LINK ANALYSIS
Redesigning the experience for Entity Relationship Analysis
Overview
Nice Actimize is a company that helps Financial analysts detect and fight financial crime. Financial Fraud and BSA Analysts use the 'Link Analysis' tool during their investigation of alerts to identify any unusual or suspicious relationships among different entities (people, accounts or transactions). 'Link Analysis' is a data analysis and visualization tool that helps explore the relationships among different nodes or entities.
The Problem
We launched the 'Link Analysis' feature in the first version of the product, but soon realized that many people were not using it because it wasn't functioning as they expected to and it wasn't helping financial analysts identify suspicious patterns. I wanted to redesign this tool to improve the experience by highlighting suspicious patterns for the users using machine learning. This would also help make the product better than what the competitors have to offer.
The primary goals we wanted to address were:
Enhance the Link Analysis tool so that Analysts can easily identify unusual links and make quick decisions with confidence.
Allow users to customize the graphs, so that it offers way more value and flexibility than what the competitors have to offer.
The Solution
We used the design thinking methodology to understand the users, define the problems we wanted to solve, evaluated different ideas and tested with using prototypes. We redesigned the tool and created a smart and fully functional link analysis tool which assists analysts in expediting their investigation experience.
My Role
I conducted initial user research and performed detailed analysis on the feedback. Used this information to clearly define the goals, and came up with multiple UX improvement ideas and iterated on them. I created wireframes and evaluated the designs through usability studies. I iterated on the prototypes based on the feedback from the studies and came up with the interaction design of the final version.
Empathize
User Research
We conducted a series of interviews with BSA Analysts to better understand their pain points and effectively address them. We optimized the interview questions to mainly focus on our objectives and that lead to actionable insights.
Some of the questions we asked were:
Tell us about your alert investigation process and how 'Link analysis' helps with investigation.
What is your experience with the current link analysis tool? What would you like to change about it?
What are key scenarios where 'Link Analysis' is most critical for alert investigation? Are you able to complete this investigation using this tool?
User Feedback
After interviewing with 4-5 participants, we identified common themes in the data we collected. Some of the major problems we identified with the current link analysis were:
There's too much information and no way to filter what you see, and not all the information is useful.
Some of the links are missing, and some nodes don't work as expected.
All information is shown to be of equal importance. So Analysts need to put in lots of time and effort to identify what information is suspicious.
They didn't trust the tool and it didn't meet their expectations
Define
After thoroughly reviewing information from the research phase, we clearly defined the goals for the redesign. The main purpose of redesigning link analysis is to help Fraud and BSA analysts visualize complex situations that may be hard to discern.
Assist in the identification of unusual/abnormal links between people, businesses and accounts.
Reduce the time and effort needed to identify and expose patterns of money laundering, fraud or other criminal activities.
Help them make quick approval or denial decisions with confidence.
Ideate
We started listing simple enhancements that would improve the experience of using the Link Analysis and how we could address them. This included:
Removing multiple links, while grouping like and meaningful information together.
Carefully eliminating duplicate/redundant links.
Allow users to customize the graph by selecting which nodes they want to see.
Showing additional information about each node on demand.
We also identified key scenarios where we could use machine learning to confidently indicate suspicious activity. These included:
Highlighting suspicious nodes or links using information we already know about the person/account.
Highlighting commonly used phone numbers and addresses by multiple people or accounts.
Fraud rings
Multiple deposit and withdrawals.
Early Ideation
Declutter the graph by removing multiple links
Most often there are multiple relationships between 2 nodes, sometimes going up to 20. Showing 20 links really cluttered the graph. The idea here shows how we could just show 1 link, but list all the relationships. Thus reducing cluttered and making the graph better.
Allow users to select which nodes to view on the graph
With Link Analysis for financial crime, the graph could easily become really large is 10+ nodes. Especially when you start to investigate further and expand on them. For example, expanding the subject node, will display additional nodes for their phone numbers, addresses, accounts, associated persons etc. After reviewing this information, the analyst may not find all the nodes equally important to their investigation. So we wanted to allow them to filter which nodes they want to view, and only display those types of nodes.
Design Considerations
Our initial idea was to use checkboxes, but we soon realized that using the space wisely was important because the graph could grow big needed all the space. So we chose to proceed with a drop down control.
Highlighting common nodes
As the analyst was proceeding with the investigation and expanding the graph, we learned that they look for common nodes like phone numbers or addresses that are common across accounts of unrelated subjects. This is one the key indications of money laundering. So we wanted to highlight these nodes as the graph expanded.
Highlighting suspicious activity
We identified common fraud scenarios which the analysts would like to investigate further. With machine learning and AI sometimes we have all the data to highlight this activity and present it to the analysts without them having to search for this information and put it together. Here is an example of a fraud ring.
Early Idea Validation
We presented these low fidelity designs and communicated the ideas to a few analysts, and learnt that they found these simple enhancements were exactly what they were looking for. All the valuable information was already present in the graph, but they just needed a simple way to process and understand the graph. They found that highlighting suspicious links and nodes using machine learning extremely valuable.
Prototype
After the ideation phase, we created simple prototypes to demonstrate this new functionality and design the full experience of using link analysis with these enhancements. Here are a few examples
Allowing users to select which nodes they want to view
Allowing the users to see additional information about each node as they click on it
Uncovering a fraud ring
Validate
We tested these prototypes with analysts to understand if this helped assisting in the identification of unusual/abnormal activity. We learned that this really helped get the analysts to focus on key pieces of information instead of trying to process all of it at the same time. These were great scenarios and most frequently attempted ones. Although we had a long way to go to correctly detect the unusually occurring ones using machine learning.
Impact
The redesign of Link analysis had a significant impact on analyst productivity and usage. Some of the key highlights are
Helped analysts save time and effort by over 50%, allowing them to confidently resolve 2x cases in the same time.
Highlighting the suspicious nodes and link improved accuracy by over 40%.
The success of the redesigning the link analysis tool added a competitive edge for our product sales.