The Numerati

by Stephen Baker

  Back when Dantzig was putting the final touches on his algorithm, IBM researchers were already preparing to apply operations research to their own business. They had the mother of all tests for it: IBM’s massive supply chain. To build its renowned office machines (which didn’t yet include commercial computers), IBM bought parts and raw materials from suppliers all over the world. Naturally, these were a major expense. If the company could use this new math to organize it all, the savings would drop straight to the bottom line.

  The math worked. In fact, IBM was able to turn this particular know-how into a business. The company’s experts helped other companies convert their own logistics into math and then optimize them. This is where the story turns inside out, a bit like that drawing by M. C. Escher, where the artist’s hand is drawing itself. In the past couple of decades, IBM’s focus moved from manufacturing to services. The company now sells more expertise than machinery. It unloaded its personal computer division to China’s Lenovo in 2005, and IBM Global Services has grown into a $40 billion business. So if IBM’s experts were to optimize their supply chain today, they would have to model and fine-tune themselves. That’s precisely what Takriti’s team is busy doing.

  Just think where this could lead. We’ve seen, with supply chains, how the company used itself as a laboratory. It mastered the process for itself and then sold the expertise to others. Now the company is modeling its workers. If this leads to big gains in productivity, do you think that expertise will remain locked up inside Big Blue? I don’t. Imagine mathematical modelers arriving at the doors of your company one day, either as a phalanx of blue-clad consultants or perhaps encoded in a piece of software. Their focus will be on you.

  SITTING IN HIS small office, one blue-jeaned leg crossed over the other, Samer Takriti confesses to me that he’s nervous. I can’t blame him. His assignment is to construct detailed mathematical models of 50,000 of his colleagues. We’re not talking about simply placing workers and their jobs into the kind of bare-bones taxonomy we described earlier. That’s complicated enough. The goal here is to build entire models, complete with a person’s quirks, daily commute, and allies and enemies. These models might one day include whether the workers eat beef or pork, how seriously they take the Sabbath, whether a bee sting or a peanut sauce could lay them low. No doubt, some of them thrive even in the filthy air in Beijing or Mexico City, while others wheeze. If so, the models would eventually include this detail, among countless others. Takriti’s job is to depict flesh-and-blood humans as math.

  Takriti is not given to bold forecasts. But if his system is successful, here’s how it will work: Picture an IBM manager who gets an assignment to send a team of five to set up a call center in Manila. She sits down at the computer and fills out a form. It’s almost like booking a vacation online. She puts in the dates and clicks on menus to describe the job and the skills needed. Perhaps she stipulates the ideal budget range. The results come back, recommending a particular team. All the skills are represented. Maybe three of the five people have a history of working together smoothly. They all have passports and live near airports with direct flights to Manila. One of them even speaks Tagalog. Everything looks fine, except for one line that’s highlighted in red. The budget. It’s $40,000 over! The manager sees that the computer architect on the team is a veritable luminary, a guy who gets written up in the trade press. Sure, he’s a 98.7 percent fit for the job, but he costs $1,000 an hour. It’s as if she shopped for a weekend getaway in Paris and wound up with a penthouse suite at the Ritz.

  Hmmm. The manager asks the system for a cheaper architect. New options come back. One is a new 29-year-old consultant based in India who costs only $85 per hour. That would certainly patch the hole in the budget. Unfortunately, he’s only a 69 percent fit for the job. Still, he can handle it, according to the computer, if he gets two weeks of training. Can the job be delayed?

  This is management in a world run by Numerati. As IBM sees it, the company has little choice. The work force is too big, the world too vast and complicated for managers to get a grip on their workers the old-fashioned way—by talking to people who know people who know people. Word of mouth is too foggy and slow for the global economy. Personal connections are too constricted. Managers need the zip of automation to unearth a consultant in New Delhi, just the way a generation ago they located a shipment of condensers in Topeka. For this to work, the consultant—just like the condensers—must be represented as a series of numbers.

  To put together these profiles, Takriti requires mountains of facts about each employee. He has unleashed a squadron of Ph.D.’s, from data miners and statisticians to anthropologists, to comb through workers’ data. Personnel files, which include annual evaluations, are off limits at IBM. But practically every other bit of data is fair game. Sifting through résumés and project records, the team can assemble a profile of each worker’s skills and experience. Online calendars show how employees use their time and who they meet with. By tracking the use of cell phones and handheld computers, Takriti’s researchers may be able to map the workers’ movements. Call records and e-mails define the social networks of each consultant. Who do they copy on their e-mails? Do they send blind copies to anyone? These hidden messages could point to the growth of informal networks within the company. They may show that a midlevel manager is quietly leading an important group of colleagues—and that his boss is out of the loop. Maybe those two should switch jobs.

  The interpretation of our social networks is an exploding field of research, from IBM to the terror-trackers at the National Security Agency in Fort Meade, Maryland. One leading lab is at Carnegie Mellon University, in Pittsburgh, where a professor named Kathleen Carley is building an entire social network empire within the computer sciences department. When I meet with Carley, she has 30 grad students jammed into just a handful of basement offices. They’re analyzing the networks of contagious diseases, such as Asian flu. They’re comparing the dynamics of different networks in the Middle East.

  What can this social network analysis divulge about workers at IBM or elsewhere? Lots. Start with e-mail. Carley’s grad students can feed a computer all of a company’s e-mails over a certain period of time. They practice with the e-mails exchanged during the frantic dying months of Enron. Released as evidence in Enron trials, they’ve been dissected ever since by social network researchers around the world. Carley’s system notes the senders of e-mails, the time the messages were sent, and the recipients. Without even reading the content of the e-mails, a software program her team has built draws various diagrams of the organization. One of them shows who communicates with whom. When she shows it to me, it looks at first like a spaghetti cook-off. The organization—if you can call it that—features different tangled piles, each one with its own set of meatballs. Single noodles extend from one pile to the others. Each meatball, naturally, represents a person within the organization, and the piles represent groups of people who communicate heavily with one another.

  Logical enough. The finance people, the gas people, the legal team, they all communicate within their groups, with an occasional e-mail to another department. But it’s not that simple. “See this group here?” Carley says, pointing to a cluster of meatballs in a swirling mass. It’s an informal network, she says. It took shape as Enron collapsed. This group sent out about one thousand messages a day and became a clearinghouse for inside dope. If the company had been studying this network, executives might have interpreted it as an insurgency taking shape. In a sense, it was, since a growing network of employees was trading ever more dire reports and rumors about a company in crisis—and helping one another prepare for life after Enron.

  Corporations elsewhere, including IBM, can draw all kinds of insights from their employees’ networks. They can map each person’s circle of contacts. They can also spot outliers, people who aren’t communicating much with anyone. These employees, Carley says, are worth scrutinizing: they may be depressed or about to leave, or even consorting with the
competition. Even without reading all the e-mails, the company can automatically spot the most common words that circulate within each group. This permits them to map not only each worker’s contacts but also the nature of those links. They can also see how communications shift with time. Two workers may discuss software programming Tuesday through Friday but spend much of their time on Monday sending e-mails about the past weekend’s football games. “The next big step,” Carley says (a bit ominously), “is to take tools like this and tie them to scheduling and productivity programs.” I read this to mean that we office workers are well on our way to being optimized.

  Sound scary? It may depend on where you’re perched on the food chain. Remember the $1,000-per-hour consultant who almost got dispatched to the Philippines? He didn’t end up going, and instead, in IBM’s scheme, he remained “on the bench.” Takriti smiles. “That’s what we call it,” he says. “I think the term comes from sports.” The question, of course, is how long IBM wants to have that high-priced talent sitting on the bench. If there isn’t any work to justify his immense talents, shouldn’t they put him on something else, just to keep him busy?

  Not necessarily, says Takriti. Job satisfaction is one of his system’s constraints. If workers get angry or bored to tears, their productivity is bound to plummet. The automatic manager keeps this in mind (in a manner of speaking). As you might expect, it deals very gently with superstars. Since they make lots of money for the company during short bursts of activity, they get plenty of time on the bench. But grunt workers in this hierarchy get far less consideration. They’re calculated as “commodities.” Their skills are “fungible.” This means these workers are virtually indistinguishable from others, whether they’re in India or Uruguay. They contribute little to profits. It pains Takriti to say this, because humans are not machines. They have varying skills and potential to grow. He appreciates this. But looking at it mathematically, he says, the company should keep its commodity workers laboring as close as possible to 100 percent of the time. Not much kickback time on the bench for them.

  Where is this all leading? I pose the question one afternoon to Pierre Haren. A Ph.D. from Massachusetts Institute of Technology and a prominent member of the Numerati, he’s the founder and chief executive of ILOG. It’s a French company that uses operations research to fine-tune industrial systems, charting, for example, the most efficient delivery routes for Coors beer. ILOG makes allowances for all kinds of constraints. For example, a few years ago, the Singapore government wanted to avoid diplomatic spats at its new airport. So officials asked ILOG to synchronize the flow of passengers, making sure that those from mainland China wouldn’t cross paths with travelers from Taiwan. Haren speaks in a strong French accent. We’re talking in the lobby of a Midtown hotel in New York, and he has to yell to make himself heard over a particularly loud fountain.

  Haren says that the efforts underway at places like IBM will not only break down each worker into sets of skills and knowledge. The same systems will also divide their days and weeks into small periods of time—hours, half-hours, eventually even minutes. At the same time, the jobs that have to be done, whether it’s building a software program or designing an airliner, are also broken down into tiny steps. In this sense, Haren might as well be describing the industrial engineering that led to assembly lines a century ago. Big jobs are parsed into thousands of tasks and divided among many workers. But the work Haren is discussing is not done by hand, hydraulic presses, or even robots. It flows from the brain. The labor is defined by knowledge and ideas. As he sees it, that expertise will be tapped minute by minute across the world. This job sharing is already starting to happen, as companies break up projects and move big pieces of them offshore. But once the workers are represented as mathematical models, it will be far easier to break down their days into billable minutes and send their smarts to fulfill jobs all over the world.

  Consider IBM’s superstar consultant. He’s roused off the bench, whether he’s on a ski lift at St. Moritz or leading a seminar at Armonk. He reaches into his pocket and sees a message asking for ten minutes of his precious time. He might know just the right algorithm, or perhaps a contact or a customer. Maybe he sends back word that he’s busy. (He’s a star, after all.) But if he takes part, he assumes his place in what Haren calls a virtual assembly line. “This is the equivalent of the industrial revolution for white-collar workers,” Haren says.

  Some of us like to think that our work is too creative to be measured and modeled. I used to feel this way. For years I would write articles, and the only metric that mattered was whether the editor in chief appreciated them. Things began to change when the articles moved online. This made it possible for managers to count how many people read each article. Some managers these days rank writers by page views or how many times each article is e-mailed by readers. Is this fair? Not in my view. I remember one time a colleague posted on his blog a video ad featuring Paris Hilton. She wasn’t wearing much, and she was washing a car with a big wet sponge in a splashy and provocative fashion. His blog attracted tens of thousands of visits that day, more than others of us got in a month. Did he outperform us? It depends on what the bosses decide to count. As the Numerati gain sway in the workplace, such questions are bound to rage.

  It’s getting late in Takriti’s office. I can see that he’s concerned about my line of questioning. This virtual assembly line sounds menacing. The surveillance has more than a whiff of Big Brother. For those of us who aren’t Carlos Beltran or a $1,000-per-hour consultant, life as a mathematical model is sounding like abject data serfdom.

  Here’s Takriti’s counterargument. As the tools he’s building make workers more productive, the market will reward them. (So there’s an economic benefit, even to us serfs.) What’s more, workers will increasingly use their numbers to open doors for themselves. We already use math programs to map our trips and look for dates. Why not use them to map our careers—and negotiate for better pay? Let’s say analytical tools show that a consultant’s value to the company topped $2 million one year. Shouldn’t she have access to that number and be free to use it as a negotiating tool? In a workplace defined by metrics, even those of us who like to think that we’re beyond measurement will face growing pressure to build our case with numbers of our own.

  Chapter 2

  * * *

  Shopper

  THE CALL COMES from my wife at the supermarket. “Do we need onions?”

  I check. “We have one big one,” I say, turning it over gingerly. “But it’s been sprouting for a while . . .”

  “Okay, I’ll get some. How about milk?”

  You know the routine. A few minutes later, whichever one of us is shopping arrives at the checkout counter. There, if we remember, we dig into a pocket or purse for the frayed customer loyalty card on the key chain. The cashier scans it. We get a discount on the orange juice or razor blades, and the supermarket learns about everything we buy. It’s a deal we shoppers have been making for years. Stores give us what amounts to a couple bucks a week in exchange for our shopping lists.

  Here’s the strange part. To date, retailers have stockpiled untold mountains of our personal data, but they’re only now waking up to what they can do with it. Sure, managers have used the scans to keep an eye on inventory. They can see when to order more mangoes or Snickers bars. They’ve learned plenty about our behavior en masse but next to nothing about us as individuals. When we walk into a store, even if it’s the hundredth time this year, the system doesn’t recognize us. It’s clueless.

  This era is coming to an end. Retailers simply cannot afford to keep herding us blindly through stores and malls, flashing discounts on Pampers to widowers in wheelchairs and ham hocks to Jews who keep kosher. It’s wasteful, and competitors are getting smarter. Look online. Whether it’s Amazon.com or a travel service like Orbitz, Internet merchants are working every day to figure us out.

  They’re tracking every click on their sites. They know where we come from, what we buy,
how much we spend, which advertisements we see. They even know which ones we linger over for a moment or two with our mouse. In the online world, businesses no longer look at us as herds but as vast collections of individuals—each of us represented by scores of equations. They prove every day that merchants who know their customers have a big edge. They can study our patterns of consumption, anticipate our appetites, and entice us to spend money.

  Personal service is nothing new for retailers. For centuries, it’s been a privilege for the rich. Shopkeepers and tailors know their names and measurements and their taste in premier cru burgundies. They also know where to send the bill. A few generations ago, the rest of us got personal service (on a far more modest scale) in our own neighborhoods. “The retail model was a shopkeeper, a millinery, a rug merchant,” says Jeff Smith, a managing partner of the retail practice at Accenture, the tech consulting giant. “You didn’t serve yourself,” he says. “They stood behind counters and found what you were looking for.” Chummy relations with customers gave these merchants an edge.