(FORTUNE Magazine) – Look for the formula of Coca-Cola, and you'll find it in a bank vault in Atlanta. Search for Intel's microprocessor blueprints or Merck's drug formulations, and you'll probably locate them in a locked file cabinet someplace. But at General Motors--once again this year the biggest generator of corporate revenues on the planet--hiding vital information isn't so easy. The company sets aside entire rooms as repositories for the diagrams, charts, and pictures--stamped GM CONFIDENTIAL--that are the soul of its machines. The secret rooms, as well as the company's design studios, have been closed to outsiders for years, but I was allowed to visit in February. Few restrictions were placed, but my tours were brief, and I could not take any pictures.

No place is more vital than the Rigorous Tracking Room, located in the basement of one of the flat-roofed, 1950s-vintage buildings on GM's engineering campus in the industrial suburb of Warren, Michigan. Rigorous Tracking contains a wall chart 45 feet long that plots 42 new vehicle programs--the very lifeblood of GM--through their three-year gestation. The programs are measured for timeliness, quality, and financial performance, and color coded by complexity: A major redesign is tinted maize, while a totally new car is colored blue.

Big Ten fans will recognize maize and blue as the colors of the University of Michigan, and indeed, something of a rah-rah spirit prevails at GM these days. It no longer teeters on bankruptcy, and ended 1996 with more than $14 billion in the bank. The spinoff of nonautomotive assets, as forecast a year ago in FORTUNE, is proceeding with the tentative $9 billion sale of defense electronics to Raytheon and with a contemplated public offering for 20% of the Delphi parts-making business. International operations are strong, as is the finance company.

There's just one big problem: GM has lost the home-field advantage. It makes almost no money in North America, where it has higher costs than its competitors and some of the weakest brands. Its cars don't make a profit, and although its trucks do, GM produces fewer of them as a percent of total output than either Ford or Chrysler.

Last year a strong performance by the rest of GM was dragged down by North American Operations (NAO), which builds and sells cars and trucks in the U.S., Canada, and Mexico. U.S. market share declined for the fifth straight year. Labor relations, never healthy, hit a postwar low as UAW strikes kept GM from building 300,000 vehicles, at a cost of $1.2 billion after tax. Overall, NAO returned a piddling 1.2% on sales. If you exclude the Delphi parts-making business, as GM began doing in March, NAO's return shrinks to 0.8%. That's well below GM's target of 5%.

Despite the debacle, GM expects major improvements at NAO in 1997. "It is going to be a great year," says the plainspoken chairman and CEO, Jack Smith, 59, looking a little haggard after losing 15 pounds on a diet. He won't have any excuses: GM is introducing 14 new models this year, representing 25% of its sales volume. The new models are better designed, better built, and cheaper to produce than the ones they replace. They are being launched into a healthy economy, though the competitive environment is fierce. GM will have to battle Ford and Chrysler for truck sales, and Toyota, Nissan, and Honda in cars.

Outside GM, expectations are lower. "The company has improved its cost structure, streamlined product development, and improved its image, but it still lags behind the industry," says Paul Ballew, a Detroit economist for J.D. Power & Associates. "The new generation of products is much better, but they are chasing moving targets." Like Chrysler and Ford. Chrysler raised profits in a flash in the early 1990s by redesigning the minivan and introducing dramatically styled LH cars. True, Chrysler is much smaller than GM, but size alone isn't the issue. Ford is two-thirds GM's size, and its North American operations run more smoothly.

GM dithers over almost everything. Explains Rick Wagoner, 44, the lanky Harvard MBA and Smith protege who runs NAO: "In 1992 and 1993 we were just trying to stop the bleeding. In 1994 and 1995 we put in some key processes: vehicle development, brand management, running lean. In 1996 and 1997 we are putting in a second round of key processes: engineering productivity, common procedures and systems, and global integration." For those keeping score at home, that's six years and counting.

Why is it so hard to get this monster moving? GM has unique handicaps rooted in its history and culture. NAO grew up as a collection of six separate vehicle makers overseen but never really managed. The common purpose for which they were joined often got lost in fierce intramural competition. Lacking unity and direction, the divisions became easy prey for powerful independent constituencies: suppliers, dealers, labor unions. Thus, NAO, with nearly half a million employees, operated without centralized manufacturing, purchasing, data processing, advertising, or market research. I visited a secret room, called Customer Satisfaction, filled with classified performance measures dedicated to enforcing one of Jack Smith's objectives: getting everyone to do things the same way--to "run common," as he puts it. Imagine finding such a room at General Electric or Exxon or Coca-Cola. (Characteristically terse, Smith describes his other three objectives in just five more words: "run lean, grow, go global.")

What a consultant might call "empowerment" became chaos at GM. Take stamping, a simple manufacturing process that GM made unbelievably complicated. As recently as the 1980s, stamping was split among several car and truck divisions that each produced their own metal parts using different presses and dies. This segmented approach resulted in some press systems that ran only 20 hours a week. Pieces stamped by Pontiac for a certain model wouldn't fit on a nearly identical car made by Buick. Now Joe Spielman, 52, the 6-foot-6 head of Metal Fabricating, is trying to impose uniformity on his far-flung organization. GM is spending $850 million to standardize the die production at 13 plants and reduce the number of press line setups from 57 to six. That's progress, though it won't send GM to the head of the class. Spielman says GM will spend about $2,700 per ton for stamping, vs. $2,200 to $2,300 at Toyota.

GM's assembly plants have been every bit as dysfunctional as the stamping process. The Pontiac Grand Prix, introduced in 1988, was built in its own dedicated plant in Fairfax, near Kansas City. But when Grand Prix sales fell off the shelf, production was cut from two shifts to one, and the factory lost money for years. Now the Fairfax plant has been reengineered with flexible machinery; it can build both the Grand Prix and a new Oldsmobile called the Intrigue, and it runs on two shifts. With the expensive equipment fully utilized, the Grand Prix has swung from a money loser to a moneymaker. But Fairfax still isn't world-class. GM takes 29 hours to put together the Pontiac. It hopes to improve to 22 by the end of 1999. Toyota already assembles a Camry in 20 hours.

GM's multiple divisions should enable it to achieve economies of scale, differentiate brands, and feed a multichannel dealer network. But one or more divisions always suffer because products overlap. Worse, GM never caught up to the truck boom because of faulty market research and stubborn leadership. So while trucks, including sport utilities and minivans, are approaching 50% of all auto sales, three of GM's six divisions--Saturn, Buick, and Cadillac--sell none. Mercedes-Benz, Lexus, Infiniti, and Lincoln will all have luxury sport utilities to sell before Cadillac does.

No division suffered more under this system than Oldsmobile. Its sales skidded from 1.1 million in 1986 to 331,287 last year, as aging customers died or defected to Buick. If GM had decided to kill Olds in 1992, it would have been euthanasia. Instead it kept Olds alive, perhaps to avoid the expense of buying out dealers, perhaps out of the fear that too many sales would be lost. One idea, briefly debated, was to merge Olds, GM's oldest and weakest franchise, with Saturn, its youngest and strongest, so that Oldsmobiles would become step-up vehicles from Saturns. The idea died aborning, but it had merit; today Saturn outsells Olds.

Having made the decision to keep Olds and all its other divisions, NAO had to figure out a better way to separate individual models and still maintain the economies of scale that come from common engineering. So it adopted a process of defining and targeting discrete groups of customers through a research technique called "needs segmentation." After completing a lengthy survey of tens of thousands of carbuyers, it divided them into about 30 "needs segments," based on such characteristics as income, age, family size, and driving habits. Buyers of sport-utility vehicles, for instance, were subdivided into six segments, represented by the $60,000 Range Rover on the high end and the $15,000 Geo Tracker at the low end.

This is a very controversial approach to building cars. Asking questions is science; interpreting the answers is art. Somehow GM must translate its cold data into hot cars, which it tries to do in its design studios, the repositories of its most closely held secrets. GM rigorously tests concepts with customers, then shows them scale models of the actual vehicle all the way through the design process. Customers even get a chance to feel how the car will drive by climbing into a $1 million programmable mockup.

GM ignored its research in the past because its engineers thought they knew best. Customers hated the beached-whale Chevrolet Caprice and the dustbuster minivans when they saw them in styling clinics, but GM built them anyway. Now information is king, carefully collected and analyzed in a single department run by Michael di Giovanni, 49, an intense, slightly built economist. "Our job is to get engineers and designers to pay attention to the research," says di Giovanni. "GM stopped listening to the customer in the 1980s, when it did things like put dual sunroofs on minivans. It also neutered the brands. We can commonize our cars where customers don't see the differences, and differentiate them where they do."

Other automakers view GM's research-driven design with everything from polite amusement to jaw-dropping disbelief. Research is fine, they say, but it produces boring cars. Executives must rely on their own experience and judgment to interpret it. Chrysler vice chairman Robert Lutz, possessor of Detroit's golden gut, is particularly disdainful. "We've spent too much time 'listening to the customer,'" he told a British auto magazine. "The customer is usually just a rear-view mirror. Why should we expect him to be the expert in clairvoyance?"

Early returns on GM's new policy are mixed. The refined Buick Park Avenue and the aggressive Pontiac Grand Prix seem perfectly styled to their target customers. But to me, the Chevrolet Malibu and Buick Century look as if their intended buyers are named Hertz and Avis. GM officials concede that some cars designed early in the process--like the Oldsmobile Cutlass--went wide of their intended audience, but they believe even the more undistinguished models will find their market. Utterly missing from this collection is a car that might stop traffic, like a 1984 Ford Taurus, a 1992 Dodge Intrepid--or a 1955 Chevrolet, for that matter. Says Wagoner, a tad defensively: "If you look at how our products have done, not with the buff books and auto writers but with the people they are intended for, we've had a lot of success."

With 82 models of cars and trucks (counting two-doors and four-doors separately), GM is a big believer in shelf space. Rather than building a single family sedan--like a Toyota Camry or Honda Accord--that sells 350,000 units a year, GM sells four sedans aimed at different segments, ranging from conservative to sporty. Says di Giovanni: "People have been compromising their tastes to buy a Camry just to get the quality. If we can provide the same quality, we will supply a car that exactly meets their needs."

Oldsmobile will be the acid test of this theory. In rebuilding the brand, NAO has decided to abandon the aging owner base of Olds and go after a new group of younger customers who usually buy imported cars. That's quite a face-lift for a dowager division that celebrates her 100th anniversary on August 21, and GM concedes that it will be difficult. "Olds had to reposition away from a portfolio of customers who are going away," says dapper marketing czar Ron Zarrella, 47. "And they need to get a whole range of new customers. We think we've got the product right. The question is: Can we get the marketing and selling right?"

Two of the products that will determine Oldsmobile's future are parked in GM's design studios and have never been seen in public. They are the compact Alero, which replaces the underachieving Achieva, and a $27,000 sedan code-named Antares. They are attractive cars, but after seeing them, I wonder if they will draw shoppers away from Honda and Lexus, especially if they carry an Oldsmobile badge. More clues to Oldsmobile's future reside in a nearby alcove, one of seven Brand Character Centers that GM has created to keep the identity of each division straight. Posted on the walls are pictures of old cars, close-ups of design details like headlights and wheel covers, and sketches of future models.

Defining the brand character of a division like Pontiac, with its boy-racer history, is simple compared with creating an image for Oldsmobile. GM's marketers have selected four adjectives they want the new Olds to embody: thoughtful, precise, international, and rewarding. Try translating that into sheet metal. In fact, new Oldsmobiles will resemble the sharply styled Aurora, a $36,000 sedan that, ironically, was praised by auto writers but neglected by customers. The new cars will all have bulges above the wheel wells, a thin, horizontal accent strip along each side, and six-spoke wheels. In an interesting ploy to position Olds against its foreign competition, GM is equipping each model a bit better than normal and pricing it a bit higher as well. That can work for a powerful marque like BMW. But should BMW and Oldsmobile even be spoken of in the same breath?

I got a graphic demonstration of how GM plans to differentiate Olds from other brands by looking inside another secret room. I saw a clay model of the 1999 Olds Bravada sport utility being sculpted alongside a Chevy Blazer and a GMC Jimmy, which share the same basic mechanical structure. While the assigned mission of the Blazer is to look "practical," "durable," and "trucklike," and the Jimmy is a somewhat flamboyant luxury version, the Bravada is defined as "refined," "more carlike" in appearance and ride, and "understated" in its luxury. I wonder if those could turn out to be distinctions without a difference.

It's one thing to design a car and quite another to engineer it so that customers get it quickly. A chart in the Rigorous Tracking Room shows GM's progress. In 1992 the company needed 42 months to start production on a new model once the final design had been set, vs. 31 months for Toyota. Now GM can do it all in 31 months, but Toyota has hardly been standing still: It has reduced product development time to 26 months. Five years from now GM's announced goal is 18 months, about where it expects Toyota to be as well. That's a big leap, as Wagoner concedes. "We'll be depending on a lot of computer simulation," he says. "We can't crash vehicles against a wall and still meet these schedules."

Like everything else at GM, product development suffered from multiple processes and personalities. Lessons learned in one program seldom appeared in succeeding ones, because nobody was reading from the same page, much less the same book. GM's effort to create a common product development system, which is still only 80% complete, is compiled on a giant chart I saw in the Customer Satisfaction chamber. The chart identifies 39 "enablers"--common processes that shorten program lead time and improve quality. The enablers tend to have jaw-bending names like "develop component/subsystem sourcing model," but they represent critical operations that are closely monitored. To ensure accountability, each enabler has an individual's name attached to it. A green circle appears on the chart if everything is running smoothly; a red circle when things are not. Engineers like to say that "red is good," because it means a problem has been detected. However, too many red circles draw the attention of executives as high as Wagoner. When I saw the chart, greens outnumbered reds by about seven to one.

Even though NAO is seeing more green these days, getting production of a new car up to full assembly-line speed requires 32 months--if things go smoothly. Honda changes models over a weekend by doing more preproduction planning. While GM dealers wait for the new cars, they have to keep selling old ones that must be heavily discounted and tarnish GM's image. The company only recently retired the midsize Buick Century and Olds Ciera, which dated to 1982, and it still sells the compact Buick Skylark and Olds Achieva, even though they would make swell patients for Dr. Kevorkian. Simple tasks, like renaming a vehicle, take an eon. The origins of "Jimmy" on the GMC sport-utility vehicle date back to World War II, and its customer equity approximates zero, but GM is only now getting around to replacing it with "Envoy." Says Zarrella: "I don't think we ought to just cut it [the Jimmy name] off. As we differentiate the product from the other ones, it makes sense to reposition the brand name."

If 1997 is going to be a great year at NAO, as Smith insists, it had better get going. Sales fell sharply in February, and GM's market share for the first two months of 1997 was 30.3%, down from 32.5% for the same period a year ago. GM blames the decline on all those "unavailable" new models. The Cadillac Catera, introduced with fanfare in November as "the Caddy that zigs," has been zagging at dealers, so GM cut the lease price to $399 a month. Labor unrest continues apace, while some 40 UAW locals negotiate plant agreements in the wake of the national contract signed last fall.

GM's market share may indeed climb back to 33% this year, as executives predict. But barring a major collapse by one of its competitors, or an unexpected explosion in new-car sales, it is difficult to see North American Operations soon achieving Jack Smith's oft-stated goal of a 5% return on sales. Despite all the smart, hard-working people running it, NAO remains burdened by its past and hamstrung by the powerful UAW and dealer groups. It sounds oxymoronic, but the future of NAO may lie overseas. GM is thinking of ways to integrate future Chevies, Pontiacs, and Saturns with German-made Opels to achieve greater economies of scale. A peek into the last secret room reveals a compact Pontiac Sunfire GM is designing alongside four other brands--some from outside the U.S.--as part of a global small-car program for the early 21st century.

That seems a long way off. Fixing GM has turned into a decade-long project that few people--investors, analysts, or company executives--could have foreseen. The question is, once it's fixed, what kind of car company will it be: outstanding or merely average?