Decision-makers in the banking sector have a unique set of business intelligence needs, and artificial intelligence has been on the radar of banking executives for several years now. It follows that AI and machine learning would find their way into business intelligence applications for the banking sector.
In this report, we discuss AI-driven business intelligence software for the banking sector and their evidence of success. As it turns out, there isn’t much. The companies covered in this report all lack banking case studies for their business intelligence software, which may be indicative of the nascency of AI for this particular application in banking. That said, these companies do offer case studies for other sectors.
- Report Generation – creating reports with a variety of visualizations that suit the needs of employees at different departments
- Predictive Analytics – correlating enterprise data to find patterns on which executives can take action
In this report, we’ll discuss the uses cases of AI-driven business intelligence applications in the banking sector by taking a look at four vendors that offer AI software to banks. We’ll begin by examining the vendors offering report generation-related AI solutions:
SAS offers software called SAS Visual Analytics, which it claims can help banks provide their lead staff with self-service analytics and interactive reports using what appears to be a combination of predictive analytics and natural language processing. For banks, this analysis and reporting may be related to customer buying patterns, loan payments, or customer experience.
This could be for strategies on gaining customers or to finding customers less likely to default on loans. SAS claims that in addition to the data analytics capabilities of its solution, it can also discern the sentiment behind text data, such as social media posts, and mark it as positive or negative. This is likely accomplished through natural language processing.
Below is an 8-minute video demonstrating how SAS Visual Analytics can segment customers. In this video, the demonstrator is able to segment the section of his customer base by 4:40:
We can infer the machine learning model behind the predictive analytics portion of the software needs to be trained on the client’s data related to banking transactions, customer profiles, and geolocation. For example, if a customer made an ATM withdrawal, the relevant data from that transaction would be the customer’s age, what gender they are, and where the ATM is relative to their bank. Customer segments could become more specific than this, however. The data would then be run through the software’s machine learning algorithm.
This would have trained the algorithm to discern which data points correlate to changes in behavior or new patterns forming within a bank’s customer base. This could be across the entire customer base or focused into a segment of customers based on age, gender, or geolocation.
The software would then be able to predict, report on, and visualize these changes or patterns. As a result, business leaders can provide organized information to other departments. Users can create various types of graphs and charts to depict the data in a more organized fashion, and these models can be shared between staff through the software. This may or may not require the user to upload information about their future plans or banking protocol into the software beforehand.
If the SAS Visual Analytics solution does leverage natural language processing to analyze text data, it would have to be trained on thousands of social media posts or other text data samples, such as emails. This data would involve customer experiences in financial aspects such as banking and loans. The data would be labeled as a negative or positive experience.
SAS would then expose the machine learning model to the labeled text data. This would train the algorithm to discern the chains of text that, to the human brain, might be interpreted as a positive or negative sentiment as conveyed by a social media post.
SAS Visual Analytics would then be able to run through public social media posts or any other text data in need of sentiment analysis, and the software’s algorithm would be able to determine if the customer’s experience was positive or negative. This may require the users to upload data about their future changes to banking operations into the software beforehand.
It’s likely that users in banking can integrate the software into existing databases for banking and customer information, as well as systems that would allow the transfer of text data from social media.
SAS does not make available any case studies showing a bank or financial institution’s success with their software, but it lists Royal Bank of Scotland, Cosmos Bank, and Erste Bank Croatia as some of their past clients.
Jim Goodnight is CEO at SAS. He holds a PhD in Statistics from North Carolina State University. Goodnight has spent the last 42 years of his career at SAS.
Thoughtspot offers software called Thoughtspot Embedded Analytics, which it claims can help banks and financial institutions easily search through large stores of data. A user can purportedly start typing in a search term into the software’s search function, such as “New York,” and the search bar creates a drop-down menu of possible insights the user may want to generate a graph out of.
The software also purportedly allows users to type questions into the search bar using everyday language, which likely indicates that the system uses natural language processing. The company claims that the insights its software generates for users are powered by machine learning. Generating correlations out of large volumes of data is a classic use case for machine learning.
The software starts generating a graph before the user has selected the kind of visualization that they prefer at the moment. It is unclear if the graph is generated based on the user’s own history of chosen visualizations or if there is a default of some sort.
The video below demonstrates the software’s search function and shows how graphs are generated as the user types:
For example, a user might want to figure out if a certain age demographic, say millennials, were more or less likely to default on their loans within a given location, say New York. They could type relevant search terms (millennials, New York, loan default) into the search function and be presented with several options for visualizing several different correlations, including one between millennials, New York, and loan default status.
Thoughtspot claims banks can integrate their stores of data into Thoughtspot within a relatively short timeframe in comparison to other data visualization tools, although they do not present a case study to verify this claim.
Thoughtspot does not make available any case studies that show a bank’s success with the software, but they list Scotiabank and Sterling National Bank as some of their past clients.
Amit Prakash is co-founder and CTO at Thoughtspot. He holds a PhD in Computer Engineering from The University of Texas at Austin. Previously, Prakash served as Software Engineer at Microsoft.
Aithent offers software called Aithent Analytics for Business Intelligence (AABI), which it claims can help banks and financial institutions transform data from multiple and disparate sources into actionable knowledge for their management using predictive analytics. Such knowledge involves profit and loss, return on investment, and customer information.
The software is reported to create templates from detected patterns to help business leaders in their decision making. The templates serve as visualizations of current trends, which is helpful in communicating technical changes to other departments and for planning ahead. Aithent claims their software’s reporting and templating of data can be scaled widely or acutely to find insights for more specific business areas.
In addition, they advertise the ability for users to generate custom alerts for patterns they want to monitor closely. With this feature, users can eliminate false positive alerts by indicating to the software that the alert was not helpful or did not accurately reflect what is likely to happen.
The company states their software’s machine learning model would need to be trained on their client’s own company data, which likely consists of transaction data, customer profiles, and loan contracts. Aithent would then expose the machine learning algorithm behind their software to the client’s data.
This would train the algorithm to discern which data points correlate to patterns in customer banking or buying behavior, such as a customer who has defaulted on multiple loans within a certain period. The algorithm could also discern the data that correlates to customer behavior changes across segments, as well as fraud or attempts at fraud.
The software could then predict and report on various customer patterns a bank manager would want to know about when making decisions for their company. These could include behavioral patterns that usually precede fraud and customers with a high risk of defaulting on a loan.
These patterns can also be found within specific customer segments, such as men aged 30-50 years old. This may or may not require the user to upload information about their recent loan contracts or future business plans into the software beforehand.
Aithent claims bank managers can integrate the software into existing databases for bank and customer information.
We could not find a video demonstrating how Aithent’s AABI software works. Aithent does not make available any case studies showing success with their software, and they do not list any major banks as clients for this solution.
Gourish Hosangady is Executive Vice President of Business Analytics at DataRobot. He holds an MS in Engineering from the University of Missouri. Previously, Hosangady served as Vice President of Business Intelligence Solutions at Inforeem.
DataRobot offers a namesake software which it claims can help banks and financial institutions gain insights from business data and create tables and models from company data. Each predictive model is a visualized representation of the predictions the software makes to organize the results for clarity and communicability. They do this using predictive analytics.
DataRobot states their software’s machine learning model is trained on their client’s company data, which is likely hundreds of thousands of customer profiles and transaction records. For example, a predictive model could display all of a given customer’s late payments and loan defaults in a visualized chart by accessing the banking data associated with that customer. To do this, the software would need to have been trained on loan and customer data relating to prepayment risk, credit risk, and late payments.
The bank or data scientist would then expose the machine learning algorithm to this data. This would have trained the algorithm to discern which data points correlate to the desired search term or terms that narrow down data results and could be used to create a predictive model in the form of a graph or table.
For our example, this would be a search term such as “will_default” as defined by a bank or data scientist. The underscore appears to be used to keep the search term recognized as a single term, and not as two distinct terms.
The software would then be able to make the user’s requested prediction and create a visualization such as a graph, spreadsheet, or table with the data that results from a search term or set of search terms.
This could be in the form of a table or chart listing each potential customer’s likelihood to default on their requested loan. This may or may not require the user to use data about their recent or future loan contracts into the software beforehand.
DataRobot claims banks and financial institutions can integrate the software into their existing databases and other stores of company data.
Below is a short 2-minute video describing how DataRobot can help banks and financial institutions predict risk for lending. At 1:29, the user’s desired prediction search term is shown with an underscore.
DataRobot does not list any explicit banking case studies, but banking is one of the industries they list on their website. They also claim USBank as a past client.
Tom de Godoy is co-founder and CTO at DataRobot. He holds an MS in Mathematics from UMass Lowell. Previously, Godoy served as Senior Director of Research and Modeling at Traveler’s Insurance.
Header Image Credit: Nairobi Business Monthly