Discovering Fraudulent Checks to catch Fraud

Overview / Product Design @Unit21
Role / Designer (me!), Product manager, Engineering Manager, 4 Engineers and Customer Success
Timeline / 2 months
Tools/ Figma
Status / In Beta

Overview

Check Fraud is a rampant problem, with banks and credit unions having lost up to $24 billion up to 2023. We dabbled into fighting check fraud through improving a highly manual alert investigation process agents have to take, while also setting the foundation for an AI-driven investigation.


I discovered gaps in the check image comparison flow, championed an image comparison tool which led to efficiency improvement by 80%. 



Needs Assessment

How are alerts investigated?

Each bank defines its own process guidelines for which pieces to investigate for a check transaction, for ex. reviewing transaction activity, the various entities involved and their relationship and comparing the check against past checks.

To better understand the investigation process and discover pain points, I conducted user research with 10 users across 5 various banks, their roles ranging from Fraud Specialists to Fraud Managers to VP of Fraud.

Findings: One of the biggest issue identified was agents having to jump between platforms to find necessary information.

How do agents feel about their process?

According to Agents...

How do fraud managers feel about efficiency?

According to Fraud Managers...

...But what functionality is actually required?

As part of user research, I conducted user walkthroughs, competitive analysis, and ethnographic research. The following user requirements emerged as a result.

As a fraud analyst, I want to be able to view check image in Unit21

As a frayd analyst, I’d like to be able to zoom into check images, since I’d like to find inconsistencies and analyze signature as they are usually small.

As a fraud analyst, I need to compare the given check image with previous check images in a single view.

As a fraud analyst, I would like to enhance check images to identify signs of alterations and correct digital artifacts that occurred during an upload.

As a fraud analyst, I would like the ability to conduct quick analysis on an entity’s past transactions

As a fraud analyst, I want to overlay old checks on the check I am viewing to identify differences in signature, check stock, handwriting etc.

To Automate or Not to Automate?

Comparing check images visually sounds like a tedious task and should seemingly be automated. I considered whether we could automate comparison and display any discrepancies on the alert itself. This could save at least 5-10mins of investigation time.

However, AI needs to be built with the context in mind. I went back to the Fraud Managers and discussed where automation lied in their investigation process.  Biggest issues identified were:

• The fraud industry is risk averse: Fraud is a risky space. Accidentally marking something as false positive when it isn’t could result in huge financial losses and damage to the bank’s reputation.
• Automation Bias: Risk of agents relying heavily on automation and taking its result as source of truth. They may not turn around and “double check” their work.
• Missing foundational work: Automation is always a good add-on to the solution. However, if there is no manual solution in place, agents cannot double check AI’s work.
• Understanding the users: If agents make mistakes, they could have their jobs in line. They would rather double and triple check the alerts before actioning. The user group is also typically 40+ years of age and isn’t comfortable trusting AI right away.

Main takeaways were, there needs to be the ability to view and compare check images in Unit21. Once agents are comfortable with the software, automation can slowly be introduced in stages. Therefore I designed the feature keeping in mind that it can eventually be automated.

Testing with the users

I conducted usability testing with 7 users, ranging from power users who use Unit21 platform for their day job to agents that conduct alert investigations in other platforms. The two major concerns we discovered were:

• Low discoverability -> 4/5 participants had a difficult time discovering the trigger point for comparing images. As a result, they were not aware that the feature even existed.
• Scrolling when compare images was un-intuitive -> Breaking Recognition Rather than Recall pattern. Participants were expected to continuously scroll when comparing the check images, or they were expected to remember information from one part of the interact to another.

Usability testing revealed that the initial assumption of choosing flexibility as the criteria at the expense of implementation time could lead to UX issues later. I therefore re-introduced option#1 and designed an end-to-end flow that
1) would work for novice users who just want to compare against a list of pre-selected images and
2) expert users who can pick their own images.

Additionally, I also updated the UI to reflect a side-by-side image comparison as compared to scrolling pattern.

Final Iteration

Implementation

Check image comparison was swiftly developed in close collaboration with the development team, as well as tested internally by the QA team and Solutions Engineering. We were able to work closely with the data integration team to simplify complex queries to load the most relevant check images. The project is now in now fully developed and ready to be rolled out to beta customers.

Learnings

User Research with Banking Managers vs. Agents

Designing a SaaS software for enterprises can be tricky. Sometimes the people you talk to are not the people who will eventually use the product. Through this project, I learnt the importance of prioritizing talking to agents over the management, whether that meant finding users through unconventional means such as linkedin reachouts. It is still imperative to balance user needs vs. business goals so I ensured to collaborate with both the management and the end users.