Recently, banks have started to apply artificial intelligence in practice in more areas, however, transition does not happen immediately. Appropriate internal and external competencies need to be built, data assets need to be standardised and cleansed, and in many cases the legacy mindset of banks becomes a barrier to the implementation of effective AI technology. This is not only about money – introducing an AI-based solution requires changes in infrastructure, operation, and culture. We talked about this topic with Szabolcs Pinter, Managing Director of UpScale, a company dealing with modern technologies, and presenter at our Banking Technology Conference on October 13, 2021.
How did artificial intelligence become such a hot topic recently?
Many believe that AI has become a trivial topic, while others think that now is the time to have meaningful debates about it. I remember a survey which showed that 86% of participants had heard about artificial intelligence in 2014 or even earlier, and it was a widespread phenomenon in Western countries already. Of course, if we consider that the first AI algorithms can be traced back to the 1950s and that by the 80s AI technologies were already commercially available, it might seem surprising that AI has only now become a real “hype”. However, given that the past decade brought about a general breakthrough in the adoption of connected devices, data storage and transmission technologies, or even infrastructure solutions, cloud-based services, “smarter” algorithms, and the maturing of the so-called “deep learning” technology, it is not so surprising, that the participants have only now realized the true potential of artificial intelligence.
Can we expect artificial intelligence to operate all solutions in the future?
Without doubt, we can expect that there will be more and more AI-based applications.
The expected business benefits have already been quantified by numerous consulting firms and there is a consensus within the industry on the high business potential. However, in our experience, there are many obstacles that are holding back the practical breakthrough of this technology.
If we take a large bank with significant capital resources as an example, what is it that is preventing them from “moving to” AI-based solutions tomorrow already?
This is not essentially, or not primarily a question of money. Typically, there are several factors holding back the digital transformation processes involving AI. The reasons are concentrated in three areas: infrastructure, operations, and culture. Not to mention anything else, a major problem is that in a significant number of financial institutions, the data required for AI-based analyses is not available at all or not in the right way. This is due to outdated systems, and the limitations of existing processes and human resource capabilities.
Modernization, in particular the modernization of data storage, could give a significant boost to the spread of AI.
What do you mean by the modernization of data storage?
Today, banking data is still often chained to different independent systems, or is available in highly aggregated form in low-availability data warehouses. The availability of the knowledge and algorithms needed to process them is also a major slowing factor. Compared to traditional data processing tools, Big Data analytics infrastructure is typically designed from scratch to enable the time-consuming task of processing billions of data records in manageable chunks.
This way, it is possible to benefit in parallel from the capacity-boosting effect of data storage distributed across hundreds of processors and servers, radically reducing the data analysis process from days to minutes. Scientific literature has summarized the defining requirements of Big Data in the so-called 5V’s. These are Velocity, Volume, Value, Variety and Veracity. In other words, the speed of analysis needed to measure data, the volume of generated data in seconds, the conversion of the collected and analyzed data into value, the quantity of typically unstructured data, and finally the quality and accuracy of data, together provide an objective criteria system for data adequacy. The modernization experience gathered by our company in recent years confirms that building the infrastructure required to effectively implement AI applications is a challenging task for a traditional organization and bringing in expert competence can make a huge difference to the efficiency of the transition, both in terms of time and money.
Why can’t AI transformation based purely on internal resources be implemented effectively?
The primary reason is perhaps not even the lack of competence within the organization. Of course, the obvious limitation is that the application of AI in the narrow sense of the term itself requires not only ‘domain’ knowledge but also ‘data scientist’, ‘data engineer’, and ‘DevOps’ expertise (and these are rarely found in the traditional organization), and moreover, the secret of successful AI projects is their truly cross-functional nature. In other words, the final product is the result of effective collaboration between teams from different fields. The availability of multidisciplinary knowledge, skills, tools, infrastructure, and management commitment are all important ingredients for success. This already requires a certain degree of agility within the organization.
But the most important reason is that AI is a new technology, and its adoption in an enterprise environment requires the internalization of significant implementation experience. Most of the time, our customers talk about their aspirations, not their practical plans. There is often a mix of goals and instruments. It is also common that the legacy mindset of banks is a barrier to the adoption of effective AI technology. Executives socialized in a corporate culture, used to bureaucratic decision making, now need to identify business cases with a strong focus, where they can invest concentrated resources in this new solution. Today, this often involves a plethora of internal approval and signature rounds. The whole transformation process is ultimately a huge change management project. An external partner with experience in modernization can greatly help in facilitating this way of thinking in a goal-oriented way.
Do banks need to change their mindset?
Indeed, a common problem is that banks try to manage AI initiatives that seem like good ideas with generic analytical skills within existing departments and end up with gigantic data-standardisation programs, which of course have no internal responsibility. Sooner or later, however, they recognize that in order to carry out more advanced analytical tasks effectively, data must be made available in the right format.
On the other hand, setting up monumental and lengthy projects is difficult to combine with the agile approach that has been brought about by the increased organisational adaptability in response to rapidly changing market processes and customer needs.
A better approach might be to select a few applications, considering infrastructure, operational and human resources, where the introduction of AI could have a direct and significant impact on a performance indicator, such as increasing revenue, reducing costs or increasing customer satisfaction. In this way, they can gain additional support from management in a step-by-step agile way.
Do you think that senior management is not committed enough to AI yet?
In my opinion, high expectations and fragile trust describe the current situation. As with all emerging technologies, trust can easily be shaken if major investments fail. Not to mention potential regulatory compliance concerns. As it stands, AI is still embodied in algorithms that are difficult to explain and are less transparent, which can be combined with system-level reflections of biases that are often difficult to justify.
Is it a challenge to comply with regulations?
Yes, AI is still essentially a “black box” solution that cannot meet the regulatory requirements for the explainability of models to the extent required. Consider, for example, the case of machine learning, which is the basis of artificial intelligence, where, given a data set of input and output factors, machine learning recognizes certain patterns and defines algorithmic relationships that can in the future reveal, in the language of mathematics, the correlations between the two factors.
However, from an economic point of view, these are sometimes difficult to interpret and, in some cases, ethically unfair. For example, why do we classify a loan as more risky for a particular demographic (age group and gender characteristics) group. The banking sector is a particularly regulated and economically dominant industry, so these considerations are of particular importance. In our time, this is coupled with the requirement to ensure compliance with GDPR and so-called Privacy legislation in general. Today it is still a challenge to audit an AI-based solution.
What is the solution to this problem? How can we make these algorithms, and AI itself, acceptable to the supervisory authorities?
The good news is that the technology is constantly evolving. Today, the term “glass box” has been added to the term “black box” in the field of artificial intelligence, and there are now modern tools to ensure GDPR compliance, for example, which can provide a satisfactory solution to the demand for legal compliance.
For example, there is a six-point system of expectations for AI-based credit scoring modernization. Although this has not yet become a generally accepted standard, it can be used as a benchmark. European regulatory bodies are also already focusing on developing a tailored legislative framework to facilitate the use of AI.
In conclusion, what results can we expect from the application of AI in the banking sector?
McKinsey recently estimated the annual incremental revenue associated with the rise of AI in the global banking sector at $1 trillion.
Interestingly, however, a study (MIT Sloan Management Review and Boston Consulting Group) earlier this year found that only 11% of organizations have achieved significant financial results from the use of AI. I am convinced that the huge gap between potential and realized results can be brought sufficiently close together with the right approach. This is what we will be discussing at the Portfolio Banking Technology Conference on October 13, 2021 at Corinthia Hotel.
