Everyone seems to know a little about expert systems these days – but usually very little about them.  Why might expert systems be important to you?  Like anything else that provides a competitive edge in this fast moving world of ours, if your competition has them and you don't, your company is liable to end up on the short end of the stick.

There was a serious article regarding expert systems in the last issue of VARBusiness.  The premise was expert systems got a lot of bad press in the early 80's because of some highly visible failures and were forced under-ground so to speak.  They are now called knowledge-based systems or intelligent systems and are supposedly so common that one of the magazine's "experts" was quoted saying "Today, almost every large corporation has (multiple) expert systems in use."  I was having trouble swallowing that line until I finished the article and realized their definition of an expert system included any form of rule based programming.  This broad redefinition didn't diminish my view of this excellent magazine one iota,  after all, this same edition mentions Pick in a favorable light in an article about "non mainstream operating systems".  Any mainstream magazine that knows how to spell Pick is OK in my book!

Now, I would certainly have trouble holding myself out as an expert on expert systems, but I did code several complex systems back in the early 70's that were clearly expert systems, at least by these standards.  At the time we called it heuristic programming.  One in particular was a lot of fun.  A major shoe manufacturer had a pending problem converting the gross orders into individual job tickets.  Shoe making is still very labor intensive and there were lots of efficiency issues involved in this process.  Unfortunately, they were all in the head of a single 65 year old individual.  Since no one else in the company understood the problems, they were concerned they would lose considerable manufacturing efficiencies if this person retired or became ill.  The computer project was almost rejected by the consulting company I worked for.  No one thought we could solve the problem.

Extracting the non-verbal rules from the "expert" was the hardest part of the job.  Over the course of four weeks, I watched over his shoulder and tried to duplicate  his decision process.  When I made a serious attempt, I would take the output to him and try to get him to tell me why his output was better than mine.  Then I would go off and try to add the newly discovered constraints to the existing rules.  After perhaps a dozen attempts, he was forced to admit that the computer was doing a substantially better job than he was.  This was in 1972, coded in COBOL, and used none of today's pre-packaged expert information engines, but was certainly an expert system.  It also demonstrates one of the hardest parts of creating an expert system – understanding and creating the decision rules.

Why would you want to consider using an expert system in some piece of your business?  There are lots of possible reasons.  For example, you might use them to apprentice or train people.  Any time there are a pool of people making decisions for your company, such as order entry, help desk, etc. you should be concerned about properly training them to making consistent decisions and have some form of a quality control check.  Any time you need to extend critical individual knowledge to a larger group or to replicate a knowledge resource as in the shoe manufacturing example, a computerized expert should be considered.  The process of creating (or even attempting to create) an expert system is also useful to expose and understand the actual rules of the game, or to develop new insights to how a system works.  An often overlooked advantage is the ability to change the rules more quickly as the business changes.  In today's regulatory environment, the blind and consistent applications of highly visible rules can reduce your company's exposure to many types of discrimination claims.

Where might you use a piece of an expert in your business?  There are thousands of possible places they might be useful.  Fairly simple business examples might include issues such as mailing list selections, inventory level management, or evaluating and granting credit.  Intermediate difficulty examples might include items such as manufacturing and shop floor scheduling, medical claims processing, various kinds of insurance rating, machine diagnostics, or route scheduling.  The heavy duty areas (at least in our context) might be assisting the help desk, or actually participating in the design of application systems or the programming of them.  Few of us need to worry about the industrial strength, life or death applications such as medical diagnosticians, nuclear power control, or aircraft flight control systems that have been the typical grist for the high profile successes and failures.

Don't look for miracles from your first project.  Think about where in your situation a better, more consistent application of known rules would be worth the investment required to address a totally new technology like this.  A small project might be attempted in-house by training a few of your own people and giving them the time and room to experiment.  Larger projects probably require an investment in a formal expert system engine.  Don't forget to look at packaged solutions if your needs fit one of the relatively few markets covered by real packages offered today.  If the project is critical or the return is clearly substantial you may want to consider using a specialized consulting firm with experience in this area to improve your chances for quick success.  If you are considering this option, be sure you deal with reputable firms.  They don't have to be large, but this is a very new and highly technical area so they should have solid credentials and a good reference base.

Using outside consultants is especially valuable the first time you venture into this kind of unknown territory.  They can train your people and help you avoid the more obvious pitfalls.  After the first attempt is successful, you could extend the project or start others on your own with more confidence.  Consultants are not a guarantee by any means, but used wisely they generally substantially improve your odds of success.  They also reduce the time and costs of the learning curve, even (or perhaps especially) if the project proves unfeasible.  By the way, I believe the ability to say "you're off base – that won't work" is a marvelous sign of a good consulting firm.  Think of the alternative and you'll see what I mean.

In general, if your department is over two or three people and isn't using at least some outside consulting in special cases, you must not be tackling any hard projects.  Well thought out but hard projects are one of the hallmarks of cutting edge competition.  If you're not doing these projects, you can sure bet your competition is.   Even MIS departments on the small side need to spend regular time in technology research and intelligently applying that information to bettering your business.  Knowing what your competition is doing is not nearly as good as doing it first.