According to Fortune, JPMorgan Chase is the largest bank in the U.S. and controls over $2 trillion in total assets. In this article, we detail the types of AI research JPMorgan is doing as well as how they are likely to be using their applied AI applications.
JP Morgan’s AI research initiatives include:
- Natural Language Processing in Equity Investing
The bank’s applied AI initiatives include:
- Anomaly Detection for Recognizing Fraud and Risk Mitigation
- Natural Language Processing (NLP) for Virtual Assistants, Utilizing News, and Client Intelligence
- Predictive Analytics for Smart Documents Intelligent Pricing
-
We begin our exploration of JP Morgan’s use of AI with their research initiatives:
AI Research Initiatives
Natural Language Processing in Equity Investing
JPMorgan and APG Asset Management, a dutch pension manager, have worked together to study how data science can improve the ways in which portfolio managers and research analysts take in and utilize information.
They used a dataset consisting of bank statements from the European Central Bank to run tests that would elucidate the implications and limitations of data science. JPMorgan is now purportedly using their findings to explore how they can be used in separate initiatives for their trading and research teams.
Researchers at JPMorgan claim they have found positive results within a relatively short time, and have been able to pinpoint multiple use-cases for this technology in developing new user interfaces. Individual employees, likely from both JPMorgan and APG, collaborated through a single data repository that allowed everyone working to observe each other’s progress.
The team was purportedly able to test out various research approaches from within the company’s databases and those from outside entities. They also built a prototype AI application which marked the end of the research project.
Augmenting the experience of important staff, such as portfolio managers, may prove to be an important part of AI transformation within banks. Similar to the customer experience, reducing friction for end users could improve satisfaction and efficiency across many departments.
We spoke with Sebastien de Brouwer, Chief Policy Officer of the European Banking Federation, about this on our podcast, AI in Banking. The interview was focused on where business leaders should be focused in terms of AI. When asked about which capabilities will matter in terms of being critical in the future, de Brouwer said:
We strongly believe that AI will have indeed a transformative effect on the banking industry…The most important aspect is certainly that it will change and hopefully enhance the customer experience. So that’s already a very important element I think for the banks who will succeed, and that is of course interactivity with the clients, because this should allow [banks to use interactivity data to create better offerings]…One activity where many banks are looking at is investment advisory or recommendations. I think that is certainly an area where no big players are looking very seriously at AI [as a solution.]…This may also expand the client segments that would have access to those kinds of services.
As de Brouwer observed, investment advisory and recommendations are business areas that stand to benefit from data science and AI-enabled business intelligence. If JPMorgan were to implement a way to use AI to not only help their staff but their customers understand what the most important information means for their stocks, it may raise their customer satisfaction in the future.
For example, consider if a customer’s portfolio manager was using an AI-enabled dashboard to view the portfolio and its associated data. JPMorgan could create an interface that displays the most important market trends for the manager to show to their client.
This way, the customer may develop a better understanding of their investment situation as well as a better relationship with the bank’s brand.
For JPMorgan, this research project proved that their co-creation model with APG is valid, which may imply success with this project. Whether this means JPMorgan will be adopting this AI application in some form is unclear. While the team found mostly business information, they claim to be continuing to use this project as a way to engage with the greater data science community.
Applied AI Initiatives
JPMorgan claims to be taking advantage of six applied AI initiatives. That said, the company does not provide much detail about how these initiatives are implemented or in which departments. In order to implement six separate AI initiatives, JPMorgan would need a streamlined method of integrating AI software across their tech stack. This may be possible with a cloud-based system similar to what they use for treasury management.
JPMorgan’s purported applied AI initiatives include:
- Virtual Assistants
- Anomaly Detection
- News Analytics
- Quantitative Client Intelligence
- Smart Documents
- Intelligent Pricing
We can infer numerous possible use-cases for each of these AI initiatives, in addition to the ways in which they were developed and implemented. The clearest use case is for their virtual assistant, which is likely a mature version of the pilot chatbot project they worked on with AI firm Kasisto in 2018 and earlier this year.
Kasisto claims to have helped JPMorgan treasury services offer customers a customer service chatbot. The chatbot was purportedly made for the purpose of helping clients navigate JPMorgan’s expansive website.
Kasisto’s platform, KAI, could be used to develop chatbots which can be deployed across multiple digital channels, such as employee dashboards and smartphone apps. Kasisto claims to have a deep learning tool for business banking that helps their software analyze data.
Chatbots made using Kasisto’s platform are intended to communicate with customers about their financial operations. These can include loan or credit card applications, product discovery, or simply customer support. Some customer requests can be fulfilled within the conversational interface, including applications or sending payments.
Kasisto claims KAI chatbots can recommend users more efficient ways to accomplish routine tasks. One example of this is a banking chatbot that recommends foreign-exchange ACH payments in lieu of numerous repeated payments of small amounts.
The demo below shows how a KAI chatbot can answer questions regarding personal data such as account balances. This particular video shows the chatbot Kasisto made for Mastercard:
Anomaly Detection
JPMorgan most likely uses anomaly detection to mitigate risk and identify fraudulent banking activities. This type of application would require the company to integrate a machine learning model into their payment processing stack and allow it to analyze a continuous stream of incoming transactions.
This would train the model to recognize a baseline sense of normalcy for the contents of these transactions or new account information should the system be used to track any other banking tasks.
An anomaly detection solution could then notify a human employee of any action which deviates from the normal pattern so they can review it. This employee can choose to accept or reject this alert, which serves as a signal to the machine learning model that its determination is correct or incorrect, respectively.
This helps to train the model to recognize that the type of deviation it detected was either fraud or an acceptable deviation from typical operations. This helps catch money laundering as it happens and may allow JPMorgan’s fraud team to stop the attack and reverse any fraudulent transactions.
Intelligent Pricing and Quantitative Client Intelligence
Both JPMorgan’s pricing and client intelligence applications are likely powered by predictive analytics. This is because in order to make accurate predictions about prices and various customer service needs based on large amounts of enterprise data, the company would need a solution that can properly understand and find trends within that data. JPMorgan would likely be able to gauge potential customer satisfaction based on proposed business plans; however, the company makes the exact use case for this initiative unclear.
JPMorgan’s website states that their intelligent pricing initiative offers “more accurate prediction and confidence intervals.” This is a strong indicator that their solution runs on predictive analytics because predictive analytics in many cases also uses confidence intervals to determine the best prediction to provide to the user.
Additionally, their client intelligence initiative would require predictive analytics at least in part in order to be able to determine if a customer was satisfied using their customer service data.
In order to make this type of solution run properly and intake large amounts of customer service data, it would need to be installed across all customer service channels within the bank’s tech stack. We spoke with Nishant Chandra, Sr. Director of Data Products at VISA on AI in Banking about how banking leaders can better handle this type of challenge that comes with AI adoption.
Chandra emphasizes the importance of integrating AI capabilities as opposed to simply layering them on top of current operations. He makes a comparison between this type of layered integration, or in his words, “lasagna,” as opposed to the vertically topped “pizza.” With regards to his specific approach to integration, Chandra said:
Each layer of software when they are talking to each other are intelligent. They have data science capabilities built in, they have Ai intelligent ways of detect data fraud built in at every layer as opposed to doing it at the very end. These are intelligent software platforms which will transform or ingrain the AI capabilities in this space. This is a fundamental transformation that is happening.
Layering AI applications into each customer service channel as Chandra recommends may be important for banks, AI startups, or possibly financial institutions that are currently developing solutions that need to be integrated into multiple channels such as JPMorgan’s intelligent pricing solution. However, it is unclear at this time if JPMorgan has taken this approach.
It is likely that this solution also includes NLP because JPMorgan claims this initiative draws from client communications to quantify satisfaction. In order to do that, it would need to be able to parse typed, written, or spoken language and translate that into numerical data.
News Analytics and Smart Documents
Much like their client intelligence initiative, JPMorgan’s news analytics solution may be comprised of both NLP and predictive analytics technologies. This is because they claim the initiative allows them to collect news from multiple sources and then analyze them for sentiment, specific topics, and trading signals.
While analysis of trading signals would require predictive analytics to provide accurate predictions from news data such as articles or blog posts, sentiment analysis would also require NLP.
In order to parse written words and subsequently evaluate the sentiment behind them, a machine learning model would need to be trained on the language they are written in, the business terms present within them, and context. The model must also be trained on sets of individual words labeled according to their usual sentiment.
The sentiments are usually categorized simply as positive or negative, and then these labels are attached to individual words such as “efficient,” or “problematic,” respectively. This allows an NLP application to determine the topic of each paragraph or article as well as how the speaker feels about that topic or how the reader might be supposed to feel.
The company’s smart documents initiative is likely impossible without an NLP application to make it work. JPMorgan states that this project allows them to find important information from lengthy text documents in order to allow their staff to focus on projects where humans are more required.
They claim this improves workflow by reducing manual operations, and because of that this solution likely routes the important information to designated project manners or any employee who requests it. It is unclear exactly which types of documents this initiative is used for, but it would be possible to implement in any business department that requires employees to search for relevant information within long financial documents.
Header Image Credit: The Block
