Reader's Digest: The Longest Goodbye

By Tom Steinert-Threlkeld  |  Posted 2005-07-08

"This is our last control tree for a weekend release. In it, we have XNB, we have the nonpromote extract, we have the sampling of the PSLO and the UMD Score files to help MSR. We have the XNB scoring tree, we have the XNB July direct mail, we have the extract for eUMD, the extract of telephone do-not-promotes, we have our typical universe counts and we have the list rental monthly. And I believe that's it. Everything else is going or has gone into MSR. So with that, I'm going to forward on to the data center in New Jersey and whenever the resources allow, it'll start. And that'll be our last P-File."

With that, applause broke out in the small, unassuming conference room along Pegasus Alley at Reader's Digest headquarters in Pleasantville, N.Y.

Tapping a few keystrokes on his laptop computer, database boss Joe Devanny had instructed the company's outsourcing service in Leonia, N.J., to perform "the last big gang run"—the selection of millions of names for a series of marketing campaigns designed to generate new business for the company, particularly at Christmas this year.

This was the curtain call for what most likely had been the world's first large, comprehensive system for tracking and managing every aspect of how a company dealt with—and marketed to—its customers.

Today, the practice is called "database marketing,'' or, in some circumstances, "customer relationship management." Whichever name it would be called by, "it may have been the largest'' such system at the time, according to Frederick Newell, marketing consultant and author of

At Reader's Digest, it was called, simply, the Unified File System.

The first system to unify tens of millions of names and addresses from different businesses into one file—and then, add a record of every marketing contact made with each customer. And then, add a record of every product sent to the customer. And then, a record of every payment, as well. Decades before the idea took hold that there was such a thing, The Reader's Digest Association was compiling a "360-degree view" of each of its customers—and allowing its direct marketing operation to figure out how much profit a campaign might produce, before the first envelope was sent out. In a time when years of information had to be crammed into the space of two letters, in a computer.

With Devanny's keystrokes on May 12, the UFS was finally being phased out, in front of a small gathering of marketing, quantitative analysis and technology staff nestled into the Pegasus C room. Corks on champagne bottles popped.

It had taken 17 years to unwind this system and replace it with one based on contemporary technology that present-day programmers could work on and enhance.

In the year the system was born—1969—a mainframe computer could juggle only 64,000 bytes of information in its brain. Customer records had to be kept in the obscurity of the mathematics of "base 16" to make the most use of limited memory. That made it possible to squeeze the date a person became a customer, the products offered to that person, and what he ordered, paid and returned into fields that might store just one or two characters of information.

That the Digest's system had lasted 35 years was partly due to the foresight of its original 30-person programming team, which left key rules, prices and data in the marketing engine open to change by computer users—a radical thought in the early days of "big iron.''

But it was also a testament to the constrictions of a complex, homegrown ball of code that only its Reader's Digest in-house devotees could understand, read and interpret, as the system evolved, struggled and ultimately became what Jeanne Plekoh, an associate director of customer technologies, would call a "cranky, brittle, elegant old lady.''

This, then, is the story of that cranky old lady, who labored in anonymity, cheated death for decades and helped turn Reader's Digest into one of the world's most famous publishers. It's the story of a digital dowager who, in her earliest days, pioneered the concepts of tracking customer behavior, customer response, cross-selling, and the basic marketing principle adapted generations later in the online world by entrepreneur Jeff Bezos: "If you like that, you probably will like this."

But it is also the story of the ladies of Pleasantville itself, who were trucked in by the busload to enter new orders for magazines, books and music, as the company built a system that would keep records on 50 million or more households in a single file—in a time when the entire country had fewer than 60 million. It is the story of the women and men in the quiet halls with the art-festooned walls and in the data center in the basement of Reader's Digest's ivy-covered headquarters, who made history, but no history books, when they gave birth to an invisible lady in the year man first landed on the moon. And kept her alive—perhaps as the world's oldest living system—until that afternoon of May 12, when Devanny told his coterie of cohorts in Pegasus C, "This is it."

Genetic Enemies

It was time to replace another cranky system. The year was 1964, and the system was a Univac II computer, the company's workhorse since 1958.

Reader's Digest had purchased one of the first Univacs in the world. In fact, Reader's Digest was buying the computers when there were fewer than 100 extant, at about $1 million a crack—almost $7 million each, in today's dollars.

Even at that price, the Univac could store only 10,000 words in its memory, and divide a few hundred numbers in a second. Sandy Koufax' fastball crossed home plate faster.

As long as it had to be replaced, systems director John Moore and programming manager Robert P. Otten had this idea: Combine all information that the company held on its magazine, condensed book and special product customers into one file. Then combine that with a history of every piece of mail or promotion sent to each customer.

"The real genius was to put data on what was mailed to anyone, into the system,'' says Robert Burns, not the Scottish poet but one of the lead developers of what would come to be called UFS.

The task in front of the developers was not as simple as the idea. Keeping track of Reader's Digest customers was akin to keeping track of the American population for the Census Bureau. By 1965, the magazine had 10 million subscribers. That put a copy of the publication in more than 1 of every 6 homes in America.

But its marketing reach was even more extensive. The number of households in its files at that time totaled 50 million. According to the Census Bureau, there were only 57.4 million households in the country.

Putting data into a new electronic filing cabinet would not be easy. Many customer records were kept on metal stencils with tiny holes punched in them. To "pull" the records on a given set of customers to whom you wanted to send a mailing, you would stick a long "chopstick" through a hole and, literally, pull the right stencils out of the stack.

For turning those records into digits, punch cards were the order of the day; and the process of getting customer information on the cards was something of a full-employment act for the women of Pleasantville. Women willing to work full-time, as well as housewives with some spare time before kids and husbands came home for the evening, became data entry clerks. About 150 would spend their days heads-down, operating key-punch machines, entering new orders, making changes in addresses and the like.

If they didn't have cars or didn't drive, no matter. Reader's Digest would pick them up in buses that ran routes around the town of 7,110 and elsewhere in Westchester County to bring them to work.

Their job was to put as many holes into as many cards as they could. The cards made it possible to turn customer records into bits of digital data. Which meant the information could be used to do more than just create mailing labels, even if labels were the end output.

Back then, it was genius enough to think of putting what marketing piece was mailed to whom, and when, into a database. If all the data about a customer could be kept in a single file, you could know who bought what, who responded to promotions, who paid promptly and who didn't. And who would make a good prospect for more of what you sold.

Moore and Otten's idea was to merge the records of the magazine business with those of the condensed book business and the single-sales group, which sold one-shot music and book products like Music of the World's Great Composers or the Reader's Digest Great World Atlas.

Moore was in charge of developing the software "systems" that could improve operations at Reader's Digest. But first he had to convince Bob Frankel, his opposite number on the hardware side, that a single file of customer information was worth the effort.

He wouldn't really have to convince Frankel to phase out the Univac. That was overdue. But Frankel made hardware decisions, and competition was keen in the computer business. Sperry, which made the Univac, was still a big name. So were RCA and Honeywell.

Moore wanted a new machine from IBM, the 360, the first standards-based computer designed for commercial information processing. But Moore and Frankel were "genetic enemies,'' according to Burns.

Frankel was in charge of the part of Reader's Digest that fulfilled orders for magazines, books and other products. Since he owned the computers, he figured he should run the systems department, Burns recalls.

But Moore did that and had to serve the marketing department, not just the fulfillment operation, along the way. Making matters worse: They were polar opposites in character. Moore had an advanced degree from a major East Coast technical institute, was Irish and known to enjoy a drink or two, Burns says. Frankel, by contrast, was viewed as a buttoned-up "CPA-type of a guy.''

"It was an oil-and-water sort of deal,'' Burns says.

Moore and Otten would need to get the backing of operations vice president Kent Rhodes and marketing chief Walter Hitesman, two men who would later become president of the company. But first, Frankel would have to go along.

So Burns and Moore waltzed down to the fulfillment conference room, on the ground floor, next to the garden—the same room once spied by a visiting set of old ladies who wisecracked, one to the other, in "Laughter Is the Best Medicine" style, "Oh, Helen, let's go in there and get fulfilled!''

Moore made his pitch. But he got stopped by Frankel, who thought what Moore was proposing was underpowered, according to Burns. The most a million-dollar IBM 360 could hold in memory were those 64,000 characters of information. That's the same maximum storage a Sinclair personal computer would have two decades later, at a starting list price of $99.95.

"We thought that was tremendous,'' Otten says. But, "it was a drop in the bucket. We needed three or four times that much.''

Frankel would go along with Moore's plan—when Moore agreed to boost the number of machines that would be bought and include a development staff of at least 16 programmers. Final approval came from Rhodes; Otten became the project's manager. Burns, Ed Adolph, Ruth Ritchie and Betty Kahrs would become the builders and dressers of the elegant lady, as her lead developers.

By the time the new filing system was launched at the end of 1969, Reader's Digest would possess three IBM 360/30 computers and three copies of the Model 40. The coding, programming, systems work, support work and coordination would consume 150,000 work hours. And $30 million of Reader's Digest cash—roughly the equivalent of $158 million today.

Moore had stuck to his guns, insisting that the value of knowing how customers would react to marketing appeals far outweighed the initial cost of the technology, even if the cost of the hardware doubled or tripled.

"It was very courageous to make a statement, back then, that information was worth a lot,'' Burns says. "That was the true vision.''

A Hex in Computing

The practical reality, in the early days, was much more prosaic.

There wasn't much discussion of "data warehousing" or "data mining" or anything you might want to call "business intelligence.''

In fact, before the Univac II came along, Reader's Digest's data warehouse consisted of 18 million stencils, little metal plates with subscribers' names, addresses and expiration dates on the front. They were used to create mailing labels, by pressing ink on them. Their edges were notched, to add marketing information to stencils selected for marketing campaigns.

Several rooms in the company's headquarters were devoted to this "prehistoric" system, as Otten terms it. About 100 women and men would toil in a stencil room, making sure each stencil was in the right sequence in the right tray for the right postal code. And, once removed, returned to its right place.

What they couldn't do was easily put customers in buckets they could do something with, like sell customers a new book or record. Or just simply put names in alphabetical order without shuffling cards by hand.

With the new file system and the battery of IBM 360s, they finally had a way of putting order into the universe.

"It was wonderfully awesome to do a sort of 10 million names,'' Burns says.

Being able to sort millions of records by name, state or street address was not the point. Figuring out what prompted each customer to buy more products was the mystery worth solving.

For Burns and 29 other programmers, that meant devising a schema that would compact a record of any offer made to any customer in an "atomic record" of four bytes per event. One byte would record the name of the product; a second the action that resulted (promoted, paid bill, canceled, etc.); a third the type of marketing effort (direct mail piece, house ad, etc.) that spurred the action; and another the month and year of the mailing or other marketing campaign.

Each byte mattered in an era of expensive hardware and expensive memory. IBM had spent $5 billion in 1964 ($28 billion in today's dollars) just to launch the 360.

In fact, Reader's Digest programmers had to figure out how to squeeze lots of information into fields that might only be one byte long. That meant writing in hexadecimal code, an approach taken to maximize use of the limited memory of the IBM machines. "That was the nature of the beast,'' Otten says.

Everyday math is based on decimal code: the numbers 0 through 9. The base is 10: the characters you know as numbers.

Hexadecimal code takes those 10 numbers and adds six letters. Readers' Digest chose A through F.

With decimal code, you can store only 100 different values in two digits: 10 times 10. With hexadecimal code, you can store 256 different values: 16 times 16.

Which happens to be the maximum amount of information that can be stored in a byte. IBM, with the 360, established the standard in the computing industry that a byte would be the equivalent of eight bits of information fed to a machine at a time. Those bits were ones or zeroes. Two different values, eight digits, multiplied into all their possible permutations equals—256.

So, November 1987 would become 1C. November 1994 would be 70 and November 2004 was E8. More than two decades of months and years could be captured in 256 two-character combinations.

How could you tell what E8 meant? By looking it up in a table, kept on paper. Or in one's head.

Kahrs and Ritchie, the two lead developers, would define much of the foundation of the system, such as what values to put in the compacted fields. Did you need hex codes for active customers? Expired? Deadbeat? Temporary? Gift recipient?

But everyday "users" of the system wouldn't have to know hex code, project manager Otten decided. Key Reader's Digest policies, such as when to stop shipping products to a particular customer, subscription rates and who was entitled to which rate, would be kept in tables that could be pulled up on screen, altered and fed back into the system.

That separation of business purpose—and putting it on screen in a form an everyday worker could see and deal with—was "unique" in a period when only gods working in air-conditioned rooms with raised floors could be experts in computing, according to Burns.

"To think of the system user was quite advanced,'' he says. Until then, dabbling in hex code or putting it to any kind of use "was strictly up the Mr. Wizards and the Mrs. Wizards."

Need for Speed

Even with information so compressed and batches of IBM computers on the job, dealing with tens of millions of subscribers and potential customers for c ondensed books, music and tapes was slow and tedious.

"In the beginning, we blew out the memory'' of the IBM machines, Otten recalls.

Sorting could be difficult. But figuring out how to deal with a cancelled charge could consume lots of code and computational time.

To make efficient use of that time, Reader's Digest came up with a filing system and method of accessing that information that were entirely its own—and still in place until the last pass at customer records stored in the "elegant old lady" took place in May.

Customer records were kept in files of great precision but little flexibility (see chart, p. 46). Each character, each byte, would mean something, in sequence. These long "flat" files might have 25 rows of information, containing everything from a unique identifier for each customer, known as a match code, to the designation of the last mailing of an offer for a cookbook sent to that household. On top of that, each segment, whether it be the master history, the promotion history or the mailing history, could be of vastly different lengths, depending on the activity in each.

Changing the sequence could have disastrous results. Marketing, circulation and payment processing programs all depended on those long files. Each would look for information to appear in a fixed sequence, so that each could draw out just the pieces of information needed, such as the date a payment was received or the mail route that included the household in question.

Even into the 1990s, this was no small matter. The company wanted to begin adding telephone marketing of products, on top of its marketing by mail. But the file had no space allocated to telephone marketing. The sequence of information in the file would have to be changed. All applications would have to be re-trained to look for information in new locations and be able to recognize it when they saw it.

Adapting the system to a new product line—Plekoh likens it to "spelunking" in and out of its crevices—might take three or four months, to change all the records to fit the new data; to make sure all applications that drew on the records knew where to find a new designation and act on it; and to change the hardware.

But in the '60s, the biggest issue was just being able to get at information on customers rapidly, to be able to pull records wanted for a specific marketing campaign and sort accordingly.

An IBM 360 mainframe computer ranged in cost from $831,250 to $34.4 million, in today's dollars. Yet its brain only ticked off 1.33 million tasks a second. Today, a $1,000 IBM ThinkPad laptop ticks off 1.6 billion tasks in the same time.

So, Moore and Otten's programmers made a radical leap to speed up the retrieval of customer information, then stored on magnetic tapes. They would forego the creation or use of IBM 's own application for retrieving data. They would even bypass the operating system. They created a proprietary data retrieval system called the Reader's Digest Access Method that allowed technicians to navigate the customer records. The direct access to the record made it possible to select names for a mailing or do tedious chores such as fixing a problematic mailing label.

Even with a unique and systematic file system and a direct-to-the-data method of grabbing information, managing an electronic library that stored more than 50 million names was not a snap. Updating the Unified File System with new subscribers, changes of address and the like would take four days to complete.

The UFS, though, was cranky, even from the start. Results would be run off five tapes and records would be lost. But the same five tapes could be run again, and different errors would result from the playback. "You know that's kind of scary,'' Otten says. "When you get different results, you know you're in trouble."

A "file maintenance" run, where updates and changes were made to records, would take 125 hours instead of 15 to 20. It took a crackerjack troubleshooter from IBM to find and solve the problem: a logic glitch in the tape drive. "It put 20 years on my life,'' Otten recalls. "It was so frustrating.''

But after the first year, the lady was ready to show off her hex-code-crunching finery. Her real attraction to stat jockeys such as Charlotte Nester would be her ability to determine who was most likely to respond to a given pitch.


Age. Income. Magazine subscriber. Recency of payment. Number of times bought from a particular series of books.

They had been just statistics that publishers had collected. But to Nester and the quant jocks who preceded her at Reader's Digest, they and hundreds of variables like them were a means to an end: making predictions.

Nester is director of database marketing. Her job is to help Reader's Digest make more money by mailing less. To get $600,000 of new revenue from a campaign sent to 1.6 million potential buyers, instead of $500,000 from one sent to 2 million.

That means finding those names on record who are most likely to respond to a particular promotion.

With the advent of a single database that would keep records on every customer Reader's Digest had, that would be seriously possible, nearly five decades after founder DeWitt Wallace started the Digest in the basement of a New York speakeasy, printed 5,000 copies of the first edition and had the perspicacity to buy a mailing list to send potential subscribers a sales pitch.

The brazenness of the Unified File System was its simple devotion to comprehensive information. Sure, the names and addresses of present and past customers would go into it. But so would their payment history, and their mailing history.

But Reader's Digest's marketers and circulation managers weren't keeping track of whether each issue showed up in mailboxes nationwide. They were instead keeping track of every mailing campaign they would conduct, now as many as 630 a year, as the system got its legs.

They would keep track of each campaign in a shorthand code, and mark off, in another field, if there was a response. They would also keep track of how fast the company got paid, after that.

Nester's job would then be to use a form of mathematics known as regression analysis to figure out which variables actually mattered. Such analysis assumes that there are significant relationships between variables. If the relationship is strong, the result can be depicted in a straight line—and knowing where one of the variables, say, age, falls on the line can help predict where the other, say, spending, also falls. If a variable was particularly useful at predicting what you wanted—for example, the propensity to open red mailing envelopes—it was lauded for its ability to "discriminate."

Even when the magazine, condensed book and one-shot businesses had separate databases, variables were analyzed. But the only variable in a magazine record might be when the subscription ended. Pulling together records of all businesses, as well as payment history, into one file vastly expanded the variables that could be analyzed. "It was like quadrupling the size of your dictionary,'' Otten explains.

Nester could use all the data to test different models of customer behavior. The trick would be to rank each variable, to find which were the best predictors. And then which combinations could really make a difference. Age by itself might give some insight into the willingness to buy a condensed book. But age and time since last order might be a whole lot more telling.

"I'm not a gambler,'' says Nester, who came to Reader's Digest 24 years ago. "I'm a statistician.''

Recency of an order, frequency of an order and the amount spent all come into play. Out of the constant iterating of models would, at some point in the '70s, popout one particularly useful predictor: Promotions Since Last Order.

The PSLO helped separate the wheat from the chaff. If a person, on average, bought one tape a year, and a year had gone by, it was time to send a promotion to that person. If the person had just bought, it was time to back off. Sending a new piece of mail was very likely a waste.

When uncovered, the reaction among programmers and quantitative analysts at Reader's Digest was, "'Wow, that baby really discriminates,'" Burns recalls.

Before Nester's forerunners started applying regression analysis, the term "PSLO" didn't even exist at Reader's Digest. But as the analysis got into full swing, such benchmarks would be codified in little reusable bits of code, called "terms."

Each term would measure customer activity in some way, such as total number of purchases from a product line or total number of payments made in a product line.

The terms would make it easier for non-technical staff, like marketing director Kathy Gilbert Haggerty, to run campaigns. Based on prior analysis, she could have a pretty good idea of which variables would provide the best results for a new gardening book she might want to promote. And she could specify the terms.

Knowing which terms really worked was not of idle importance. After 35.5 years in operation, the results of each model, each set of variables, had been fed back into the system. Haggerty could pretty much predict what the profitability would be of any given campaign, based on the choices she made about who should receive which offer.

That's critical, because Haggerty is now responsible for the results of condensed book and one-shot sales efforts. Even at the height of the company's mail marketing prowess, "On a good day, 95% of your customers don't respond,'' she says.


A decade after the arrival of the Unified File System, the Digest reached 100 million readers around the world. Founder DeWitt Wallace and wife, Lila, ran the business almost as if defying profit to show up, which it did anyway.

Besides the housewives who were brought in by bus from Pleasantville to work on the database, commuters from as far away as Connecticut and New Jersey could ride to work on company buses. The workday ended at 4 p.m. Each May, Fridays became holidays, so workers could tend to their gardens. Each worker famously got a 20-pound turkey at Thanksgiving, as a token of appreciation.

Monet, Renoir and Matisse paintings hung on office and hallway walls. The quarter-mile headquarters building, built in Georgian Renaissance style, overlooked its own exit off the Saw Mill River Parkway in Westchester County, N.Y. Even though it actually sat in Chappaqua, the founders got the post office to agree that its headquarters would be—for mailing and address purposes—in the bucolic-sounding Pleasantville, one town over.

The company's postal bill in the late '70s probably exceeded $100 million, since mail was its only method of promotion. Because of the vast nature of its filing system, Reader's Digest was called on twice to help improve the nation's postal system, in the establishment of ZIP codes in 1963 and the national change-of-address system in 1986. After all, if you wanted to get a sense of where people lived, who better to check with than Reader's Digest?

At its peak, Reader's Digest was sent to 18 million subscribers in the U.S. each month. Twenty million volumes of condensed books, now known as Select Editions, were sold, along with 12 million other books. It became one of the world's largest marketers of music, moving an estimated $150 million worth of records and cassettes in the mid-'80s. Once, it sold 8-track tapes.

By September 1986, the marketers had used a singular view of their customers and resolute devotion to statistical analysis to create a $1.5 billion-a-year business, worldwide, off Wallace's notion that the best stories in the best magazines could be converted into a print portal of their own.

Every day, the company's 1,228 data processors in Chappaqua, a.k.a. Pleasantville, handled 116,000 orders, deposited $2 million of customer payments, processed 490,000 responses, and sent out 2.32 million letters, postcards and labels. Three quarters of a million magnetic tapes stored information on customers.

But even then, a team of systems architects and marketers had decided to explore the creation of a new Customer Information Management System. The new system would "handle all marketing, fulfillment and customer service data processing needs well into the 21st Century," the company's in-house newsletter, Pegasus, reported. The new system would be easier and cheaper to maintain, make it easier to read customer records, and perform at least as well in selecting households that would respond to mailers.

The time had come to pull the plug on hexadecimal code, the long customer records and the expensive, brittle Unified Filing System.

By 1988, Pegasus was ready to put an exact date on when CIMS would replace the increasingly cranky old lady, UFS: December 1990.


In 1990, there was no vendor of customer relationship management software like Siebel Systems. The company would have to write the replacement for the Unified File System from scratch.

It would have to serve the operations of both marketing, which tried to create orders, and fulfillment, which had to satisfy them. But each was a separate, complex business of its own.

The original aim of CIMS was to replace UFS in one fell swoop. Turn off the old, turn on the new.

"That's not feasible. You couldn't do it all at once,'' says Rob Hilliard, the current vice president of customer technologies. "It's like building a house in a day. A carpenter has to lay the foundation before the electrician comes in."

Nineteen-ninety came and went. By 1992, the company was re-examining the entire idea of a "big bang" conversion. By 1994, Andersen Consulting had come in, to try and assess what could be done to save the project.

The effort was redubbed "Chrysalis,'' as if the initiative had been a cocoon for giving birth to moths. Instead of a complete cutover to a new system, the CIMS project would turn into a series of individual tools, to run on top of the Unified File System. The old lady would keep singing.

One of the new tools would be a Selection Tree, introduced by Kari Regan, at the time the director of quantitative analysis. The tree was designed to make it easy to choose terms for mail campaigns, and then run all customers' profiles through it.

The tree was set up as a series of questions— from whether the customer is currently paying for a product, to whether the customer has received a subscription as a gift in the past, to whether he or she should be excluded for bouncing checks or having filed for bankruptcy.

The tree would act as the coding equivalent of a pinball machine. Any customer profile run through the tree would wind up in a final bucket on one of its branches—even if it took 180 or 200 questions to get there.

That took brute-force computing. Which is one of the reasons, according to Otten, that the elegant lady got as old as she did. Her files on customers were complete. The programs that would analyze their behavior worked effectively. And computers kept getting cheaper. It was always more economical to buy more horsepower for crunching code than to rewrite the code.

That would hold true even if every one of the company's 56 million records on customers would have to be run through the entire tree, every time, to get a final selection of names for every single marketing campaign. Even if you were doing just a 1-million-name campaign.

Several selections is exactly what Devanny was about to launch on the "last big gang run" on May 12.

The "XNB scoring tree,'' for instance, would determine which names were most likely to generate new business at Christmastime this year. Different materials for different types of offers would be sent to a letter shop to produce and mail out in July. The differences between materials might be as simple as coloring; or as involved as what kind of free gifts or incentives might make a difference.

But first, the file system would be asked which customers are most likely to respond well to 2-for-1 offers and settle those people into different buckets.

Some might get to renew their subscriptions at $19.95; others, at $21.95. Counterintuitively, the longer you've been a subscriber, the higher the price you will be offered. The more years you've been around, the more you've shown you value the product. So why cut the price down?

But after this last pass at the Unified File System, all future selections would be run through a new combination of software from Oracle (database), Unica (customer selection) and SAS (quantitative analysis), running on faster servers (Sun) using a low-cost operating system (Unix).

After 17 years of trying in earnest to ease out the elegant lady, a new system for managing tens of millions of customers was finally moving in. What had been a 33,000-square-foot data center in its heyday, would be 1,500 feet square in its new incarnation. And most of the computing wouldn't take place under the long greensward of the Digest's headquarters campus.

Confusion Matrices'">

'Confusion Matrices'

The new clearinghouse of information held on Reader's Digest's computers—or those of its outsourcer—would have an equally straightforward and forgettable name: the Marketing Systems Replacement, or MSR.

This attempt to move off the Unified File System was pitched in late 2002 by Regan, now vice president of database marketing services, and Hilliard, who came from Andersen Consulting during the push to introduce pieces of the previous big effort, CIMS.

In that campaign, the first idea was to convert to a wholly new system in a single "big bang." Of course, any conversion, then or now, would be like "flying a 747 across the ocean and changing the engines in mid-air,'' says the company's chief information officer, Jeffrey S. Spar.

Most critically, in the '90s, the UFS handled both marketing, to generate new orders; and fulfillment, where books, magazines, tapes and the like were sent out to meet those orders.

But after Reader's Digest went public in 1990, its conservative readership began to literally die off. The company compounded the problem by flooding its market with direct mail. A new CEO lasted only three years. Thousands of workers were cut loose, including hundreds in Pleasantville. Net income of $264 million in 1995 dropped to $49.5 million last year. Revenue, topping $3 billion in 1996, dropped below $2.4 billion last year.

Grappling with the slide, the company decided its core competitive advantage was how it used information to market to customers. The fulfillment of orders would be contracted out, to Communications Data Services in Des Moines, Iowa, starting in May 2000.

Information on those orders would come back from the fulfillment house and be kept in the Unified File System, ensuring that the view of customer activity remained complete. The ability to send out magazines, books and other products would not be affected by any system downtime—or changeover.

If, during the changeover, selections of new marketing targets didn't work properly, that was not a critical problem, as long as the disruptions didn't get out of hand.

Through 2004, Hilliard—who had become the leader of the replacement project—and Regan were trying to transfer UFS files into tables in the new system. They figured they could transfer data a bit at a time, then update those records directly in MSR.

But the systems would not sync easily. The old system was accurate only to the month; the new one, to the day, Regan says. If a conversion was off by only one-thirtieth of an answer, it was the wrong answer. There were other miscalculations, such as double-counting the write-offs attributed to some customers who never paid their bills.

The biggest bugaboo? That most reliable predictor of customer readiness to respond, Promotions Since Last Order, was not working.

For inexplicable reasons, the PSLO, which should have a value between 2 and 4, kept "zeroing out,'' according to Regan.

"Every day, it was, 'Can't we get a break?'" she says.

Hilliard would stay on e-mail and instant messaging until 2 a.m. on Saturday nights, trying to fix the problems.

They had to redo the PSLO calculation several times. But that was just one term that had to be right. There were 1,400 terms, plus 300 equations that used the terms. Each had to be translated to run properly in the MSR.

Regan and Hilliard started plotting results of the new system against those of the old system. Results of the old system would be depicted on a straight line from the bottom right to the upper left of a graph. Results of the new system would be plotted wherever they fell. If they all fell on the line, fine. If not, check it out.

Take the Individual Age matrix. Reader's Digest reset any birth that was before 1900—at 1900. But the logic of the replacement system did not carry out the 1900 rule accurately. Too many subscribers and customers appeared to be more than a century old. Only by repeatedly reviewing the raw data in both systems could the confusion be eliminated.

Regan and Hilliard would run these "confusion matrices" thousands of times, before being satisfied the new system would work as well as the old lady.

Her time was almost up. The cost: By one estimate, nearly $1 billion had been spent on hardware, software, consulting services and staff time since the advent of CIMS in 1988. The reality: No one knows. "Those unkind have said 'Countless Innumerable Millions Spent,'" Burns notes.

But the decision on whether the new system was ready to support a business did not fall to Regan or Hilliard. That was the statisticians' call: Nester; Al Virgo, the database marketing director for books and music; and Aiyi Wang, associate director of database marketing. After all, finding the right names to send marketing appeals that produce orders is what matters.

Ultimately, it was Nester who made the call. On April 27 she gave the high sign that the new terms were working and the replacement system ready to perform as the database of record.

The first bulk mailing, a music campaign, was hard. "Really hard," Regan says. There was no template to follow or copy over from a prior campaign. Every step had to be checked thoroughly, from the elimination of names to the equations used. The elegant old lady had made sure that every name ended up in only one box, the target of only one mailing in a campaign. In the new system, a name could end up in more than one box. Selections from box A had to be run against box B to make sure there were no duplications—a time-consuming task. Where database marketers like Virgo had gotten used to having selections made two weeks in advance of when they had to be delivered to a letter shop, this mailing went down to the wire.

In fact, the first bulk mailing in MSR was completed in the second week of May—just as Devanny was enunciating his eulogy for the original one, the Unified File System.

"The fact that a 35-year-old system could support a business that well ... I'm amazed at the architecture. That's kind of magical,'' says Regan, who retired on May 20. "Hopefully, [the new system] will be as robust and successful.''

Or, at least, live as long. Whether MSR will be as effective remains to be seen. Also in doubt is how much it will change. No system runs forever—or untouched.

"It's a constant challenge to keep them up-to-date; it's a challenge to make sure that they're accurate, make sure that they're doing what you think they're supposed to be doing,'' Regan points out. "These systems are never done," she says.

Except for the elegant lady. Now retired, too, having taken no time off since Nov. 16, 1969.

Reader's Digest Base Case">

Reader's Digest Base Case

Headquarters: Reader's Digest Rd., Pleasantville, NY 10570

Phone: (914) 238-1000

Business: Marketer of magazines, books, music

Chief Information Officer: Jeffrey S. Spar

Financials In 2004: Net income of $49.5 million, or 2.1% of its $2.4 billion in revenue. Profit down 62.5% since 2001.

Challenge: Update the heart of its business, its numerically based methods for marketing its products to customers, from a system put in place in November 1969.

Baseline Goals:

  • Increase cash generated by business to more than $175 million a year, from $162.1 million in 2003.

  • Realize "low double-digit profit growth" in each of three operating segments. In 2004:

  • Reader's Digest International operating profit gained 16.3% to $57 million, from $49 million.

  • Reader's Digest North America gained 16.3% to $71 million, from $61 million.

  • Consumer Business Services, which include books and home entertainment services, fell 29.7%, to $64 million.

  • Keep monthly circulation of its Digest, at 10 million or more.

  • Grow annual revenue, past $2.5 billion.