Frequently Asked Questions
Q: Why is fraud prevention challenging?
A: The main reason is that fraud is a very rare event. For example,
out of $1,000 credit card transactions the fraud loss for a bank is
usually less than $1. But $1 of loss for every $1,000 spent costs
banks millions of dollars each year. Detecting fraud is like finding
a needle in a haystack which makes the right prevention techniques
extremely important.
Q: What is the difference between rule-driven and data-driven fraud detection systems?
A: With a rule-driven fraud detection system, analysts set up judgmental rules that generate
alerts for transactions or accounts that are considered risky. Data-driven systems usually
use statistical predictive models to more accurately calculate risk scores for transactions or accounts.
These predictive models are built based on real historical data using data mining technologies.
Q: Why use data-driven predictive models for fraud detection?
A: One of the reasons is that data-driven predictive models are more adaptable. Since the models are built based on historical data,
we can always refresh the models using more recent data. This is particularly important given the ever-changing economic environment.
Q: What is the opportunity loss caused by fraud detection?
A: When a fraud prevention system gives false alarms, wrong holds
may be placed on good customer accounts. Since good customers
are inconvenienced, there is a chance that they may close their accounts and companies
will lose future revenue. Smarter fraud detection minimizes this
impact.
Q: Why is the goal of fraud detection to minimize total cost?
A: It is the total cost, i.e., fraud loss plus opportunity loss,
that impacts the bottom line of a business. To reduce fraud loss to
zero, we can simply place holds on every account but then there will be no business. Fraud prevention is always a compromise between fraud loss and opportunity loss. It is important to find the balancing point where the total loss is minimized.
