How artificial intelligence and robots help the bank become more efficient and smarter

SEB Pank has been actively using a range of modern technologies for a long time already. One such modern solution is also entrusting part of the work to robots. While most robots work on processing simpler transactions, combining robots and machine learning is also an evolving trend.

The growth of any company goes hand in hand with the ability to adapt to conditions and use advanced capabilities, including technologies that allow processes to be digitised and automated, thereby increasing efficiency and reducing costs. When we talk about the potential help that robots can provide, many people first imagine humanoid robots whose task is to do something physical; however, when (digital) processes require automation, it will be the software-based robots that come to the rescue (robotic process automation, i.e., RPA). In essence, robots in this case will be taught a specific work process that needs to be repeated (for example, copying some data from one data source to another), which is then either repeated according to a given command, or the robot constantly repeats the start command to then perform the repetition pattern as necessary. Of course, such simpler processes could and should be replaced by better integration of systems and the development of ‘real automation’, but often the respective development is too costly or takes too much time, which is why sometimes building robots to perform a specific job is a reasonable compromise. Often, the speed of robot development is also a key argument; SEB, for example, developed robots to help it cope with the exponential growth in the number of clients due to the flow of refugees from Ukraine, proceeding from an idea to a working robot in just a few days.

Robots enable employees to focus on more important topics
In a bank, as in any business, there are monotonous everyday tasks that do not require complex skills or knowledge, but are necessary and still take quite a lot of working hours. These monotonous jobs that repeat with a certain regularity, as well as processes related to the processing of large amounts of data, are also the most suitable candidates for the involvement of robots. At the bank, a specially created Robotics team is responsible for this area; the team operates in each of the Baltic States and, together with the various structural units of the bank, constantly identifies processes with a potential to be automated. The team always conducts detailed surveys – even if it turns out that robotisation is not necessary or possible for a particular job, the survey will still simplify the process and introduce additional improvements.

Robots at SEB Pank not only make everyday life easier for employees, allowing them to focus their energy on more important tasks that robots would not be able to handle, but they also ensure the faster processing of clients’ wishes. For example, robots in the bank are currently engaged in creating credit agreements and insurance policies, processing requests for audit reports, as well as creating and processing leasing offers and agreements.

SEB first began introducing software robots a few years ago, and today there are 75 robotic colleagues working in the Baltics, who perform the work of more than 50 full-time employees, making an average of 200,000 repeat movements in a month. While the largest number of robots work on transactions and simpler operations, the greatest efficiency is seen in performing the ‘know your customer’ (KYC) procedures. A third of these robots work in Estonia.

New solutions emerge when the consistency of robots is combined with other AI capabilities
Lennart Kitt, the Head of Customer Analytics and Data Science at SEB Pank, explains that robots have been taught to repeat certain human activities, but they cannot come up with anything new on their own. At the same time, new opportunities have opened up to make robotics work in conjunction with other branches of artificial intelligence, such as machine learning or optical character recognition. The development of smarter solutions created in this way is also one of the main tasks of the Robotics team for the near future.

For example, tasks are already performed in the bank by a robot which, in the process of archiving photos of real estate objects, must detect whether a photo depicts the exterior, interior, etc. of buildings, using artificial intelligence. At that, the initial machine-learning-based version of the work section repeated by the robot was developed together with students from the University of Tartu, and was later adjusted and retrained according to actual bank data and processes. Going forward, the direction will continue to be developing robots that can recognise and process not only digital, but also scanned documents and images.

Automation – an opportunity for everyone
The democratisation of automation, i.e., process automation at the level of their usual performer, has been introduced as a new direction. Its essence is the principle that the automation of processes should not be exclusive to a few (top) developers, but that everyone should have the opportunities and skills to make improvements to their segment of work. As part of a pilot project, mentor-mentee pairs of employees were formed, where the mentor directed the mentee to learn suitable digital tools so that they would be able to automate their own daily tasks, which are not complex by nature, but require quite a lot of time. Democratisation of automation at SEB has not only increased efficiency, but also developed the competence of employees and increased their motivation.

Five tips for robotising processes
SEB Pank has been developing software robots for several years already. Based on its experience, the bank has prepared the following control questions to help select new processes to undergo robotisation:

1. Is the process stable? As robotisation requires a developer resource, SEB has set the aim that it is worth robotising processes in which no changes are foreseen within one year. Since robots mimic human labour, then every time something changes in the process, the robot also needs to be retrained. An unstable process, however, can lead to a situation where the constant retraining of a robot consumes more labour than performing this section of work the old way. At the same time, if variables in the process are known in advance, then it is possible that the robot can be taught to take input from, for example, a dynamic input file (e.g., team formations, days off, etc.).
2. What is the realistic level of efficiency that can be achieved with robotisation? It is possible that if the gains in terms of labour are very high, then it may still be worth it to review the priorities of the developments and find a way to create systematic automation. If the gains are very low, it should be checked whether the time invested in the development of the respective robot will pay off within a reasonable period (such as 1 year). Also, when calculating this indicator, one should critically consider whether the given robot could also be used in the work of another department. However, if the process is very small and simple, then perhaps it is a good process for democratisation of automation. The practice of SEB has shown that it is reasonable to robotise processes that take up the working hours of at least 0.2 employees, i.e., in total, one employee should repeat the given section of work for at least one full working day per week.
3. Will a (temporary) increase in costs be acceptable? Additional costs may be involved in the implementation of robotics. It is necessary to train developers, set up the infrastructure, possibly pay licence fees, and the robots that work also sometimes require maintenance and improvement. Well-chosen processes make it possible to cover this cost on account of making things more effective, or increase revenues by directing the newly available labour to more profitable activities. However, during the implementation process, an increase in costs will also probably have to be taken into account.
4. Is the input standardised? Robots are not creative. Thus, they also have difficulty in special cases or when processing unstructured data (photos, handwriting, unstructured text). If solutions can be found for the processing of unstructured data then, when managing special cases, it will become necessary to consider whether it is reasonable to describe each possible individual case and the potential rhythm of behaviour, or solve only the most typical cases with the help of a robot and continue to leave the more complex / rarer cases in the hands of employees.
5. Are people ready for these new ‘colleagues’ and skills? As robotisation does not mean full automation, and the corresponding section of work still remains (but is carried out by a robot), robots should be viewed as colleagues who help the team within the framework of their skills – how they have been trained. They need to be trained in a specific way, which means, among other things, that even the smallest movement that an ordinary worker would perhaps do intuitively must be described step-by-step. The results of the robots’ work should also be systematically monitored, random follow-up checks should be carried out, and it should be analysed how to reorganise its work to achieve better results or modify its work to keep up with upcoming changes to the process. In other words, a ‘performance and development interview’ should also be conducted with robots. These are new skills and new working methods that may not be very easy to adopt.

Lennart Kitt
Head of Customer Analytics and Data Science at SEB Baltic.