(FORTUNE Magazine) – Just like computers, people can be programmed to think the way experts do. Lev Landa, a Soviet emigre, is trying to show American business how to turn novice office workers into experts in days or even hours. He does it by duplicating the mental processes of an expert in the form of an easy-to-follow decision tree, or algorithm (see example), that leads a novice through a task step by step. In effect, he creates a computer program to be run not on a machine but by the human brain. After pushing American business to try his methods for nearly a decade, Landa, 59, a bearded psychologist with a bass voice and a huge grin, is winning converts. Clients such as AT&T, Citicorp, Du Pont, Hartford Insurance, MCI, and Procter & Gamble have been appraising his ideas. Landa has found that an expert can describe only 30% to 35% of what he does. That's the bit that appears, for example, in training manuals and income tax instructions, which is why they are so often muddled. The novice has to learn the unexplained remainder by trial and error, which can take a long time. One of Landa's favorite seminar tricks is to ask people to pretend they are writing instructions telling a foreigner how to use an American public telephone. Although his listeners are all clearly proficient at the task, they invariably leave out such key steps as listening for the dial tone -- illustrating Landa's point that experts are not aware of much of what they do. He discovered this human failing as a high school student in Leningrad when he became frustrated by his teachers' inability to explain their thought processes. While working for his Ph.D. in psychology in Moscow, he began developing ways to uncover mental processes and break them down into simple steps that can be diagrammed. He tested his methods on geometry students and then at the Academy of Pedagogical Sciences in Moscow. Landa's son, a political activist, left the Soviet Union in 1975. Landa and his wife, a psychotherapist, followed the next year and eventually settled in , New York City. He made business his target because he observed that companies seem unable to write clear manuals for training vast numbers of people to do routine tasks. His company, Landamatics International, consists of himself, a part-time secretary, and offices in a run-down building in Queens. Clients have included the Dutch internal revenue service, which since 1984 has used a Landa-designed algorithm in place of a typically maddening tax form, roughly the equivalent of the U.S. W-4, for determining the proper number of exemptions for withholding purposes. The Dutch Ministry of Finance pronounces the results ''a great success.'' Landa says Dutchmen make half as many mistakes on the new form and fill it out faster. Starnet, a San Diego long-distance telephone company, called Landa in because its billing analysts were making costly mistakes. Every month Starnet gets a pile of computer readouts about 17 feet high with bills from AT&T and other long-distance carriers. Even after three months' training, novices were making mistakes on about 40% of the bills -- for example, calculating a charge by using the rate for the wrong time period. After studying minutely all the decisions a Starnet clerk had to make, Landa drew up detailed algorithms that take a novice through each step to the correct answers. Although the whole process is complicated, involving dozens of decisions, at each point the clerk has to make just one simple, clear choice. Training time for new clerks dropped from three months to 15 hours, and the proportion of mishandled bills sank to 5%. Starnet's savings in 1986 amounted to $312,000. On a recent assignment at a major bank's credit card operation, Landa found only 20% of the employees working at what bank managers considered an expert level after nine months' training. Once he designed some algorithms, Landa says, novices could handle cases nearly twice as fast as workers using the old manuals, and did so without mistakes. While Landa's clients tend to be pleased, some find his claims a little overblown, and they point out that algorithms don't come cheap. It takes Landa ten to 25 days, at $1,500 a day, to create a human expert system. Also, his decision trees don't suit every company; they're most useful for training a lot of people to do a few jobs repeatedly. But in the right circumstances, says a client, ''the results are astonishing.''

CHART: TEXT NOT AVAILABLE CREDIT: NO CREDIT CAPTION: What to Do in Case of a Billing Error The murky legalese that comes with a Visa card bill prompted Landa to devise a decision tree that would help people correct mistakes. This is an adaptation of the first page. DESCRIPTION: Decision-tree for consumers with credit-card errors.