JP Morgan: Expanding Machine Learning Capabilities in the Financial Services Arena
Brief overview of JP Morgan's adoption of machine learning into its day-to-day operations
Given high transaction volume, accurate historical records, and the quantitative nature of the financial ecosystem, few industries are better poised to reap the process improvement rewards associated with machine learning (“ML”) than the financial services industry. JP Morgan (“JPM”), arguably the largest financial institution in the United States, has made a strategic decision to incorporate ML into their operations in order to sustain competitive advantages in today’s historically competitive financial services market.
Representing half of the Firm’s 2017 net revenue,[1] JPM’s lending business can benefit tremendously from the process improvement associated with ML. Fiercely competitive dynamics in the lending market, one in which consumers have come to expect the immediacy they find in the online retail sector and in which regulators have established “zero-tolerance” policies for compliance mistakes, have led to industry-wide adoption of ML software used to streamline loan application processes, reducing the lengthy time commitment and paperwork gathering of the past.[2] ML software can also be utilized to meet lofty compliance standards set forth by lending regulators, primarily in helping to produce accurate financial reports and expand the scope of stress testing and risk monitoring.[3]
JPM also has the potential to recognize meaningful benefits from ML implementation across the rest of its day-to-day operations. By 2030, the financial services industry is expected to recognize a 22% reduction in operating expenses (approximately $1-1.3 trillion) resulting directly from ML adoption.[4] Use-case examples include incorporating ML into compliance and anti-money laundering functions (expected to save nearly $220 billion).[5] Lastly, and perhaps of most long-term importance, ML processes data to suggest the optimal financial solution for an individual, thus providing firms with the ability to better understand customer needs and forecast the demand of financial services, both without human intervention.[6]
In 2017, JPM’s Quantitative Investing and Derivates Strategy Team issued a 280-page report highlighting how the Firm’s forward-thinking management team has made cautious adoption of ML a short and medium-term priority.[7] Highlighted in the report was JPM’s recent incorporation of ML software which trains algorithms on millions of samples of consumer data (age, job, habits, etc.) and financial results (credit history, insurance status, etc.) to make better lending decisions.[8] JPM has also allowed ML software to trade securities on their investment management platform: “Machines have the ability to quickly analyze news feeds and tweets, process earnings statements, scrape websites, and trade on these instantaneously,” says Marko Kolanovic, JPM’s Head of Quantitative and Derivatives Research. “This will help erode demand for fundamental analysts, equity long-short managers and macro investors.”[9]
JPM recently launched a “Contract Intelligence” platform that leverages Natural Language Processing, a common ML technique.[10] The solution helps to improve back office functionality and reduce administrative expenses by processing legal documents and extracting essential data.[11] Manual review of 12,000 commercial credit agreements would typically occupy approximately 360,000 labor hours; ML allows for the same review in a just a few hours.[12] Aside from applying ML to its current business processes, JPM has improved its chances of long-term success in ML adoption by creating a corporate culture and building a workforce that is supremely focused on maximizing the ML opportunity. The report highlights JPM’s hiring of some of the world’s top data scientists and further willingness to “hire an army of people to acquire, clean, and assess data.”[13]
While JPM has established itself as an ML thought leader, hurdles remain in ensuring that the sizable opportunity is maximized. First, as most consumers and regulators remain wary of ML applications, particularly in financial services, JPM must build and incorporate its ML capabilities with the upmost transparency to secure market trust.[14] Much of this trust will be established with continuous operational improvement and enhanced customer experience, but JPM must be able to quickly explain the “black box” to optimize the impact of its major strategic initiative, a daunting challenge given that the technological capabilities of ML are limited in the sense that ML software cannot easily explain the reasoning behind its decisions. JPM must continue to invest significant time and resources to combat both this existing reputation of ML in the market and the other inherent limitation of adopting the technology: Although data is being created at an accelerated pace and the robust computing power needed to efficiently process the data is available, most massive data sets are not simple or financially feasible to create.[15] Investment of today’s resources must continue to make the adoption of ML easier and more financially appealing over the medium and long-term.
The most significant looming questions that merit comment from the HBS student body revolve around wide-spread adoption of ML in the financial services industry. How will adoption of ML impact the labor force currently in-place? Will adoption of the technology displace millions of employees or will it grant employees increased availability to work on more meaningful tasks? (800 words)
[1] JP Morgan, “2017 Annual Report,” JPMorganchase.com, April 5, 2018. https://www.jpmorganchase.com/corporate/investor-relations/document/annualreport-2017.pdf, Accessed November 2018.
[2] Price Waterhouse Coopers, “Leveraging Robotic Process Automation in Mortgage Lending,” PWC.com, March 1, 2017. https://www.pwc.com/us/en/consumer-finance/publications/assets/pwc-rpa-robotic-digital-labor-mortgage-lending.pdf, Accessed November 2018.
[3] Ty Kiisel, “7 Ways Automation Improves the Bank Lending Process,” Lendio.com, May 13, 2013. https://www.lendio.com/blog/small-business-tools/automation-improves-lending/, Accessed November 2018.
[4] Lisa Joyce, “Artificial Intelligence and The Banking Industry’s $1 Trillion Opportunity,” thefinancialbrand.com, May 29, 2018. https://thefinancialbrand.com/72653/artificial-intelligence-trends-banking-industry/, Accessed November 2018.
[5] Lisa Joyce, “Artificial Intelligence and The Banking Industry’s $1 Trillion Opportunity,” thefinancialbrand.com, May 29, 2018. https://thefinancialbrand.com/72653/artificial-intelligence-trends-banking-industry/, Accessed November 2018.
[6] Lisa Joyce, “Artificial Intelligence and The Banking Industry’s $1 Trillion Opportunity,” thefinancialbrand.com, May 29, 2018. https://thefinancialbrand.com/72653/artificial-intelligence-trends-banking-industry/, Accessed November 2018.
[7] Vincent Granville, “J.P. Morgan’s Comprehensive Guide on Machine Learning,” uci.edu, November 21, 2017. http://faculty.sites.uci.edu/pjorion/files/2018/05/JPM-2017-Summary.pdf, Accessed November 2018.
[8] Vincent Granville, “J.P. Morgan’s Comprehensive Guide on Machine Learning,” uci.edu, November 21, 2017. http://faculty.sites.uci.edu/pjorion/files/2018/05/JPM-2017-Summary.pdf, Accessed November 2018.
[9] Vincent Granville, “J.P. Morgan’s Comprehensive Guide on Machine Learning,” uci.edu, November 21, 2017. http://faculty.sites.uci.edu/pjorion/files/2018/05/JPM-2017-Summary.pdf, Accessed November 2018.
[10] Konstantin Didur, “Machine learning in finance: Why, what & how,” towardsdatascience.com, July 11, 2018. https://towardsdatascience.com/machine-learning-in-finance-why-what-how-d524a2357b56, Accessed November 2018.
[11] Konstantin Didur, “Machine learning in finance: Why, what & how,” towardsdatascience.com, July 11, 2018. https://towardsdatascience.com/machine-learning-in-finance-why-what-how-d524a2357b56, Accessed November 2018.
[12] Konstantin Didur, “Machine learning in finance: Why, what & how,” towardsdatascience.com, July 11, 2018. https://towardsdatascience.com/machine-learning-in-finance-why-what-how-d524a2357b56, Accessed November 2018.
[13] Vincent Granville, “J.P. Morgan’s Comprehensive Guide on Machine Learning,” uci.edu, November 21, 2017. http://faculty.sites.uci.edu/pjorion/files/2018/05/JPM-2017-Summary.pdf, Accessed November 2018.
[14] Lisa Joyce, “Artificial Intelligence and The Banking Industry’s $1 Trillion Opportunity,” thefinancialbrand.com, May 29, 2018. https://thefinancialbrand.com/72653/artificial-intelligence-trends-banking-industry/, Accessed November 2018.
[15] Konstantin Didur, “Machine learning in finance: Why, what & how,” towardsdatascience.com, July 11, 2018. https://towardsdatascience.com/machine-learning-in-finance-why-what-how-d524a2357b56, Accessed November 2018.
Hey Mike,
I really enjoyed your post. Specifically, good job referencing market statistics and stakeholder quotes. I wrote my essay on the ML opportunity at Goldman so I see a lot of similarities in our findings. Two areas that I am still struggling with are the following:
1 – Do Goldman, JPM and the like have legal rights to their clients’ underwriting data? Yes, this is a perfect use case for ML, but so far I dont see any real-life success stories.
2 – Because of the inherent difficulty of monetizing B2B data (institutions tend to actuall look at data privacy agreements), I see the universal banks doing a lot more with ML in the consumer / retail segment. Did you get a similar takeaway?
Mike,
I enjoyed reading your post. I have seen an increasing amount of news coverage about using ML to aid lending decisions. The core of each company’s algorithm is the access to data sets that are relevant to the borrower’s willingness and ability to pay. Some internet companies are starting to use personal data (including chat data, what kind of news someone read) in developing the algorithm, which I believe has gone too far and will eventually run into a regulatory wall. By then, the government is likely to put a stop on using personal data acquired to make credit review process easier.
The projection I outlined above answers part of the question you posted at the end of the article. Using ML for lending decision or credit review is more of a hype and hope, which is complicated by the regulatory barriers. Therefore, the risk of replacing labor in my opinion is possible but remote.
I wonder whether banks like JPM have been able to quantify the new risks that arise from the adoption of ML. If ML algorithms are making trading decisions, do the humans who designed the algorithms have sufficient control to stop the system from making large and risky trades? I worry that broader adoption of algorithmic trading at banks and hedge funds creates new systemic risks and undermines our ability to understand how the market behaves. What happens when all algorithms across multiple firms start to behave in the same way due to a convergence of specific signals?
I agree that ML will radically transform the financial services industry, but I’m not equally excited about all potential applications. Using ML to reduce compliance costs and legal paperwork is probably a good starting point. ML-based trading and lending probably requires a more cautious approach.
Very interesting piece – thanks for the research. Machine learning also has large implications for other operations within financial services, including risk management. It will be interesting to see the impact of machine learning on the development and usage of models for risk management, including pricing models, liquidity models and capital adequacy planning models.
It is great to see big banks innovating–they must in order to stay competitive! One thing that I think JPM should think about as they delve deeper into ML is biases. Particularly because of the impact that financial services can have on the lives of consumers, it’s imperative for JPM (and its competitors) to make sure that their models are promoting, not prohibiting, financial inclusion; biased data is an easy way for these models to swiftly fail the millions of American families that cannot access the banking system. I like the idea of using ML to better customize products. Imagine having product terms that perfectly fit your needs at a smart price, all predicted by ML models. This is the type of product customization that could help big banks stay competitive as fintechs enter the market. Larger, established financial institutions have a greater amount of data to leverage to make better, predictive ML algorithms but, due to their size, they may be under greater scrutiny from regulators. Helping regulators understand “what’s inside the big black box” is just as important as helping customers understand the method behind certain product decisions. Great article!
Really great post and very informative. The “Contract Intelligence” initiative is a real time saver. There is so much time spent by backoffice/operations employees going over contracts that do not change in content by a whole lot. Appreciate the stats on the man hours saved by using ML technology to extract pertinent information from contracts. I can see law firms harnessing type of ML technology to reduce the amount of time associates / paralegals spent reading through contracts.