ACM Interactions

Researchers are rewarded for publishing, but this time, my heart wasn’t in it.

It was 2006. IBM software let an employer specify an interval—two months, six months, a year—after which an email message would disappear. This was a relatively new concept. Digital storage had been too expensive to hang onto much, but prices had dropped and capacity increased. People no longer filled their hard drives. Many saved email.

When IBM put automatic email deletion into practice, a research manager asked her IT guy to disable it. “That would be against policy,” he pointed out. She replied, “Disable it.” Another IBM acquaintance avoided upgrading to the version of the email system that included the feature. When she returned from a sick leave, she found that a helpful colleague had updated her system. Her entire email archive was irretrievably gone.

“We call it email retention, but it’s really email deletion.”

Word got around that Microsoft would deploy a new “managed email” tool to all North American employees, deleting most messages after six months (extended grudgingly to 12 when some argued email was needed in preparing for annual reviews). Because of exceptions—for example, patent-related documents must be preserved for 10 years—employees would have to file email appropriately.

Many researchers, myself included, prefer to hang onto stuff indefinitely. I paused another project to inquire and learned that a former student of mine was working on it. A pilot test with 1,000 employees was underway, he said. In a company of 100,000, it is easy not to hear about such things. He added that it was not his favorite project, and soon left the team.

Our legal division had assembled a team of about 10 to oversee the deployment. Two-thirds were women, including the group manager and her manager. They were enthusiastic. Many had voluntarily transferred from positions in records management or IT to work on it. My assumption that people embracing email annihilation were authoritarian types quickly proved wrong, it was a friendly group with bohemian streaks. They just didn’t like large piles of email.

I had assumed that the goal of deleting messages was to avoid embarrassing revelations, such as an admission that smoking is unhealthy or a threat to cut off a competitor’s air supply. Wrong again. True, some customers clamoring for this capability had figured prominently in questionable government contracting and environmental abuse. But it is a crime to intentionally delete inculpatory evidence and, I was told, litigation outcomes are based on patterns of behavior, not the odd colorful remark that draws press notice.

Why then delete email? Not everyone realized that storage costs had plummeted, but for large organizations, the primary motive was to reduce the cost of “ediscovery,” not hardware expenditures.

Major companies are involved in more litigation than you might think. Each party subpoenas the other’s correspondence, which is read by high-priced attorneys. They read their side’s documents to avoid being surprised and to identify any that need not be turned over, such as personal email, clearly irrelevant email, and any correspondence with an attorney, which as we know from film and television falls under attorney-client privilege. A large company can spend tens of millions of dollars a year reading its employees’ email. Reduce the email lying around to be discovered, the thinking went, and you reduce ediscovery expenses.

Word of researcher unhappiness over the approaching email massacre reached the ears of the company’s Chief Software Architect, Bill Gates. We were granted an exemption: A “research” category was created, similar to that for patent-related communication.

Nevertheless, I pursued the matter. I asked the team about the 1000-employee pilot deployment. The response was, “The software works.” Great, but what was the user experience? They had no idea. The purpose of the pilot was to see that the software deleted what it should—and only what it should. The most important exception to automatic deletion is “litigation hold”: Documents of an employee involved in litigation must be preserved. Accidental deletion of email sent or received by someone on litigation hold could be catastrophic.

The deployment team was intrigued by the idea of asking the early participants about their experiences. Maybe we would find and fix problems, and identify best practices to promote. This willingness to seek out complaints was to the team’s credit, although I was realizing that they and I had very different views of the probable outcome. They believed that most employees would want to reduce ediscovery costs and storage space requirements, and about that they were right. But they also believed that saving less email would increase day to day operational efficiency, whereas my intuition was that it would reduce efficiency, and not by a small amount. But I had been wrong about a lot so far, a not uncommon result of venturing beyond the ivory tower walls of a research laboratory, so I was open-minded.

“Doesn’t all that email make you feel grubby?”

My new collaborators often invoked the term “business value.” The discipline of records management matured at a time when secretaries maintained rolodexes and filing cabinets organized to facilitate information retrieval. Despite such efforts, records often proved difficult to locate. A large chemical company manager told me that it was less expensive to run new tests of the properties of a chemical compound than to find the results of identical tests carried out years earlier.

To keep things manageable back then, only records that had business value were retained. To save everything would be painful and make retrieval a nightmare. Raised in this tradition, my easygoing colleagues were uncompromising in their determination to expunge my treasured email. They equated sparsity with healthy efficiency. When I revealed that I saved everything, they regarded me sadly, as though I had a disease.

I have no assistant to file documents and maintain rolodexes. I may never again wish to contact this participant in a brief email exchange—but what if five years from now I do? Adding everyone to my contact list is too much trouble, so I keep the email, and a quick search based on the topic or approximate date can retrieve her in seconds. It happens often enough.

I distributed to the pilot participants an email survey comprising multiple choice and open-ended response questions. The next step was to dig deeper via interviews. Fascinated, the deployment team asked to help. Working with inexperienced interviewers does not reduce the load, but the benefits of having the team engage with their users outweighed that consideration. I put together a short course on interview methods.

“Each informant represents a thousand other employees, a million potential customers—we want to understand the informant, not convert them to our way of thinking. For example, someone conducting a survey of voter preferences has opinions, but doesn’t argue with a voter who differs. If someone reports an intention to vote for Ralph Nader, the interviewer doesn’t shout, ‘What? Throw away your vote?’”

Everyone nodded.

“In exchange for the informant trusting us with their information, our duty is to protect them.” After they nodded again, I continued with a challenging example drawn from the email survey: “For example, if an employee says that he or she gets around the system by using gmail for work-related communication—”

White-faced, a team member interrupted me through clenched teeth, “That would be a firing offense!”

At the end of the training session, the team manager said, “I don’t think I’ll be able to keep myself from arguing with people.” Everyone laughed.

The white-faced team member dropped out. Each interview save one was conducted by one team member and myself, so I could keep it on track. One interview I couldn’t attend. The team manager and another went. When I later asked where the data were, they looked embarrassed. “We argued with him,” the team manager reported. “We converted him.”

My intuition batting average jumps to one for three

The survey and interviews established that auto-deletion of email was disastrously inefficient. The cost of the time that employees spent categorizing email as required by the system outweighed ediscovery costs. Time was also lost reconstructing information that had only been retained in email. “I spent four hours rebuilding a spreadsheet that was deleted.”

Workarounds contrived to hide email in other places took time and made reviewing messages more difficult. Such workarounds would also create huge problems if litigants’ attorneys became aware they existed, as the company would be responsible for ferreting them out and turning everything over.

Most damning of all, I discovered that managed email would not reduce ediscovery costs much. The executives and senior managers whose email was most often subpoenaed were always on litigation hold for one case or another, so their email was never deleted and would have to be read. The 90% of employees who were never subpoenaed would bear virtually all of the inconvenience.

Finally, ediscovery costs were declining. Software firms were developing tools to pre-process and categorize documents, enabling attorneys to review them an order of magnitude more efficiently. At one such firm I saw attorneys in front of large displays, viewing clusters of documents that had been automatically categorized on several dimensions and arranged so that an attorney could dismiss a batch with one click—all email about planning lunch or discussing performance reviews—or drill down and redirect items. That firm had experimented with attorneys using an Xbox handset rather than a keyboard and mouse to manipulate clusters of documents. They obtained an additional 10% increase in efficiency. However, they feared that customers who saw attorneys using Xbox handsets would conclude that these were not the professionals they wanted to hire for a couple hundred dollars an hour, so the idea was dropped. Nevertheless, ediscovery costs were dropping fast.

Positive outcomes

At the ediscovery firm, I asked, “Are there changes in Exchange that would help improve the efficiency of your software?” Yes. A manager in Exchange told me that ediscovery firms were a significant market segment, so I connected them and a successful collaboration resulted.

We found ways to improve the interface and flow of our “email retention” software, reducing the inconvenience for anyone who would end up using it.

I learned about different facets of organizations, technology use, and people. I loved working with the records management team and the attorneys. Attorneys in tech companies are relaxed, funny, and have endless supplies of stories. They never let you record an interview, but they are invariably good company.

As the date for the deployment to 50,000 Microsoft North America employees approached, word of our study circulated. The executive vice president overseeing our legal division convened a meeting that was run with breathtaking efficiency, like a scene in The West Wing. He turned to me. “We created a ‘research’ exemption for Microsoft Research. Why are you here?” I said, “I’m doing this for Microsoft, not MSR.”

The deployment was cancelled.

The product was released. Customers wanted it. A partner at a major Bay Area law firm heard of the study and phoned me. He was interested in our analysis of efficiency, but noted that for some firms, profitability was the only issue. “Consider Philip Morris,” he said. “One of their businesses is addicting people to something that will kill them. As long as that business is profitable, they will stay in it. If it ceases being profitable, they will get out. Efficiency isn’t a concern.”

Collateral damage

I saw a cloud on the horizon. The raison d’être of the team that had welcomed me was to oversee a deployment that would not happen. What would they do? I formed a plan. When a subpoena arrives, all affected employees are put on litigation hold. Their email and documents are collected and read by attorneys to identify relevant material. Determining relevance is not easy. It could be signaled by the presence or absence of project code names that evolved over time. People involved in discussions may have left the company. It is often difficult to determine which project a short email message refers to. Some employees file information under project names, but others rely on message topic, sender, recipients, date, urgency, or a combination of features. Some don’t file much at all, relying on Inboxes or other files holding thousands of uncategorized messages. Attorneys sit at computers trying to reconstruct a history that often spans several years and scores or hundreds of people.

I thought, “Here is the opportunity for email management.” Armed with tools and procedures, the team could help attorneys sort this out by working with the employees on litigation hold: identifying attorneys with whom privileged email was exchanged, listing relevant project code names, indicating colleagues always or usually engaged in communication relevant to the subpoenaed project and those wholly unrelated, and so on. This could greatly reduce the time that expensive attorneys spent piecing this together.

I worked on a proposal, but I was not fast enough. The legal division makes an effort to reassign attorneys whose roles are no longer needed, but it does not generate jobs for surplus records managers. The team was laid off, including the manager and her manager.

They had heeded a call to take on important work for the company. The positions they had left had been filled. They had welcomed me and worked with me, and it cost them their jobs. “Our duty is to protect our informants,” I had taught them, and then I failed to do it.

A few found other positions in the company for a time. None remain today. Before leaving, they held a party. It could have been a wake, but it was labeled a project completion celebration. Lacking a “morale budget,” it was potluck. An artist in the group handed out awards. Mine has rested on my office window sill for almost a decade. At the party, someone told me quietly, “We had a discussion about whether or not to invite you. We decided that you were one of the team.” I was the one not laid off, for whom the study was a success. But not a success I felt like writing up for publication.

Jonathan Grudin