Innovative machine learning can be a boon for financial services, if they can get over the hype.
The “mystique” of machine learning will begin to wear off over the next two to three years, according to Caspian CEO Chris Brannigan. It will instead be replaced by conversation that focuses on the challenges that need solving and pave the way for a true next generation of opportunities to take shape for financial institutions.
For the past few years, machine learning and artificial intelligence have both been swamped with hype. Most financial institutions are either in the early stages of implementing the technology or exploring how it could revolutionise what they do. This has led to masses of solutions flooding the market, and yet there is still so much more to be achieved.
Financial institutions remain plagued by legacy systems and data silos that keep large chunks of key data locked out of reach to new technology. These setbacks are restraining machine learning solutions from the freedom they need to really make a difference, but they are not insurmountable and large strides are already being made to solve these issues. In a few years’ time, bank’s will have moved further towards cloud environments which in turn, will help bring more clarity to accessibility and privacy issues, as well as creating cleaner data, Brannigan said.
The excitement of machine learning has led to streams of companies building solutions around it. However, Brannigan believes a lot of current vendors are sometimes “trading on hype” to capitalise on the mystery around the technology. There are solutions on the market that may not do what they claim or indeed, are attempting to improve a process that will not benefit from optimisation.
He observes lots of banks who are still experimenting with machine learning based financial crime solutions, and some are getting burned or disillusioned as a result. However, in time Brannigan expects this to disperse, and open up the confidence required for financial services firms to really innovate with its capabilities. At first it is likely to be the odd solution and some big success case studies that come to the fore, but this will be the catalyst for innovation to accelerate.
Brannigan said, “in two years’ time, a lot of the pitfalls surrounding machine learning solutions will be very well known, and that mystique will be gone. At that point, we’ll hopefully just be talking about business requirements or simply about productivity or results. I think at that point, we’ll start to see some real change and innovation really starting to snowball.”
Comparisons can be made between the evolution of machine learning and the creation of the internet when we consider how vendors get to grips with their full capabilities. It took several years before banks truly began to understand web-based solutions and the new opportunities they could create for customers. It might not take as long for machine learning to come to fruition, given that the market is now more naturally tuned into the possibilities of technology innovation, but it will still take time.
Caspian is successfully showcasing how machine learning can be used to bring material change to the way financial institutions manage high volume, complex risk investigations. The company’s Financial Investigation Platform (FIP) automates the process of gathering and analysing alert data, before producing overall risk decisions that are fully evidenced and explained to investigation analysts.
This takes technology beyond just alert generation to support the point in the process at which a human analyst would ordinarily have to perform an investigation. Caspian achieved this by first building a model of how expert investigators think and act, then building a system to replicate it.
“We have spent the past 24 months working with banking partners and financial crime experts to observe, capture and test how expert investigators gather evidence, judge and evaluate evidence, make risk decisions and then explain those decisions. The resulting cognitive map is a detailed blueprint that means the machine can perform the wide range of cognitive tasks that an expert investigator would perform whilst evidencing the rationale against the standard of expert human consensus.”
Brannigan highlights why machine learning is crucial for tasks like financial investigations. Beyond the obvious speed benefit of having a computer assess hundreds of documents, a machine is much more reliable than a human is for judgements. “Humans come to conclusions in many different ways and so there’s a huge disparity in quality – [For example] if I was to read something and investigate it versus you, the analysis or outcome would likely be different. Whereas, a machine will always give you a very reliable and very consistent answer, that can constantly be improved upon. This means you can accurately measure risk at a very granular level and as new scenarios or vulnerabilities are found, or criminals invent new ones, then the machine can very quickly be updated and trained to best practice again.”
Earlier in the year, the RegTech company rolled out its Transaction Investigator solution to a global tier one bank. The tool gives banks the ability to automate the investigation of alerted anti money laundering cases and the associated alerted transactions. Fully automating or significantly augmenting this process ensures that banks can get more accurate decisions at a quicker rate. Instead of needing human analysts to spend hours sifting through buckets of data, often stored in separate locations, the technology can do it in seconds and advise analysts of any further required steps. The solution is also trained on bad information so it can generate judgements based on ambiguous data and then explain why it came to that decision.
Machine learning is not infallible though, and a big challenge still surrounds the transparency of many solutions. Human analysts are systematically asked to explain how they came to their conclusions to satisfy regulatory and audit requirements. Any automated system being utilised to identify incidents of fraud or financial crime and make serious judgments also needs to be fully auditable. Outputs of such solutions need to be clear-cut with the routes and evidence to support judgements being fully explainable in a human friendly manner. Asking a probabilistic or non-linear machine learning system to reproduce something can be difficult if the models are not created with this capability. And this is where some banks are being misled or ill advised with regard machine learning systems.
“These solutions need to be able to trace the data that was used to train it, right the way through to production and making predictions in the real world. These are not necessarily things that machine learning vendors build from the outset. Often banks may not be aware of this consideration until it’s too late and inadequacies are highlighted through independent Model Reviews or similar.”
Brannigan cites that “the validation and model maintenance required in financial crime compliance is built into the Caspian DNA and that model-risk management can actually be turned from a major barrier into a significant enabler of AI. By providing risk executives the ability to determine how a machine investigated and made a risk judgement, we are empowering them with insights that are not possible to confidently diagnose when it comes to human cognitive judgements.”
The market is littered with money laundering fines. The most notorious in recent memory was the €200bn Danske Bank scandal. Money laundering looks to be a scandal that will not quickly disappear and research from Encompass Corporation found that a total of $352.5m in fines were handed out globally for AML failures in just the past four months.
Brannigan believes the reason there are so many issues with money laundering is because policies banks deploy in their rule-based systems are founded on regulations that were put in place in 2008/9 to cope with situations relevant at the time. They have since become antiquated and yet firms are still trying to add to existing systems to meet compliance with various new regulations, instead of implementing a new system that makes use of better technology.
He said, “The banks are having to spend enormous amounts on building extremely complicated and multi-layered systems that rely on high numbers of humans to carry out often low levels of administration to effectively comply with regulations, as opposed to, a system where that maximises the higher order capability of humans alongside technology that automates or significantly augments the process of identifying financial crime.”
Brannigan concludes that “A lot of bankers I talk to are really keen to move towards these intelligence-led operations that utilise more expert humans and technology to drive improvements. But the regulatory world is still very much in a compliance mode.”
Copyright © 2019 FinTech Global
