One of the hottest technologies is machine learning in banking. Many of us experience the outcomes of machine learning every day. For example, Netflix and Amazon use algorithms to give us a movie or shopping recommendation based on our prior behaviors. CitiBank utilizes machine learning to evaluate “big data” to prevent fraud and monitor potential threats to customers. Even the United States Postal Service performs character recognition of handwritten characters using an algorithm and a computer vision system behind it.
Machine learning is related to Artificial Intelligence (AI), which is computational models of human behavior and thought processes (thanks to MIT for the definition). These models (aka algorithms) learn how we behave in certain situations and replicate those behaviors, so computers act the same way.
“Artificial intelligence (AI) essentially refers to a machine taking over a task that a human normally performs, such as extracting pertinent information from documents. AI used in this way can save up to 90% of cycle time and can also root out human error, but it is more challenging to train computers to complete tasks that require common sense. The real value of AI is to augment human effort, enabling elevated human expertise through more valuable time spend.” – Thomson Reuters – What FinTech Trends Are Experts Seeing?
There’s a tremendous opportunity cost in spending time on massive amounts of repetitive tasks that can be easily outsourced by process automation. Not only can tedious work be done faster with software, but advances in machine learning can nearly eliminate the risk of human error. This all, in turn, frees up employees to spend time on more fulfilling and customer-focused projects. According to Forrester’s Predictions 2018: Automation Alters the Global Workforce, companies that can master automation to “squeeze performance and insights out of previously commodity operations” will dominate their industries.
Advances in artificial Intelligence in banking and machine learning in banking are helping commercial lenders differentiate themselves by:
- Identifying bottlenecks in their operation workflows and bring in significant improvements in process efficiencies and efficiency ratios
- Compensating for a shortage of talent and hiring budgets devoted to regulatory compliance
- Reducing errors and risk
- Using repetitive process automation to free employees to focus on skill-based work and more engaging experiences for customers
- Predict outcomes (see the machine learning in banking use cases below)
Machine learning in banking case studies
Based on research conducted by Tech Emergence, below are illustrative machine learning in banking use cases.
The Bank of New York Mellon Corp experienced an 88 percent improvement in processing time thanks to over 220 AI digital robots programmed to process automated tasks that previously required humans to sift through mounds of data manually. They’ve been able to cut down response time to 24 hours from 6 to 10 business days. Examples include “data requests from external auditors” and “funds transfer bots” which help “correct formatting and data mistakes in requests for dollar funds transfers.”
“One of the objectives of BNY Mellon’s bot strategy is to help the organization “get rid of the mind numbing tasks” so that employees could focus on different activities.” – Doug Shulman, EVP, BNY Mellon
IBM’s Watson uses machine learning to assist banks with compliance. For example, rather than manually looking up each regulation, Watson scans a document to identify areas where regulation applies. The recommendations support the work of a compliance officer by providing insights and recommendations that a compliance officer may have missed on her own.
J.P. Morgan Chase discovered that nearly 80% of loan servicing errors are due to contract interpretation errors. They began using machine learning to analyze documents and extract critical data points and clauses. What used to require thousands of hours manually reviewing and transcribing documents can now be done in seconds.
“Manual review of 12,000 annual commercial credit agreements normally requires approximately 360,000 hours per year. Results from an initial implementation of this machine learning technology showed that the same amount of agreements could be reviewed in seconds.” TechEmergence
Use Case: How does machine learning help with construction loan management?
In this use case, imagine a commercial construction lender with a $500M portfolio. They manage dozens of loans and the deluge of administrative tasks related to draw requests. For example, within a single draw disbursement request submitted to the bank, a general contractor pulls together 100’s of legal and compliance-related documents including lien releases, invoices, the AIA G702, the AIA G703, receipts, change orders, inspection reports, approvals, and more. Each draw request requires the approval of several parties including but not limited to the loan administrator, third party inspector, title company, and other loan participants.
Historically, due to flexibility and familiarity, Excel has been the construction loan administration tool of choice for years. Meanwhile, recognition of potential errors and manual overload have often been overlooked.
The problem is that, in addition to an abundance of manual work, Excel does not aggregate project data and notify lenders of potential risks based on historical information. With an excess of documentation flying around in PDFs, emails, and Excel spreadsheets, it is often difficult to know precisely where an approval, loan, lien release or draw is in the process.
With the volume of work involved in administering their portfolio of loans, the example bank embraced the benefits of modern machine learning, process automation, and construction loan software. Through a combination of optical character recognition, computer vision, machine learning algorithms, and rule-based predictive modeling, loan management software parses the information from the countless emails and PDFs included within a construction loan draw request. It analyzes the data and performs “searches” to identify errors. Spreadsheets and forms capture the clean data, and the software creates recommendation reports based on rules and the data.
A process that had taken hours of human interpretation is now done in seconds. With all of their loan information in one location, The First Federal construction loan administrators now quickly run their project and portfolio level reports like loan composition, cash flow projections, draw processing, and missing lien releases. Moreover, they have access to real-time information about payment status. This visibility allows both First Federal and their borrowers to see who is being paid, and when.
The outcomes have led to more profitable loans, faster disbursements for borrowers and their contractors, fewer mechanic’s liens over non-payment disputes, and an opportunity for employees to spend time and effort on things that provide a greater sense of purpose and profits for the bank.
By aggregating and parsing the information from the countless emails and PDFs, machine learning and construction draw software has helped First Federal quickly answer questions:
“Is this invoice for this project?”
“Has this invoice been invoiced before?”
“Is the invoice within the contract amount?”
“Is the project within budget?”
“Does the invoice have the correct documentation?”
“Is the project on schedule?”
Other examples of how construction loan software has assisted the bank with the monthly draw request process include:
- The borrower, lender, inspector, architect, etc. all quickly track approvals and documents related to the monthly draw process. Coordinating and collecting the approvals have become much more efficient and approvals can now take place in the field by way of mobile applications.
- The use of a single software platform prevents data entry duplication and manual entry like contractors transferring invoices from their subcontractors, developers transferring invoices from their contractors, and lenders transferring information from their borrowers.
- First Federal is distributing the appropriate amount of funds as quickly as possible while always ensuring that enough funds are remaining to complete the project and that lien releases are tied to fund disbursements.
- The bank not only tracks draw disbursements but also monitors the required interest to be billed each month and withholds those funds from the loan amount.
Conclusion
American Banker published their 2018 predictions for how the leading trends in FinTech will impact banking. They report that an essential focus for the banking industry is to develop their AI capabilities in critical areas including operational efficiency and risk mitigation. By upgrading their technology and enabling paperless processing, commercial lenders can compete for clients beyond the footprint of their branches.
It’s an exciting time for the banking industry, and the impact of innovation is just taking root. The applications for machine learning in banking are benefiting some of the most prominent banks in the world, as well as banks without billion dollars in IT budgets.
We’re here to help you navigate this journey.
Jump To
