SOLUTION + IMPACT
In the summer of 2024, I designed a cheque fraud detection dashboard for web and a refined two-factor authorization feature for mobile that led our team to win an award for best use of Amazon Web Services at BMO’s annual digital innovation competition.
SETTING THE SCENE 🎥
Imagine Shark Tank, but instead of pitching to Mark Cuban, we were pitching to BMO’s senior executives in the heart of Toronto.
That was Destination Digital 2024, a 30-hour hackathon setting where teams were assigned business problems, team members were assigned at random, and technology sponsors (e.g. Microsoft, AWS) and mentors were all working together to design innovative digital solutions to enhance the technology and reputation of BMO.
TEAM/ROLES + DATE
Volunteers signed up and were then placed onto teams that drew on each individual’s strengths. I was the Designer/Marketer, in a team with an Operations Specialist, a Data Guru, a Programmer, and a Business Strategist.
This event took place in person from June 12-13, 2024.
With our teams made a couple of weeks beforehand, we had time to deliberate and decide on a challenge.
Ultimately, we decided to tackle the challenge of cheque fraud.
THE CHALLENGE: cheque deposit fraud detection
The following how-might-we statement was given to us by the challenge sponsors:
“How might we effectively use AI to properly and credibly identify counterfeit cheques presented at branches or through our online applications?”
The context of the challenge given by the challenge sponsors:
Several tools and systems are used by fraud ops team members to determine if a check is fraudulent or legit, as opposed to one centralized space.
Triggers to detect fraud include font anomalies, handwriting variations, and out-of-sequence cheques, leading to missed cases due to human error.
What would the ideal solution include, according to the challenge sponsors?
Allow fraud ops analysts to more quickly determine whether or not a deposited check is fraudulent or not.
A more advanced solution would allow mass check deposit review with a web page that flags identified fraudulent/valid deposits with a confidence level from 0 to 100%.
How did we learn more about the problem?
We had an ace up our sleeve - a team member who spent years working adjacent with the Fraud Team and knew some fraud analysts personally.
This connection allowed us access to interview 3 fraud analysts via MS Teams, where we could learn more about their current systems, and the problems they face.
Identifying the core issues
From speaking to the analysts, we were able to find out the following:
There are primarily two types of cheque fraud:
account-based/behavioral fraud (e.g. a large deposit on a newly-opened account)
physical-cheque fraud (e.g. cheque forgery).
Manual processes involve fraud analysts reviewing one cheque, one person at a time, which takes up a lot of time (an average of 15 minutes per case).
The current technology that we have leads to a lot of false positives and missed fraud. The current software is very old and archaic.
Fraud analysts have to manually create rules based on reviewing the analytics and statistics of past and current cases which takes time.
Making the solution, an MVP (minimum viable product)
We didn’t want to reinvent the wheel.
For our solution, cheques would still flow through regular channels (mobile app, ATM, branch deposit.
But now, all cheques would flow through AWS Dynamo DB, and into a new and refined dashboard for fraud analysts which I designed on Figma.
Now introducing... BMO Fraud Fighter Central!
Home Page - a streamlined view
Fraud analysts could see the number of fraudulent transactions in a given period, graphs showing the statistics of fraudulent transactions by location and by channel, rule efficiency score, as well as to-do items.
Cheque Review
A fraud analyst would be able to review a singular cheque, with fraud indicators calculated from 0-100 that they could review. Also, they would be able to see that person’s account and any past reviews, notes, alerts, and changes.
Rule Creation
Fraud analysts can create, edit, and delete rules that would create triggers/alerts to potential fraud. Here, we use AI machine learning to suggest rules that analysts may have missed through observation alone.
My Alerts
Fraud analysts can review fraud cases in a table view, view and assign statuses to a case, as well as escalate to a senior analyst if needed.
But how about the customer?
While this is great from the fraud analyst’s view, what if there is a way to further prevent fraud right at the source, from the customer’s vantage point?
I explored the current BMO process of depositing a cheque and introduced two more solutions to detect fraud:
Leveraging two-factor authentication
Input field verification
Two Factor Authentication
If a transaction seems suspicious, the app will ask the customer to authenticate themselves through a separate method (e.g. SMS, or Email) If this effort fails, the transaction will be blocked and the fraud team will be notified.
Input Field Verification
Users have to re-enter values taken from the cheque if it’s unclear or suspicious.
Results - We won! 🏅
We won Best Use of Amazon Web Services for our pitch. Although we didn’t make it to the grand prize, the judges loved the demo shown above and appreciated its potential.
By using dummy data fed to the data model, we found we could reduce the time spent on each fraud case from 15 minutes to 5 minutes!
Conclusion
What would we redo? ⏪
We did not have a project manager on our team. If we were to redo this, I’d focus more on an execution plan for the project, as well as estimating the funding and resources needed for the project to proceed.
How would we move forward if given the opportunity? ⏩
I would speak to more members of the fraud team, and get their feedback on what we’ve demonstrated so far. (e.g. What works? What doesn’t work? What could be improved?)
What did I learn? 🧠
Using pre-existing components (e.g. through a design system) speeds up the design process!
I leveraged BMO Commercial Banking’s design system Lexicon and decreased lag that would have been spent on designing new UI components and instead got to focus more on the user experience and flow for our fraud analysts, saving valuable time.