Machine learning in APAC banks: What are the best practices and who's doing it right?

Simon-Kucher's Iraklis Kordomatis tackles how the technology is being implemented in banks in the region.

Asian Banking & Finance caught up with Simon-Kucher & Partners' senior consultant, Iraklis Kordomatis, to talk about machine learning adoption in Asia Pacific and what banks in the region should best do to leverage the technology to enhance customer experience and internal services.

What is the status of machine learning adoption in Asia Pacific banks? Are they at the forefront, are they lagging, or are they in the process of experimenting?
A good indicator for the adoption of machine learning for banks in Asia Pacific compared to other markets, is the number of VC-backed fintech deals around the globe. The reasoning behind is that most banks acquire machine learning (ML) knowledge from external partners via R&D partnerships, outsourcing (ML as a service) or strategic investments into fintechs, which usually have a steeper learning curve when it comes to applying machine learning methods. The latter is most transparent to observe in the market and provides the opportunity to compare different regions with each other.

Source: CB Insights, Global Fintech Report Q2 2018

The main takeaways are that Asia is not top in number of deals, with North America being the market leader. However, a change of leadership in the near future is on the horizon with Asia having the highest growth rate amongst the regions. Europe cannot keep up with the pace of Asia and North America and is expected to fall behind further.

How can banks specifically in the Asia Pacific leverage machine learning to offer better products and services and improve their core systems?
The opportunities to leverage machine learning are vast. The main areas of improvement can be divided into three categories: enhanced client experience, new products & services, and improved internal services.

Enhanced Client Experience: Customer behaviour is very heterogeneous in Asia Pacific economies. On the one hand, countries like China, Japan, Australia, Singapore, and South Korea have a tech-savvy client base, who demand a great digital banking experience. Whereas on the other hand, countries with a low smartphone penetration have a higher demand of face-to-face- or telephone-banking.

Tech-Savvy Economies: For basic functionalities, the implementation of ML techniques are usually not required. However, if a bank wants to go the extra mile and increase the likelihood of a high client satisfaction, applying ML techniques can do the trick. Applications of ML algorithm can range from payment authentication with face or voice recognition, automatic credit approvals, dynamic pricing to tailored client recommendations for products and others.

Face-to-Face: Providing relationship managers with tools and systems which enable a short handling time to increase the request throughput and reduce the waiting time in branches. These tool’s and system’s decision-making can be enhanced with ML techniques. Usually, most of the applications are in the lending area in those markets.

New Products & Services: Based on the nature of Asian economies, the highest impact of ML can be reached in the following product offerings: crowdfunding & crowdinvesting, microfinancing, and real-time fraud detection for payments.

Crowdfunding & Crowdinvesting: Alternative lending becomes increasingly popular in Asia. Due to the large population, it is an attractive business area for providers. Current developments indicate an increase in alternative lending for SMEs. In order to run such a crowdfunding or crowdinvesting platform, certain services need to be provided. For some like analysing funding or investing proposals and providing client ratings, ML techniques should be leveraged.

Microfinance: Asia has the highest amount of people at the bottom of the pyramid. Serving those clients efficiently is challenging. An enhanced decision-making and optimised risk management with ML algorithms would provide the potential to maintain a healthy and profitable credit portfolio.

Payments - Real Time Fraud Detections: Asia’s e-payment providers are amongst those with the highest transaction volume. The need for fraud detection is challenging when expected to happen in real time with a broad set of transactional data. With the right ML techniques, those requirements can be met. For textual (e.g. emails) or acoustical (e.g. telephone recordings) input data, an upgrade to deep learning fraud detection systems needs to be considered.

Improved internal services: In Asia, labor costs are significantly low, which makes large investment in automation of systems relatively unattractive. As a result, business processes in banks are less automated compared to the North American or European peers. Given this stage of low business process automation in Asia, ML is not the game changing technology yet. However, risk management is a domain where ML can have a huge impact. Measuring risk more accurately than via classical methods leads to better decision making and reduces the need of too high safety margins. Eventually, the sum of those effects results in lower cost of capital for banks and therefore increased profitability.

What are the best practices in implementing machine learning in banks? Is there a particular Asia Pacific country that is doing it right?
That depends on the goal and the current state of the bank. The fastest way is to acquire a fintech with the required technology. However, banks usually acquire a fintech, because they want to offer the fintech’s product or service to their client base. Banks do not want to learn from the ML capabilities which have been leveraged for the product or service offering as such. Another disadvantage of acquisitions is that those are usually quite expensive compared with other approaches.

Another approach is to team up with a company, which provides ML services. Australia’s top four banks are going into this direction. An advantage is that this approach is rather inexpensive and depending on the setup, short time to market still can be achieved. However, building a trustworthy relationship with a company, with whom a bank usually shares quite sensitive information, potentially at the other side of the globe is usually not as easy.

The last approach would be to build the capabilities in-house, which is generally more expensive than teaming up with an ML service company, but less costly than acquiring another company. However, the time to market for a potentially ML enhanced service, product, etc. will be likely the longest. Building ML capabilities in-house do not necessarily mean to do everything in-house. Having an ML expert in-house as a sparring partner for the business will be already beneficial as the expert can quickly assess which ideas can be realised with applying ML algorithms and will help funnel ideas into concrete projects. Another potential role for an internal expert is to advise the board on strategic acquisitions of fintechs. A third role is to be the relationship manager for the ML service company the bank is mandating, to delivering the ML solutions. Anyhow, it has to be said that establishing a team of ML experts within a bank is strongly recommended.

What benefits will machine learning bring to banks that will use it?
Derived from question two, the three main advantages of applying ML is the increase in client satisfaction, reduced cost structure, and increased profitability.

Increased client satisfaction: Serving clients better than peers results in a reduced client churn and increased consumption (happy clients spend more). All this results in higher revenue potential for a bank.

Reduced cost structure: Optimising internal process and decrease in cost of capital can significantly reduce a bank’s cost structure.

Increased profitability: An increase in revenue potential with a decrease in cost will lead to increased profitability of the bank.