helpful for performance assessment and result orientation.

Indeed, rushing for AI projects banks faced many difficulties and here come recommendations for making such projects effective:

• start small and expand your approach,
• find out the needs and demonstrate the benefits,
• first of all, think about business results,
• assign a responsible manager from the main business, not from the IT office.

How banks can develop an effective AI strategy

Many banks initially treated data lakes as a sacred cow when working with AI. However, some of these data lakes have unfortunately turned into swamps. As a result, data processing and analysis specialists and developers began to move to financial technology companies where there were no problems with data and getting access to it, unlike in banks.

Gartner analysts estimate that 70% of the AI effort is spent on data management: getting, cleaning, preparing, and processing.

To ensure that customers do not have to spend time moving data, updating it, and ensuring its integrity, it makes sense to implement machine learning algorithms, such as R and Python, directly in the data source in real time.

At the same time, it is possible to implement innovative tools for autonomous data management to reduce the load and significantly simplify all standard AI operations. This radically changes the paradigm of working with AI data, as users can fully focus on their strategic analysis.

Thus, banks will be able to extract the gold that is hidden in various internal data sources faster and with much less effort.

Artificial intelligence under control

AI conclusions reliability is of increasing concern to regulatory authorities in many countries, that is why they are increasing their supervision of banks ‘ initiatives in this area. Banks have to explain the logic of machine learning models to them. In particular, this applies to the use of AI in robots-investment consultants, for recommendations on transactions or monitoring compliance with regulatory requirements.

Regulatory authorities also have requirements for data localization. To meet these requirements, it is possible to create a virtual machine learning platform in the private cloud inside the data center, protected by firewalls. If data sets do not contain personal data, sometimes they can be processed in a hybrid environment or a public cloud.

Thus, continuous cooperation between banks, regulatory authorities, and technology partners is very important for transparency, reliability, and compliance with ethical values when banks work with AI.