First adaptors are companies that take the first risk in trying out new technologies. They are the first buyers of technology because they understand the potential benefits, have the money and personnel to make it work, and are willing to take a large technological risk because the benefits of success far outweigh the sizable possibility of failure. American Express, like the U.S. Department of Defense and large research labs, fits very neatly into this category. Second adapters are usually companies that are part of the Fortune 1000 who wait to make sure that the first adapters have proven that the technologies are valuable. Third adapters are the rest of the world, those that invest in and use the technology only after it is foolproof and economical.
American Express's AI journey had been a long one, and as a first adapter, it suffered a minor failure on its initial foray into AI. Its first exploration was the ill-fated K:Base back in 1984. A Lisp-based system, the application undertook the job of handling fluctuations in foreign exchange rates to obtain the best possible value for funds transferred internationally. The project, like some of the earliest AI undertakings, proved too broad in scope and thus was burdened by its complexity and lack of definition.
It was ultimately abandoned amidst bad politics and charges of vendor over-promising, but the company was not deterred. It jumped into another expert system and ended up with the highly successful Authorizer's Assistant. That project, which has become one of the most publicized of all corporate expert systems, initially involved the use of Inference's ART expert system tool and Symbolic's Lisp machines and has now migrations. Today it does the work of 700 authorizers, and savings estimates run in the tens of millions of dollars.
At last summer's 1992 Innovative Application of Artificial Intelligence conference, American Express was singled out for another of its expert systems, this one referred to internally as "Son of Authorizer's Assistant." Named the Credit Assistant, this expert system helps the company's credit operation review accounts for fraud situations. The system has reduced informational management systems transaction for review from an average of 22 to one, insuring worldwide consistency on credit policies. The company estimates that the Credit Assistant improves productivity by 20% and saves the company a minimum of $1.4 million annually. Again working with Inference, Amex developed the program with ART-IT and deployed it in a UNIX environment using IBM RS/6000 workstations.
Most of the publicly known work that American Express has done using AI technologies is developed in its Travel Related Services (TRS) business division: the one that handles charge cards, travel agencies, and so forth. But American Express's Shearson Lehman Brothers investment division, while being more secretive than TRS, probably has as much if not more AI work going on behind its doors.
Shearson Lehman is especially fond of neural nets and probably uses them more than it does expert systems. This business division looked into neural nets more than three years ago and now employs them as strategic part of its business operations. Like other financial houses on Wall Street, Shearson Lehman uses neural nets to squeeze extra percentage points out of its money management-points that can't be realized with statistic analysis packages.
HOW DO THEY DO IT?
So much for the application notes. We're going to do things a little different in the rest of column and continue it into looking at specific applications—the who, what, and wherefores—we're going examine the how of putting AI in practice. Specifically, we're going to look at the how of getting intelligent applications reorganized as a valuable corporate tool—which is probably the single biggest obstacle to getting AI implemented in most environments. Using one of the best adapters of AI as a model for discussing the assimilation of AI into a corporation is the best place to start. So what follows is a part of American Express technology-development story. Think of it as my version of a Harvard Case Study—only better—because I don't give you a test after you've finished. Plus, you're saving a bundle on tuition fees alone.Let's start from the beginning, shall we? In late 1984, Louis Gerstner, the CEO of American Express's TRS division, decided the company was negligent in pushing the envelope of new technology research. It wasn't taking enough big risks that might result in the big payoffs, and as a result, the company stagnated in doing exciting, and potentially valuable, research and development work.
It was also evident that the company wasn't spending the kind of dollars for the long term that it needed to get really big win. After all, this was the 1980s, and short-term outlay was considered the only way to go. What to do?
Gerstner set up a technology research fund that gave American Express's business unit the chance to pitch some neat application ideas with the possibility that the fund would pay for 100% of the work. Obviously, free money is a very attractive incentive for budget-conscious departments, especially when it runs into seven figures, and Gerstner's fund attracted some 80 application proposals right off the bat. Out of all those submitted, five were chosen, the most notable to readers of this column. Thus was born the impetus to get business units into doing big picture technology development.
That was all well and good to start with. American Express corporate jump-started some pretty advanced technological projects within its business units, ranging from TRS to Shearson Lehman to its First Data Corp. subsidiary. And it fully expected a failure rate of roughly 50%—yes 50%. While that might sound shocking to many executives, consider the ledger sheet on this percentage: If you try to create innovative applications at the "edge of the envelope" then you have to consider that some of the failure is part of the parcel of exploring new territory. A 50% failure rate means that you're probably doing the best job of using your resources to get the maximum benefit out of new technology.
All right, so failure is expected. Conversely, so is success, at the same 50% rate. The American Express corporate fund, though, found that footing 100% of the bill for new technology development meant that some business units were looking at it literally as play money and perhaps did not pursue the most relevant applications vis a vis long-term benefits. A little tweaking was in order, so American Express corporate decided to go to a 40/40/20 split with its money. Forty percent would be put up by the corporate technology fund, 40% from the interested business unit, and 20% would come from another business unit that had a vested interest in the outcome of the proposed application.
This way, the company started delving into technology areas that were not only near and dear to individual business units (since they were now using part of their own budgets) but also that could be cross-fertilized into other regions of the company. From this structure came the first inklings of research into neural nets, fuzzy logic, and even genetic algorithms.
As all of this was being put into place, American Express had a "corporate champion" that saw the benefits of technology development and made sure that it was protected from assault by other internal organizations. As many of you know, technology development is something that doesn't usually result in immediate pat back, and plenty of other groups—marketing, sales, administration, engineering, operations, and so on—feel that such "speculative" spending would be put to better use in on their projects.
The corporate champion usually acts as a defender of the technology faith, and American Express technology had an unlikely protector: chairman James Robinson. Robinson has made sure that the company maintained a view of technology development as a strategic issue and thus not to be tampered with. Rather, it is to be nurtured to bring benefits to the company that could not be realized without its efforts. So all that corporate champion that used to sound kind of pithy and even archaic—well, it's true, no matter what size company you're talking about.
So American Express has put into place a means by when technology can be fostered. Next, that fostering requires a lot of support, which involves people, politics, money, and corporate direction. I'll pick up on this thread next month by talking with Neal Goldsmith, a director of American Express's Corporate Technology Strategy group. His insights about the future of expert systems versus neural networks as part of corporate technology development and technology transfer are as well thought out as any I've ever heard from an "end-user."
And, as promised, no homework over the holidays. Is this a cake course or what?
Harvey P. Newquist III is CEO of The Relayer Group, Scottsdale, Ariz., which publishes the AI Trends newsletter. He can be reached through AI Expert.
