Skip to Content

Research in Development

IBM builds services-based R&D.

Visitors entering the T. J. Watson Research Center in Yorktown Heights, NY, are greeted by glass cases displaying historical calculating machines, from the abacus to Leon Bolle’s multiplier. The objects do honor to IBM’s history. They also serve as reminders to the people who work here that IBM’s products today are not at all what they used to be.

IBM’s track record in corporate research is almost unparalleled. The company routinely nabs the greatest number of U.S. patents in a given year. Its researchers have won two Nobel Prizes in physics. Its laboratories invented magnetic storage, the first formalized computer language (Fortran), fractals, the relational database, and the scanning tunneling microscope. If quantum computers ever do arrive, it will be in large part because of developments at IBM Research.

“There is no product in the IBM company that does not start in research, with minor exceptions,” says Paul Horn, a solid-state physicist who has run IBM ­Research since 1996.

But while Horn is proud of IBM’s achievements, he believes that the company needs to change the way it thinks about research, for the simple reason that its product mix keeps shifting toward the ethereal. That started with software. Once seen as a mere hardware accessory, software grew in importance after the 1956 consent decree between IBM and the U.S. government, which created the market for packaged software. For decades thereafter, IBM sold more software than any other company (in 2004, its software sales totaled $15 billion, about 16 percent of its revenues). But although software can’t be taken apart on a lab bench, the role that R&D can play in its development has always been clear to researchers at IBM. Computer languages, the relational database, middleware, security software – they all met obvious operational needs. But IBM’s most recent category of product – services – has more than a few of its engi­neers scratching their heads.

Speaking about IBM Research, Horn says, “If we were to disappear, there’d be a sudden stop in our products side. I can’t make that statement about services.”

But what, exactly, are services? Within IBM, the word is used in two ways. First, “services” is one of the company’s three broad product categories (the other two being physical products and software). Most pure services are sold by IBM’s 180,000 consultants and range from wholesale IT outsourcing to training, ­human-capital management, and the On Demand Innovation Services effort, a broad (that is, vague) effort to make widely disparate systems communicate more effectively, and in real time. These services don’t have profit margins as high as those of IBM’s proprietary hardware and software, but service sales often follow product sales (and sometimes drive more product sales).

The term “services” is also used by people at IBM to mean any work that helps improve a product or a process. The product could be a piece of hardware or software; the process could be the way consultants present data to clients. But whether services are thought of as discrete products or product enhancers, IBM sees them as critical to its future.

Research at a Crossroads
Horn and the rest of IBM Research find themselves in the midst of a somewhat awkward transition. While products that the company makes have, as a whole, moved across the hardware-software-­services continuum, the research that underlies them hasn’t quite kept pace. Thus IBM Research appears, in certain instances, as if it’s attempting to answer the wrong questions.

Its Zürich laboratory, for instance, is close to completing work on Millipede, a nanomechanical device that can store data at a density 20 times higher than that of magnetic storage. But if Millipede proves commercially viable, it probably won’t be built by IBM, because IBM sold its storage division in 2002 as part of an effort to shed low-margin businesses. This February, it sold its once vaunted personal-computer business for the same reason. At the time of the sale, services already constituted slightly less than half of IBM’s $96 billion in annual revenues. Now that the hefty – $12 billion – but often unprofitable PC business is gone, services will likely check in at closer to two-thirds of revenues.

Horn’s challenge, then, has been to take a $6 billion research organization dedicated to work that advances tech­nology products and get it to do work that benefits service businesses. IBM is thus in the process of answering an important question for all technology companies: can corporations perform useful research in the services arena?

At the very least, say many observers, they’ll have to try. Roland Rust is a sometime consultant to IBM and director of the Center for Excellence in Service at the University of Maryland’s Robert H. Smith School of Business, which is partnered with the company. “It’s a no-brainer,” Rust says, that as the economy in general shifts away from goods, companies will need to pursue services research. He notes that even General Electric, an industrial giant, now thinks of itself as a services company. And he believes that IBM’s approach to services research will ripple through the rest of the industrial world, both because it has a highly regarded research lab and because it has made the hard transition from being primarily a goods company to being pri­marily a services company.

Henry Chesbrough, executive director of the Center for Open Innovation at the University of California, Berkeley’s Haas School of Business, is also unsurprised that IBM is shaking things up. He contends that the academic field of computer science exists in part thanks to IBM’s ­insistence that it wasn’t something that belonged to the physics, engineering, or mathematics departments. Chesbrough thinks IBM may help services in the same way, by putting its corporate stamp on the idea that services research should involve operations, marketing, and supply chain management. “Left to our own devices, we’ll remain in our silos,” Chesbrough says. “I’m excited about it because they’re taking a fresh look at a very old problem, and they’re among the very first to do something about it.”

But being among the first to do something is not necessarily the same as doing it quickly. “There was always a feeling we had to do something in services, as soon as we’d had a services business,” Horn explains. But the company had problems determining just what it should do. IBM ­Research had pursued formal studies of services projects, and it had even developed corporate strategies, such as the ­autonomic-computing initiative Horn unveiled in 2001, that involved elements of the service business. But these projects were geared toward physical products, not service products.

Then everything changed. IBM’s tentative approach to supporting services ended in July 2002, when it announced that it was planning to buy PricewaterhouseCoopers Consulting. By the end of 2003, the two companies’ first full year of merged operation, close to half of IBM’s revenues were coming from services. In contrast, services accounted for less than 15 percent of R&D spending. “So did that mean Lou [Gerstner, then the CEO] could say, ‘Do I need only half the R&D spending?’” Horn recalls wondering. “Those things get you thinking.”

“Services Is a People Business”
Horn wasn’t the only one thinking. In May 2002, Paul Maglio, a cognitive scientist in IBM’s Almaden research lab in San Jose, CA, also had change on his mind. Maglio was working on user-interface research, and over Mai Tais with friends at a Web conference in Hawaii, he spun out a scenario he’d been working on with IBM colleague Rob Barrett, in which the company looked at technology research differently. “Services is a people business,” Maglio says now. “I’m a cognitive scientist; I’m interested in people generally. What we were doing [in human-computer interfaces] was fine, but it didn’t get the whole picture. I thought we could create a new breed of research.”

In particular, Maglio thought that IBM simply didn’t know what its customers actually did with technology. He also didn’t think anyone knew whether IBM’s consultants understood what its customers wanted. He thought there was a need for a broad study that drew on what he called the “human sciences.”

Maglio and Barrett were proposing work that would be much different from what IBM, with its roots in hard science, was used to doing. But they found a willing ear, and a champion, in Jim Spohrer, then the chief technology officer of IBM’s venture capital unit. After a series of discussions starting in July 2002, around the time IBM announced the PricewaterhouseCoopers deal, Spohrer agreed to take the idea first to Robert Morris, head of the Almaden research lab, and ultimately to Horn. He nixed the human-­sciences part of the pitch; instead, he proposed using the new approach to help Horn solve his biggest problem: determining how to help the services business. When Spohrer brought the idea to Horn, he described services as a “human business that needed human research.”

The timing was excellent. Horn was already interested in seeing the company develop the ability to do softer research. “What we were doing in services was very much on the quantitative side, and I thought we needed a break from it. We hadn’t done anything at all on how technology affects people,” Horn says. His push to support services thus converged with Maglio’s push to make research more human. That convergence meant not only that services would become an important subject of research, but also that hardware and software research would begin to include some “soft” work. This marked a turning point for IBM – and a good one, Spohrer contends: “Research reinvents itself every 10 years. So it was a good thing to do.”

That December, Horn signed off on the idea. Spohrer was tapped to direct services research at Almaden. His first outside hire was a business anthropologist, Jeanette Blomberg. He wanted her to study work practices, including how technology users collaborate. Meanwhile, Maglio began to investigate what systems administrators actually do. He found that they spent between 60 and 90 percent of their time communicating with other systems administrators about systems issues. Armed with that knowledge, IBM began developing tools to help systems admin­istrators write and share “scripts” – the short, simple programs they use to coördinate the work of other programs.

So far, it may be that the mathematics arm of IBM Research has done the most work on services. Part of the group’s charge is to devise algorithms to improve how businesses use information tech­nology. Brenda Dietrich, manager of the mathematical-sciences research department at the Watson Research Center, says her 90-person unit sees great potential in even the softest research being done on services. “What Jim’s group is doing is a good source of data for my models,” which are currently too simplistic to reflect the real world, she says. Dietrich believes that if IBM can get better data on what people actually do with technology, her group can produce more-useful algorithms. “If we can really get the data [Spohrer is] talking about, that’s powerful.”

About half of those 90 researchers are now engaged in services work at any one time, either traveling with consultants or working on projects that will help consultants. Dietrich says that IBM’s expectations for its research arm have changed gradually but dramatically in the 20 years she has worked there. “It used to be that you got asked, ‘Have you got anything into products?’” she says. But an analogous question about services is harder to answer. Development work on services tends to wind up as part of a process, not a product. Discrete product features are easy to point to; the parts of a process are broader, and mushier.

Does All This Work Increase Profitability?
Two and a half years after starting to pursue services science, IBM can point to some successes. The company has a fixed procedure for measuring what it calls “accomplishments,” and Horn estimates that about 15 percent of the accomplishments by research now involve services. Research has had 250 direct consulting engagements since 2002. And in the fourth quarter of 2004, it spawned two new practice areas, WebFountain and the Center for Business Optimization.

Both are fledgling but promising. WebFountain is a set of processes for organizing and analyzing huge sets of disparate data. A WebFountain application might involve surveying text on the Web and deducing what people are actually saying, rather than just returning instances of keywords. That work is of obvious interest to IBM customers that have, say, customer service departments that need to gauge complaints or answer questions.

The Center for Business Optimization helps clients tighten up operations. It recently created what it calls a Pharmaceutical Production Refactoring Tool, which helps large drugmakers reduce the risk of having to shut down manufacturing facilities because of failed health inspections. This tool could prove especially useful now, in light of recent changes in drug-manufacturing regulations, which mean that if the U.S. Food and Drug Administration finds a problem with a single production line at a manufacturing facility, it will shut down the entire site until the problem is resolved. Thus, “something low-margin can wipe out your entire facility revenue,” notes Krishna Nathan, head of IBM’s Zürich Research Lab and vice president of services research. “It’s a huge problem for the industry. [Drug manufacturers] need to refactor the risk and restructure it across all their sites.”

To create a product that would help meet that need, IBM researchers in November 2004 worked with IBM’s Business Consulting Services unit to devise algorithms that would quantify the risk of failure presented at each site and suggest changes, based on historical data from drugmakers. According to Nathan, in a pilot case involving one drugmaker, initial tests showed that the IBM tool had the potential to reduce risk by 30 percent.

The push now is to do more, faster. No one at IBM Research thinks the division is getting enough done on the services side. Over the past year, IBM has started referring to its research work as “services science,” but there are people even in the services group who can’t say those words with a straight face. Nathan added “services” to his title only in September, when Horn decided he wanted the research department to accelerate its services efforts.

“We’re making a lot of progress, but would I like it to be faster? Yeah, I would,” says Horn. He hopes that in five years, half of IBM Research’s “accomplishments” will come from its work on services. But he thinks it may take years for the department to become as services oriented as he thinks it must be to best serve IBM’s needs.

Services have clearly been good business for IBM. Less clear is whether services research is good business; it’s too new an area for us to know for sure. But there are encouraging signs: IBM’s On Demand Innovation Services program, which basically farms out the talents of the research staff, last year generated more than $300 million in revenues. That might seem an insignificant portion of a $46 billion service business. But it more than ­triples the revenues from the year before. Growth like that is hard to ignore – and could be a sign of more to come.

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

The problem with plug-in hybrids? Their drivers.

Plug-in hybrids are often sold as a transition to EVs, but new data from Europe shows we’re still underestimating the emissions they produce.

Google DeepMind’s new generative model makes Super Mario–like games from scratch

Genie learns how to control games by watching hours and hours of video. It could help train next-gen robots too.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.