Quickly assess whether there are likely to be any repayment problems
Complex challenges can sometimes quickly be narrowed down to the start of a solution. When PwC became involved in ING’s efforts to develop an Early Warning System (EWS) for credit risks, one of the first steps was to construct a rudimentary demo version. “It was a real eye-opener,” says Anand Autar, head of portfolio management at ING. “My reaction was ‘If we can already construct this in just two weeks, then a lot is going to be possible.’” Mr Autar is the driving force behind the digital credit risk alert system that ING is developing in collaboration with Google and PwC. “As a lending institution, we need to be able to properly monitor customers who borrow from us, so that we can quickly assess whether there are likely to be any repayment problems. For us, credit risk management is therefore a core competency, not just from a commercial perspective but also because it’s our fiduciary duty towards savers who entrust their money to the bank.”
From left to right: Anand Autar (ING), Petra Stojanovic (Google) en Pieter Verheijen (PwC).
Keeping a grip on data
The challenge is to keep a grip on the tsunami of information, which is often in highly fragmented form. A multitude of information providers flood the world with data every day,” says Mr Autar. It’s therefore quite conceivable that a credit manager – who sometimes has dozens of companies in his portfolio – won’t notice relevant news among all the information that he receives. Take the example of a report in a local newspaper about a borrower being involved in a fraud case, in a language the credit manager can’t read.”
That’s why ING started experimenting a few years ago with developing a digital warning system. Doing so helped it to clarify the requirements. The bank wanted the system to be scalable and easy to connect up to a wide variety of data sources. It was also important for ING to be able to explain the system’s underlying machine learning models to regulators, which was a major reason for it to develop the EWS in house.
Google services as LEGO bricks
It isn’t really surprising that in its search for a solution ING subsequently turned to Google. “Google Cloud provides a digital environment that’s ideal both for experimentation and later for scaling up,” says Petra Stojanovic, financial services lead at Google. ”And Google News is an information source that ING would very much like to use. As a tech and data company, we also have extensive expertise in the field of machine learning and artificial intelligence. aspects that play a major role in this project.”
Stojanovic compares the services and products of Google Cloud with LEGO’s building bricks because they which enable companies to construct customised solutions for specific challenges. “The technology already exists,” she says. “It’s all about connecting up the ‘bricks’ in the right way. The modular structure allows you to experiment rapidly. And if you take a wrong turning, you can easily go back to the starting situation and take a different route from there again.”
Fast, efficient, and effective core processes
These insights were the starting point for collaboration between ING, Google, and PwC. “Our role is to create the solution that ING proposes using the technology that Google provides,” says Pieter Verheijen, senior innovation manager at PwC Advisory. A multidisciplinary team was therefore assembled within which ING’s credit risk experts worked with a cloud architect, data engineers, a user experience (UX) designer, and a front-end developer.
“We had a proof of concept within just twelve weeks,” says Anand Autar. He believes that on the ING side, support at the highest level within the bank has been crucial in developing and rolling out the Early Warning System. “The bank has a clear vision when it comes to applying digital technology to make core processes faster, more efficient, and more effective. Development of EWS fits in with that vision.”
Market data and news from public sources in real time
The EWS is an application based on machine learning which processes, organises, and presents available data that can have an impact on borrowers’ creditworthiness. The system offers credit risk managers a solution for information overload, fragmentation of sources, and language barriers, thus supporting decision-making.
For the time being, ING is only using the system to identify credit risks in large companies, public authorities, and financial institutions. The EWS is fed with real-time market data from Thomson/Reuters and news from public sources, including Google News. In its current set-up, the system processes approximately 80,000 news items each day.
Based on natural language processing, it determines whether a news item actually concerns the company in question, or whether the company is only referred to coincidentally. The application also checks whether the item is relevant for the credit manager. For example, a report of an accident in which the make of car is mentioned is probably not relevant, unlike one about a downgrade by a credit rating agency. The last step is for items concerning the same event to be clustered.
Credit risk managers can add their own alerts, for example if a company’s share price falls by more than a pre-set percentage, it’s named in connection with fraud, or a certain report represents a highly negative sentiment. The algorithm can be refined further because the end-user can provide feedback about the results that are found.
"As far as ING is concerned, the proof of concept is just the beginning. Ultimately we want to have a tool that can predict credit risks."
Success factors for cooperation
In addition to the multidisciplinary composition of the team and support for the project at senior level within ING, other success factors have played a role too. “We opted for ‘agile’ working,” says Pieter Verheijen, “based on brief (two-week) development sprints concluding with a demo for a restricted group of future end-users. Their feedback can then immediately be incorporated into the following sprints. That way, development takes place in close cooperation with the end-users and you can constantly carry out adjustments.”
According to Mr Verheijen, carefully defining the scope also contributed to rapidity achieving proof of concept. “We limited the scope by only making use of public information sources, so at that stage it wasn’t yet necessary to link the system to the bank’s own data sources. That made developing the proof of concept much easier for all parties involved.”
Refining the risk models and algorithms
At the moment the system is only being used by a limited number of ING’s credit risk managers. Anand Autar explains: “The proof of concept has demonstrated that the EWS actually works. The next stage is to roll out the system throughout the bank and integrate it into the internal processes and the IT infrastructure. We’ll also be training our credit risk managers, because they are the ones who’ll eventually be working with the system.”
As far as ING is concerned, the proof of concept is just the beginning. “Ultimately,” says Mr Autar, “we want to have a tool that can predict credit risks.” That aim requires further refinement of the underlying risk models and algorithms. “Just like a human, the system learns from previous situations. For example, it needs to become increasingly effective at alerting us to news and market developments that may impact a borrower’s creditworthiness. Of course we don’t want the algorithm to actually take decisions; that will still be done by humans!”
Contact us
Follow us
