How Analytics Fights Consumer Fraud

By Erika Morphy
www.CRMDaily.com, 
Part of the NewsFactor Network

November 8, 2002

Analytics applications use complex rules-based techniques, neural 
networks, pattern recognition and other profile settings within peer 
groups to identify certain transactions and set thresholds for what is 
considered "normal" shopping behavior.  

Do you ever get the feeling that your credit card company has a better 
handle on your day-to-day activities than friends or family? Increasingly, 
card and other consumer financial services companies have been using 
analytics  applications to detect and prevent fraud. Indeed, these 
applications are becoming part and parcel of their CRM  systems.  

Consider my recent experience: Over the course of 24 hours, I rented a 
car while mine was in the shop, took my dog to a canine oncologist 30 
miles away, and to cheer myself up, splurged on a high-end home 
entertainment system -- all transactions definitely out of my normal 
pattern of spending.  

By evening, when I tried to buy gas in a station off my beaten track, my 
credit card company had had all it could stand: It cut me off cold turkey 
and left a message to call and confirm that these were indeed legitimate 
charges.  

Financial Services Leads

So far, the financial services and, to a lesser extent, the insurance  
industry have been the primary adopters of analytics applications. For 
example, MetLife Auto & Home of Warwick, Rhode Island, a subsidiary 
of MetLife of New York, licensed Computer Sciences Corporation's 
Fraud Investigator data mining software earlier this year. MetLife Auto 
& Home also agreed to team with CSC to develop an early fraud detection 
system.  

"Fraud monitoring in the financial industry has developed some of the 
most complex techniques," Alex Veytsel, an Aberdeen Group analyst, 
told CRMDaily.com.  

Industries that have the most to lose from failure to catch deviations 
from normal spending patterns have pushed the envelope in what 
Aberdeen Group calls "event stream management" u- analytical 
capabilities that flag exceptional or unusual events.   

The Transaction Pipeline

"There are a number of programs now that sit on top of a pipeline of 
transactions -- that analyze and link transactions between payers and 
beneficiaries," Vijay Murthy, senior manager with Cap Gemini Ernst & 
Young's Financial Services sector, told CRMDaily. These applications 
use complex rules-based techniques, neural networks, pattern 
recognition and other profile settings within peer groups to identify 
certain transactions and set thresholds for what is considered "normal," 
he explained.  

The process only touches the customer when a flag is raised. "Each 
company has different standards [concerning] how a case is handled 
and what to do with a false positive," he said, referring to legitimate 
transactions that the system identifies as fraudulent.  

Too many false positives clearly can irk a customer, especially when 
credit is cut off at the pump -- in my case, literally. On the other hand, 
too few false positives mean that fraudulent activity is probably filtering 
through the system undetected, Murthy said. "Through trial and error, 
companies can strike a good balance between these two concerns."  

Defining Normalcy

The trick is in how companies define what is a normal event. One way is 
to use an expert's opinion as to what attributes are considered to be 
normal for a particular transaction in a particular sector. Another way is 
to study a series of transactions and use a system to model what is 
considered normal. This latter option is the more popular one used, 
Murthy said.  

These systems usually have other components as well, including a 
mechanism to report abnormal transactions and workflow technology  to 
review them and decide whether further action, such as cutting off 
credit, is warranted. "One abnormal transaction is usually okay. A lot of 
low charges that are out of the norm may be fine too," Murthy said. "It 
all depends what is the company's definition and threshold of normalcy." 

                                -==- 

Source: E-Commerce News - http://www.ecommercetimes.com/perl/story/19913.html


Cheers, Steve..

---