Fraud Prevention Web Application
We designed a fraud detection web application for a client that would enable users to quickly identify and prevent fraud. The key question we were trying to answer here is how can a company rapidly detect small frauds across millions of transactions?. The primary place that fraud occurs for the client is at the cash register, where workers can pocket small amounts of money by deleting transactions or adjusting coupon rates. These small thefts add up over time and can have a big impact on the company. Their existing method of detecting fraud took hours and involved going through over 100 million rows of data and 350+ spreadsheets. I worked on a team with 2 other designers and was involved more in the second version of the app.
Interviewed the users to understand all the ways fraud could occur
Documented the traces different activities can leave in the data trail
Observed investigators go about their jobs of identifying fraud to understand how they analyze data
Designed custom visualizations that map the team’s real-world process to the data, and zeroes in on anomalies and events that were more likely to reflect fraudulent activities
Understanding Fraud Types
Subbing: Punching a ticket up with 0 value by using an illegitimate pass or coupon
Stub Selling: Letting people in without a real ticket, usually by providing them with a "fake" ticket
Short Selling: Undercharging a customer (i.e. selling a child ticket but charging for an adult ticket)
Refunds: Issuing a fake refund
Deletion: Deleting a transaction
Coupon: Using an illegitimate coupon
Refund: Issuing a fake refund
Identifying metrics and their impact on fraud
Defining the Persona
See a high-level snapshot of metrics across stores, and identify anomalies that help direct where to dig deeper and investigate further.
Identify employees in a given store that may have suspicious activity, based on patterns across days and between employees.
See a “punchcard” of long-term employee activity, to see not only that there may be a problem, but also when the problem began.
After 2 years of using the app, the client came back to us for a second iteration. Some of the asks were
Enable analysis for seeing multiple stores week over week, quarter over quarter, and compared to last year
Enable analysis to compare trends of multiple metrics in a single store
Ability to create customized filters for better groupings of stores with common attributes
Store Overview Redesign
See a high-level snapshot of metrics across stores, and identify anomalies that help direct where to dig deeper and investigate further. The major change we made here was being able to compare the current year metrics to prior years.
Investigate a store further by analyzing multiple metrics in one view for the selected store. Identify if there are patterns or a correlation among the metrics that indicate fraud. This was a new view that we added as a follow up to the previous view where they identify suspicious stores.
Single Employee Breakdown
We added a view to analyze data on a single employee in-depth. This view serves as a way to confirm the fraud and provide evidence.
We also added the ability to save current selections and lock tooltips on a screen so that they're always displayed. This would be helpful when taking screenshots as evidence.
Overall this was a very successful project. The client was able to utilize this tool to identify fraud quickly and save the company money.