Many banking decision-makers still don’t think that machine learning technology, a subset of artificial intelligence, is worth the hassle. Or the money. Here’s why they’re wrong.
Let’s start with crunching some numbers, shall we? If Deloitte is right, the number of machine learning pilots and implementations doubled in 2018, and will double again by 2020. In other news, J.P. Morgan expects machine learning-driven platforms to grow at a 13% CAGR to reach a whopping $4.8 billion in value by 2021. Meaning that it will outperform the broader business intelligence and analytics market, which is predicted to grow at an 8% CAGR over the same period. When it comes to pricing: sure, getting into ML will come with a hefty price tag. But it looks like the majority of companies have realised by now that not investing in artificial intelligence and machine learning will cost them more in the long run. In 2018, more than 60% of organisations chose ML and AI as their key data initiative for the next year.
Machine learning may still sound like something from a distant future but it has actually been part of our everyday lives for a while. ML technologies, which apply algorithms to analyse datasets, spot patterns and create customer insights, have been rolled out and used by Google Maps, Netflix and Amazon product recommendations, to name but a few. And before anyone brings up the good old “alright, but they’re big tech not banks” argument, ML already has tried and tested applications in banking. Ajay Vij, Vice President responsible for European financial services at Infosys, lists repetitive work (e.g. back-office functions), high-accuracy tasks (e.g. loan underwriting) and even informed decision-making (like providing financial advice), for example.
5 ways banks can put self-learning algorithms to work
Risk assessment: ML tools can be a tremendous help for lenders with evaluating risk scores and making credit worthiness decisions more accurate when extending loans. They basically speed up the entire approval process. Take Norwegian digital bank Instabank, for example. They have teamed up with risk analytics provider Provenir to launch a cloud-based platform so they can make decisions on customer loan eligibility in a matter of seconds. The system mostly uses traditional information but it’s possible to even incorporate social media data into the assessment methods.
Customer service: Virtual banking assistants or chatbots use machine learning to better understand customers’ behaviour, track their spending and saving habits, and assist them in everyday transactions. What’s more, they can even give them tips on how to manage their finances better. Bank of America’s virtual assistant, Erica, is a prime example. Introduced in 2016 as a new feature of the mobile banking app, it makes use of predictive analytics and cognitive messaging to “provide proactive guidance” to customers and “anticipate their financial needs”.
Equity trading: Brokerages have actually been using ML methods for a while to make predictions for equity trading by completely automating the assessment of factors, like market volatility. But they’re just learning that ML tools can come in handy for a bunch of other things too: J.P. Morgan, for instance, launched a predictive recommendation engine in 2016 to zoom in on clients who should issue or sell equity. This application was later expanded to debt capital markets, similarly basing predictions on client financial data, issuance history and market activity.
Detecting fraud: Machine learning programs can flag up anomalous actions for near real-time fraud detection. They can spot patterns based on the past behaviour of account owners and generate alerts in case of suspicious transactions. That can be a huge competitive advantage for banks, considering that in 2017 identity fraud hit an all-time high with a staggering 16.7 million victims and $16.8 billion stolen in the US alone. According to Infosys, such predictive analytics can also be used as an anti-money laundering tool to trace the true source of money by identifying disguised illegal cash flow.
Regulatory compliance: Fintech companies are taking ML solutions to brand new territories. Hexanika, operating in the US and India, has developed a software application that uses ML-based processes to cut the burden of banks in statutory reporting on accumulated data. The tool automates data management and streamlines the compliance process, which are traditionally labour-intensive manual tasks. It also keeps up with all regulatory updates and can be easily connected to existing systems without architectural changes.
Save costs and increase productivity with machine learning
In his article, Infosys’ Ajay Vij points out that using ML tools can save training costs and enhance customer experience by offering personalised service at contact centres. It’s also a land of opportunities for cross- or up-selling new products through laser-targeted messaging. Machine learning automation can work wonders for productivity, too, by speeding up the completion of routine tasks and making the outcome more reliable. To top it all, ML can be programmed to follow protocols to better meet regulatory standards.
But truth be told, artificial intelligence or machine learning is not all blue sky. Here are a few pointers on what to consider before deploying ML at your organisation. First, it’s crucial to have a plan at hand for dealing with employees in manual jobs who are soon to be replaced by machine learning. You also need to pick the right strategy to avoid using ML for just routine applications with repetitive tasks. And remember to get all the protocols right when handling large datasets containing sensitive and protected customer data to prevent violations.
Lack of experience is a barrier
There are barriers stopping banks and other financial service providers from leveraging the full potential of machine learning. Let’s take a look at insurance companies, for instance. Most of them are well aware of the many advantages of machine learning in automation, productivity and cost saving. But only 12% do actually use these opportunities, according to a survey by analytics platform developer Earnix.
Why? The biggest problem is undoubtedly the severe knowledge gap: 82% of respondents think of themselves as beginners in machine learning. Another obstacle is the lack of experienced hires. With about 300,000 artificial intelligence professionals available for millions of vacancies worldwide, an AI talent crisis is well underway. Small wonder, considering that only a handful of universities teach courses in machine learning yet.
This post was originally published on November 16, 2017 and has been updated to include recent developments.