In general, the sectors that embrace faster innovation are those in which there is a constant struggle for clients, who are increasingly unstable every day - the banking sector being representative of this point of view.
Some banks prefer to adopt innovation as an operating mode, others to wait and analyze the market reaction before implementing similar technologies or models. When adopting the ML, there seems to be a consensus that it will improve personalized interaction with the customer, reduce sales costs and response time, and improve the accuracy of predictions in operational activities, such as sales performance, the risks of non-performance product, market evolution, or customer behavior.
If we analyze more advanced markets, the investment banking sector (especially commercial) has been using advanced mathematical models and last generation infrastructure (including private cloud) for automation of selecting profitable assets and executing transactions for nearly a decade.
Early adopters of Machine Learning algorithms
The first ML versions were used to help detect fraud and security risks by using a complex set of rules. Newer algorithms provide active protection, adapt to real-time threats and eliminate potential threats by continually analyzing internal and external data flows.
Another area was the automatic management of portfolios (or robo-consultants).
These involve mathematical algorithms combined with automatic learning, which monitor and make semi-automated decisions (depending on certain thresholds) that calibrate the financial portfolio so that it is consistent with the revenue and risk exposure objectives set by people.
In other areas, such as credit and underwriting, automatic learning algorithms have been experienced to increase the accuracy of predictions or to reduce the insurer's risk in general insurance products.
In line with customer-oriented policy, ML has also begun to generate a better level of customer interaction, increase loyalty, and, as mentioned, to anticipate threats to cyber security.
Machine Learning: How Does It Affect Us?
The question that comes to mind when we hear about ML is related to the role we will have after the emergence of an ultra-fast robotic intelligence that will guide us and perhaps monitor and will be based on data that we did not necessarily have access to ( see "fake news").
What we know for the moment is that ML has transformed most of the workplaces, improving human knowledge by faster analysis of a larger data volume. Take the example of our customer support centers: ML-based robots respond to frequently asked questions, thus reducing the need for staff.
At the same time, ML also created many jobs. These solutions require the workforce to structure and classify the existing knowledge base for each client and also to "train" the algorithms to give specific answers. And intelligent technologies need intelligent people.
I believe that the tendency of ML-based customer support centers will mature in the Romanian banking sector in the next 12 months - we already see many banks that have launched or are experiencing this concept such as Banca Transilvania, Libra Bank or Raiffeisen Bank.
We are only at the beginning of the road, and the current interactions generate at the same time a lot of frustration for the clients.
Many banks that want to differentiate themselves further in their client engagement, while retaining cost optimization plans, have recently focused on synergies offered by RPA (Robotic Process Automation, Automation of Processes through robotics) and ML, with improved results both in quality and efficiency. Combining Structured Data with Unstructured Text is the Holy Grail for most departments of marketing, sales, and risk analysis.
As voice interaction reaches the forefront, I anticipate a greater integration of mobile banking with voice-based robotic applications.
What to do?
Banks are under increasing pressure to provide their customers with the platforms, dates and experience they expect, while respecting the privacy, security, legal and ethical requirements of the GDPR.
To really use ML as a competitive differentiator, banks will first need to identify commercial usage scenarios and understand what relevant data are available and where, then, clean, classify and improve through external data.
A decision on infrastructure and applications (currently many of the "open source" type) is required, together with the identification of capabilities and an ecosystem of partners for the successful implementation of the solutions.
Once the usage scenarios are agreed and implemented, performance improvement and ML training need to be monitored to improve accuracy.
I am convinced that banks that have a clear and achievable plan that has identified the available data sources and will soon adopt ML algorithms will benefit from a positive reaction from the market, improve their profit margins by reducing global risks and, rightly, consequently, provisions.
They will also improve their operational performance by identifying inefficiencies and gain more market share, as attracting new customers and personalized interaction will generate loyalty and, implicitly, higher revenue.
