Excerpt from article "Business Rules are the Key to CRM and One-to-One Personalization" published by Rolando Hernandez, Founder & CEO, BizRules.com, in the August 2001 issue of the Business Rules Journal.

Expert systems are a great way to deliver expert advice and product recommendations over the Web and in a CRM personalization application. Expert systems are one of those emerging technologies that have been around 20 years before they were 'discovered', like the Internet. The Internet was 20 years old when Tim Berners-Lee wrote three programs called WWW, HTTP, and URL and invented the World Wide Web in 1991.

Although rule-based expert systems haven't had to change much during the last 20 years, they have evolved in order to support new technologies like eBusiness. Today expert systems are contributing to many eBusiness and CRM personalization success stories, although it is hard to find a company willing to talk about it.

"Expert systems didn't really go away. They went undercover," says Richard Barfus, President and CEO of MindBox. MindBox was one of the pioneers of expert systems and AI. Originally known as Inference in the 1980's, it became Brightware in the 1990's, and spun off into MindBox last year.

"When we implement a successful system we'll want to do a press release and they say 'Are you crazy? We love it, but we won't talk about it. We love it, it works great, by the way here's all the money we're saving and here's all the money we're making, but I don't want anybody to know about this.' There are many, many examples of expert systems where people see it as their secret sauce, and they just don't want to publicize it because they don't want anyone else to be doing it," said Richard.

American Express' Authorizer's Assistant (AA) is a good example. AA is one of the oldest and largest rule-based systems ever built. AA is one of those classic expert systems that people read about and studied in the early 1990's. Then the publicity stopped and now you don't hear anything about it. I asked Richard whatever happened to AA because his company, MindBox, developed it in 1988. He said, "Authorizer's Assistant is a perfect example of all the wonderful things of these types of applications. It's been in continuous production for 13 years, and it's never been down. When we originally built it for AMEX it was for the Green card only. Now it applies to all of their card products both domestically and internationally. And only five people maintain a core system like that. If you look at a core system of that magnitude in most organizations, you're talking about 30-40 people in the IS department assigned to maintaining it."

It's one thing to hear that from a vendor, but another to hear it from the customer. So I called American Express and spoke with Marilyn Stark, Director, who leads the Authorizer's Assistant expert system. AA is an integral part of AMEX's Credit Authorization System (CAS) transaction processing system for receiving credit requests from merchants, processing those requests, and transmitting the approval or denial back to the merchant. AA analyzes the small percent of requests requiring human attention, and it uses business rules to simulate the decision process of experienced credit agents. [3]

According to Marilyn, the core AA application contains over 3,000 business rules, and the entire AA system contains about 35,000 rules. AA originally ran on a Symbolics machine, and it now runs on an IBM RS/6000 under AIX. It went into production in the 4th quarter of 1988, and it has been running 24x7 ever since. Even though it has evolved over the last 13 years, AMEX has not had to rewrite the rulebase. The original rulebase is still in production. I asked Marilyn why AMEX stopped talking about this system, and she said that the industry moved on to other new things like neural nets and other new technologies, but that "Authorizer's Assistant never went away."

If you use your American Express Card to buy something online, it is very likely that the Authorizer's Assistant expert system will process your transaction. Rule-based expert systems are key components of many large OLTP and CRM personalization applications.

References

[3] Steven Alter, Information Systems, A Management Perspective - American Express Authorizer's Assistant, Prentice Hall.

Copyright © 2001 Rolando Hernandez

BIZRULES is a registered trademark of BizRules.com.  Copyright © 2002-2008 BIZRULES. All rights reserved. For further information email or contact us. Revised: 1/18/2008.