23 Sep 2019
Machine learning will change the way banking works – but the technology will only ever be as smart as the people who design, implement and maintain it.
Ahead of the Sibos 2019 financial services conference in London, ANZ institutional sat down with Adam Grant, Head of Trade & Supply Chain and Phil Cubbin, Head, Trade & Supply Chain Operations at ANZ to discuss how the increasing application of machine learning technology in financial services will impact the sector - and how the customer experience will be revolutionised as a result.
We’ve edited the highlights of the conversation below. They started by addressing the pace of change.
GRANT: Trade finance is very paper-oriented business and has been for a very long time. I would describe the pace of technology driven change around trade finance as rather modest, certainly over the past couple of decades.
"Machine learning is really just a term for something that's been around for years - coding a computer to take certain actions in certain scenarios.” – Adam Grant
However the emergence of blockchain and machine-learning technology has already revolutionised the way we look at trade finance globally, providing us with a unique opportunity to shift gears.
Certainly, at ANZ we have meaningful programs of work that leverage those two technology capabilities which impact a material component of our trade-finance business.
ANZ’s guarantee business represents a significant portion of our revenue line and a large portion of our operations capacity as well, by volume.
We're in live pilot phase on a blockchain-based platform that shifts those instruments from a paper-based to a fully digitised instrument. As a comparison, this could bring a similar impact we experienced through the introduction of internet banking for bank accounts.
CUBBIN: As we move towards a digital environment across all parts of the international trade supply chain, it is translating into opportunities to use of a range of different and emerging technologies.
Beyond blockchain are two other mediums. One is robotic processing automation (RPA) and the second is automation using artificial intelligence (AI), or machine learning, and ANZ is applying both within our operations.
The distinction between the two is RPA can be applied to structured data that appears in a document - in other words, a templated banking invoice with a consistent field which contains data in the same place each time that the robot can recognise and interpret.
With unstructured data, the invoice number may appear in a different place on different invoices. That’s where machine learning comes into play.
For example, if there's different data on a different part of the page or on a different page for every customer, a machine using OCR (optical character recognition) technology in tandem with artificial intelligence will take an image of the document and review it, identifying the piece of data you want.
And then on the next document if the machine finds the same data in a different place it will learn from that process. The time after that it will not just go to the same spot but look for the data throughout the whole document, in a similar way to how a human operator would.
GRANT: Machine learning is really just a term for something that's been around for years - coding a computer to take certain actions in certain scenarios. That's not a new concept.
While the sophistication in coding and software has advanced, the fundamental change that has allowed the progression of machine learning is the ability to ingest, store and reference that code against an enormous amount of unstructured data. That's fundamentally different.
The adoption and impact of distributed ledger technology (DLT) will take a while, as it needs a large-scale transformation of business processes, a technology shift from existing legacy applications, and more importantly we face the difficulty in bringing all parties of a transaction in to a unified platform.
Ultimately, these instruments will need to be fully digitised and DLT is the answer we have for now. That’s why at ANZ we're embracing DLT (as well as RPA and machine learning) in a number of circumstances. We're currently involved in DLT projects in Singapore and Hong Kong, as an example.
Until then, machine learning and deep learning will be a huge asset, particularly in operations, to efficiently process paper-based transactions into a digital format.
The Sibos financial services conference is here again.
Kicking off on September 23, one of the world's premiere international finance conferences will attract the most-innovative minds in the sector, setting the agenda for 2020 and offering insight into the trends and development which will shape the future of the industry.
In the lead up to and during the event ANZ Institutional will offer an insight into those themes, giving you a sneak peek at the ideas set to dominate the conference from ANZ’s industry experts and attendees themselves.
CUBBIN: Ultimately due to the emergence of a number of technologies the role of machine learning may change, as the products, use and flow of data also changes.
At the moment though, machine learning is starting to deliver value for us and our customers, with an improvement in processing times of up to 40 per cent, increased consistency of quality, and increased operational efficiencies. And as you said, we are just in pilot stage for blockchain.
GRANT: Yeah, I agree. The problem we have with robots is they aren’t intelligent enough to react to different scenarios unless they are taught how to handle it.
Ultimately the machine is only as smart as the rules that get coded into it and applied to the processing of paper documents, while blockchain provides new opportunities for products - and how they are used to change the world of trade finance.
Without the human, the subject matter expert, to code the rules that guide the actions of the machine, it is of no value to anyone. That’s your quality driver.
And obviously the amount of coding and maintenance on these robots should be kept at check. If there’s code written for basically every scenario, it may soon become counter-productive.
CUBBIN: It’s very similar to humans. You can teach a human the rules and teach them how to do something. Then when you give them the experience, their knowledge and capability increases - and quality and efficiency improves further.
At ANZ, when we started our first machine-learning use case - for sanctions screening - we were getting what's called ‘capture and accuracy rates’ in the mid, high 80s, in percentage terms. Over time, as we pumped more transactional data through that machine, these metrics increased as the machine ‘learned’. The difference is complimented with human-augmented workflow to ensure overall quality is maintained.
The machine capture and accuracy rate now for us is around 95 per cent. It’s almost perfect and the variance in performance is actually better than a human ‘maker’.
And for all the use cases we’ve implemented, the operational efficiencies are substantial. The machine could achieve operational efficiencies of up to 30 per cent in some cases, if not higher.
But we're not just doing it for that purpose. There are also several customer benefits from deploying machine learning, including from more consistent quality, as well as faster processing times. Additionally, the operational risk is also reduced and our ability to capture and use data is also enhanced through the automation.
Ultimately when we deploy all our use cases, customers will see an increasing improvement in turnaround times, allowing us to enhance our service proposition and gain a competitive advantage.
And our teams at ANZ are also embracing the use of new technologies, seeing the benefits and becoming ‘champions’ in identifying, and helping to develop, and deploy new machine learning use cases.
GRANT: I think that what's important to point out is that ANZ’s progress around this space is right at the very front of the global queue in certain cases. I think that would be fair to say.
But there are plenty of other banks who are working on this. Ours is probably still immature in the grand scheme of things but more mature than many others, including banks which are significantly larger trade banks than we are.
Shane White is content manager & Arun Kayal is AD, Communications, Institutional at ANZ
This article was originally published on ANZ's Institutional website
The views and opinions expressed in this communication are those of the author and may not necessarily state or reflect those of ANZ.
23 Sep 2019
20 Sep 2019