Within the asset servicing industry, how is AI being used? Over the last 12 months have you seen huge developments?
Keiko Kataoka: As far as we understand, we haven’t seen any huge developments in the last 12 months in the industry. A lot of AI projects are still being developed. Most of the projects are linked to back-office operations, such as compliance and data quality improvement.
How is the implementation of AI changing roles in compliance and operations?
Kataoka: We understand the expectations for AI in the back-office operations area but to others, the compliance and operations in the back-office area can seem very complicated.
We have the routine type of work as well as errors outcomes, at this time it is very difficult for AI to operate under all of these issues. We understand the potential of AI but when considering the type of work that is being done in the compliance and operations area, it’s about handling all of the data.
AI is not relevant to every area. At this moment, we don’t think firms are truly ready for AI implementation because accuracy is so important, and we don’t think that AI can do better work compared to a human right now.
Mitsuhiro Tsunoda: Continuous human intervention is indispensable for quality control of AI. In the initial ruling of AI, a large amount of future data that is created by humans is required. It is necessary to continuously learn, even after the first time of incorporating AI into your business model.
Would you agree there is an unnecessary hype surrounding AI, or do you think it is the only way forward for the back office? Should all asset servicing companies be embracing it?
Kataoka: There are still unnecessary hypes surrounding AI, and we do not think that AI will be the substitution for the human in the near future. From this point of view, AI is not the only way forward for the back-office.
Considering the current environment in Japan, we have an ageing population problem as well as a shrinking workforce, so we need to think about sourcing the right staff, replacing the staff we have with the younger generation who understand the most up-to-date technology when it is time for the older generations to retire.
AI can be a good candidate and everyone needs to understand how to work together with AI to have the correct result from operations. From this point of view, we need to be ready for AI. Firms need to understand what AI is. In the past five or ten years, AI has been the buzzword, but its meaning is very broad.
When we say AI, it could mean anything from robotics to natural language processing. We can’t transfer everything under the umbrella of AI or robotics. Humans need to have the experience and ‘know-hows’ in every company concerning how AI can bring opportunities to asset servicing. To do that we need to evaluate the tools correctly and implement and combine the work appropriately as an industry.
Before implementing AI into companies, what should firms be doing to ensure it’s the right thing for them?
Kataoka: Evaluating what AI can do is the most important element. When we say ‘proper evaluation’, this includes ideas like evaluating what AI can and can’t do. We also need to understand how AI can affect other processes.
We need to look at the limited processes to which AI could be implemented but also utility processes in order to optimise the whole operational processes.
For instance, NRI developed a data extraction tool in which AI techniques are being used. If you implement digital signatures with the tool, you can maximise the performance of data extraction tools because you do not need to rely on OCR. And that potentially can happen anywhere.
What did NRI’s specific proof of concept studies with AI and robotic process automation (RPA) reveal in 2018? In terms of the back office, are there any specific trends that stand out?
Tsunoda: In terms of AI and robotic process automation, natural language processing (NLP) is very compatible with data extraction work. In terms of the task for re-entering data for fund attributes and for mutual funds, they were used to do contract attributes for structured notes.
It may take a long time to search for certain target data from a large number of documents which are unstructured. So we verified our approach which confirmed the data work greatly improved.
By combining AI and RPA, we can expand our range of operation that can be streamlined. If there is a high-load operation that can be put on the process to be carried on AI, is it not possible to streamline the entire business flow with just introducing AI?
In such case, automating high-load operations with RPA, will optimise the entire workload.
AI quality control processes are indispensable, as long as the accuracy of AI doesn’t reach 100 percent. It depends on human beings at present.
We believe that the existence of a human in the loop will report AI’s liability and the continuation of the latest AI operations.
What opportunities and value can AI bring to a financial services company?
Kataoka: From the aspect of fraud detection and legal tech, AI can deter so many possible cyber attacks, and there is demand—people and business want that kind of protection. We’ve come to expect more from AI and how it can protect us from cyber attacks and viruses.
What studies will you be working on in 2019 that have an asset servicing focus?
Kataoka: What we have learned so far from 2018 is data work is very suitable for AI. A lot of gathering meta-data is currently created by human so data is one of the bottlenecks concerning AI. We are thinking about introducing AI and NLP in some parts of our back service operations in our BPO.
We collect information in office operations from various types of companies and AI is more suitable for larger types of operations rather than for smaller volume type operations. Having AI capability with BPO offerings makes sense.
That’s what we are trying to achieve in 2019.
