(Re) Writing Code

Senior executives who want to transform their organization can't focus only on strategy or structure. Ultimately, they need to rescue their company's data from digital oblivion.

In the summer of 1991, the Russian mathematician Andrey Terekhov stood in the middle of a potato field outside St. Petersburg. As he scanned the dry, nearly infertile land, he felt a sense of humiliation. "I, a world-famous professor, was preparing potatoes," he recalls. Terekhov, the creator of key telecommunications systems for the KGB, had been a mentor to hundreds of Russia's best mathematical minds. But in post-Soviet Russia, the government had stopped paying him the small salary that he had earned as a professor at St. Petersburg State University. So he had forfeited a labor of love — the study of algorithms, geometry, the theory of graphs — for hard labor.

Then something occurred that changed Terekhov's life. Two American IT executives traveled to St. Petersburg to find the mathematician. They wanted to hire Terekhov to tackle a problem that was confounding U.S. programmers — and frustrating many large U.S. corporations. The problem was as simple as it was dire: Computers and software languages were evolving so fast that companies were accumulating vital programs, and critical business data, in languages that were going out of date. Well before the Internet frenzy — which only made the problem worse — companies were struggling with how to translate their financial records, their inventory systems, and the programs that ran their factories into new, more flexible languages.

The standard solution was clumsy and expensive: Fill a room with engineers, have them manually dissect an existing program — and then have them manually rewrite that program in a new language. It took too much time, it cost too much money, it resulted in buggy code, and it was unrewarding for the people who did it. (It was the programming equivalent of potato farming.) The American executives had heard that Terekhov might be able to automate the entire conversion process. The Russian jumped at the chance: "What else could I say but 'Yes,' " he recalls.

Ten years, three companies, and several versions of software later, Terekhov and the team that he assembled have created a product, RescueWare 6.0, that does for software languages what science-fiction writers have imagined machines doing for human languages: translate one into another quickly and seamlessly. Thanks to RescueWare, information that was previously locked in mainframes can now be displayed on Web pages, and programs that used to run only on big boxes in COBOL can now run on servers in Java.

RescueWare is a huge innovation — one that has already had a tremendous impact on Terekhov. The 51-year-old mathematician, who resembles a smiling Fidel Castro, now earns 10 times as much as the going wage for a Russian mathematics professor. He runs Lanit-Tercom, a St. Petersburg-based telecommunications-software company. He owns a four-story house with a swimming pool, a Jacuzzi, a basketball court — and no potatoes in the yard. Meanwhile, the two men who sought him out, Vivek Wadhwa and Len Erlikh, run a company called Relativity Technologies, which produces and sells RescueWare. Based in Cary, North Carolina, Relativity has a client list that includes Charles Schwab, Ernst & Young, and the U.S. Air Force. Wadhwa and Erlikh compare Terekhov's achievement to cracking the atom. "This guy is a Russian Einstein," says Wadhwa, 43, CEO.

Hyperbole aside, what Terekhov and his team have built could change the game of change for big, information-driven companies. If RescueWare works as promised, it will allow companies to become as digital as they want to be. With a tool that makes change easier, faster, and less risky, companies may start changing their systems more often, reaping benefits both for their business and for their customers. And automating change means increasing the pace of change. Not only does RescueWare have the potential to liberate companies from outdated programs; it also has the potential to liberate programmers from outdated careers as software translators. That would leave them free to pursue more useful, and more creative, work — free, in other words, to unleash even more change.

Past, Present, Future

At Charles Schwab, huge supercomputers power the operations of a financial-services giant: Mainframes process stock trades, calculate returns on mutual funds, and pull up and display customer accounts for service reps. As recently as a year ago, many of those functions were run in Seer*HPS, a proprietary computer language that is rapidly becoming obsolete. Schwab's IT directors wanted to transfer those tasks to a version of COBOL, in part because the company's army of programmers knew that language fluently.

This isn't glamorous work, but it's exactly the kind of work that companies must undertake if they are to improve their operations dramatically. Not only does Schwab have all kinds of data locked up in old formats, but, equally important, it relies on all kinds of formulas for using that data. One key function, for example, calculates the average cost bases for Schwab's 2,000-plus mutual funds. Conversions that handle such "deep" code are what Relativity aims to specialize in.

"In that kind of translation, it's easy to generate the worst kind of spaghetti code," explains John Bilotta, 53, VP of operational-support services at Schwab, who served as program manager for the conversion. "We risked spending a lot of money and time, and ending up with applications that didn't work as well as the old ones." But Schwab's only alternative to using an automation tool like RescueWare was to throw out the legacy program and rewrite it from scratch. "We really didn't want to do that," Bilotta says. Given how high the stakes were, Schwab decided to send Seer*HPS to Relativity. After three months of analysis, automated code writing, and product testing, Relativity returned the mutual-fund functions to Schwab in COBOL. Hardly any new computer program operates perfectly during its first run. Glitches are a fact of life even in word-processing software, let alone applications that handle complex financial calculations. But Relativity's conversion ran perfectly the first time out. And it performed the mutual-fund calculations 50% faster than the old version had done. Today, all of Schwab's customers depend on RescueWare-generated code for the maintenance of their portfolio. "Relativity had the expertise that we needed," Bilotta says.

The Schwab conversion also came in on time and under budget — a major feat for any conversion, and all the more so for a project undertaken by a new company. In the past two years, Relativity has taken on five other conversion projects for Schwab, including one that speeds up the process by which service reps access customer accounts.

But even that success story isn't enough for Relativity. The Schwab experience was about moving technology from the past to the present. RescueWare's real challenge — its true promise — involves acting as a tool for converting present technology into future technology. Relativity doesn't just want to be a one-shot solution for companies trying to extricate themselves from ailing computer languages. It also wants to be a one-stop resource for companies seeking to keep pace with advances like the wireless Internet. Financial-services outfits like Schwab are starting to enable customers to trade stocks using cell-phones and other mobile devices. Using wireless devices, customers can also access research and account data wherever they happen to be. These days, even the most Webcentric company needs to be moving its operations to a wireless platform. So Relativity has equipped the most recent version of RescueWare to translate any program into a version of WAP (Wireless Application Protocol).

"The way that people deliver information keeps changing, but the way that business decisions are made doesn't change as often," says Wadhwa. "Our value is to make use of existing information, so you don't have to throw away your work and start over."

A Slap to the Forehead

Like many other middle-aged techies, Vivek Wadhwa grew up hacking computers. But he's a lot more engaging than the average hacker. He is an entertaining storyteller. He knows when to sprinkle his company saga with telling anecdotes and when to lay off the technobabble. He offers a simple explanation for why company leaders don't know what their company's software does: "Real men don't do documentation."

Wadhwa, the son of an Indian diplomat, grew up in Australia. He spent his days at the University of Canberra hacking into university and company systems. He then came to the United States and earned an MBA from New York University, and in 1986, he joined the it department of First Boston (which became Credit Suisse First Boston [CSFB] in 1988). There, he encountered a problem that his hacking experience hadn't prepared him for. First Boston needed to upgrade its massive information systems. The bank wanted to be able to calculate trades 24 hours a day, 7 days a week — instead of leaving such calculations "pending" until batch computer runs could take place each night. Wadhwa and his team were unable to reconfigure the existing system, and in its place they wrote hundreds of programs from scratch. The new system cost $150 million to create — and still it couldn't carry over any code from the old system.

Wadhwa believed that there was a better way to approach such projects, and he wasn't alone. In 1990, he was asked to join a company that did for other organizations what he had done at First Boston. That company was Seer Technologies, a joint venture between CSFB and IBM. "Customers were demanding a program that would do conversions," he recalls. "I wish I could tell you that I was a visionary, but I wasn't."

But Seer's vision for how to create such a program was limited. So Wadhwa, who had become the company's CTO, began searching for people who might take a broader view of the conversion problem. In 1991, he and Len Erlikh, a former First Boston colleague who had joined him at Seer, hit upon the idea of tapping talent in Russia. (Erlikh, now 42, is a Russian Jew who had immigrated to the United States from the Soviet Union in 1978.) Both men had heard about Terekhov from former students of his who were working as programmers at Seer, and they decided to seek him out.

Terekhov captivated the Americans with his intelligence and resourcefulness. He had built a telecommunications system for the KGB that enabled Soviet officials to establish phone connections that were both clear and secure. Since that system used computer standards that had been created for the U.S. Department of Defense, Terekhov had to develop a primitive version of the translation software that Wadhwa and Erlikh had in mind. "He's like quicksilver," marvels Erlikh. "He's able to respond to change, and he's full of energy." Wadhwa and Erlikh sent Terekhov PCs and modems, along with sample code from the original First Boston project. Then, after working on the problem with his best students for more than two years, Terekhov sent back a formula that seemed to work.

In 1994, after some initial tests, Wadhwa declared victory. But he had fallen victim to his own bravado. Seer tested Terekhov's process on its customers' systems, and the process failed. Terekhov had built a model so specific to the sample First Boston code that when Seer applied it to a customer's 40,000-line program — at First Boston, a typical program was 2,000 lines long — the system choked. Wadhwa lost credibility, and Seer had to return customers' money. "We went from being rock stars to being losers," Wadhwa says.

That moment was like a slap to the forehead. Members of Terekhov's team realized that a translation algorithm that worked on only one kind of software was useless. They revised their approach, but even as they did so, Seer encountered business problems. The company decided to concentrate on proven technologies and existing customers, and it stopped funding Wadhwa's little-understood R&D. Again, just as Wadhwa and Terekhov seemed to be closing in on victory, it was snatched away.

But Wadhwa had choices. He could stay at Seer but abandon the Russians and the dream of automated translation. He could leave Cary, North Carolina (where Seer was headquartered) for Seattle, home of Microsoft, which had offered him a good position in a completely different area of software development. Or he could leave Seer, turn down Microsoft, and set out on his own with the Russians — and with his vision. "I had to make a tough decision: whether to abandon something important to me or to risk what little I had after a big setback," he says. He left Seer and took the unfinished translation technology with him.

Determined to build a company around that technology, he paid the Russians (and his own bills) out of the $240,000 that he and his wife had saved for their children's college education. That company, founded in February 1997, was Relativity Technologies. (The notion that software translation is akin to cracking the atom may seem melodramatic, but it was in homage to Einstein's Theory of Relativity that the company name was chosen.)

Wadhwa realized that he needed to sell a product that met a wide range of needs — from clarifying what a program actually does to offering customized formula conversions. He, Erlikh (who had left Seer to become CTO of Relativity), and the Russians started building more features for RescueWare (as the product had been called since its first beta version, in 1994). They added tools that compile important functions in a program's old code. And they improved the program's ability to remove and translate particular pieces of code, such as a bank's formula for calculating interest on a loan.

Today, RescueWare is starting to live up to its promise. Pol Mac Aonghusa, manager of solutions development and services at IBM, worked with Relativity last year to salvage functions from the ailing computer system of a large European bank. The system contained about 20,000 functions, or "rules," and before it could be translated, many of those rules had to be recast for new business needs. Mac Aonghusa was impressed by RescueWare's ability to customize conversions without losing any functions in the process — something that the Relativity product hadn't been able to do two years previously. "RescueWare is ready for prime time," he says. "It actually adds value by allowing companies to use their old software in ways they haven't thought of before."

Small Company, Huge Challenges

Relativity's software may be ready for prime time, but what about its business strategy? The market opportunity is huge — but the company is still tiny, with revenues that came to just $10 million in 2000. The needs of giant organizations like commercial banks and the Air Force are clear — but how Relativity will meet those needs productively and profitably remains open to question.

Some days, the company's Russian-based expertise makes Relativity seem like an organization with a Russian brain and an American face — like two teams working across different cultures, across different spoken and machine languages, and across half the globe. At Terekhov's telecom company, 65 of his 160 employees work full-time on Relativity projects. In the United States, another 10 Russians work full-time out of Relativity's North Carolina headquarters. Sasha Aprelev, for example, spent six months working with National City Corp., a Cleveland-based bank that operates in five states. A 23-year-old hotshot who started working for Terekhov at age 13, he helped the bank move private-client accounts to the Web so that tellers could access information faster.

Given that an estimated 200 billion lines of COBOL code are now in use worldwide, and that at least two-thirds of corporate data now resides in mainframe computers (rather than on servers), the prospects for selling RescueWare seem bright. Relativity already has 40 customers. But for now, the company is sitting on a lot of overhead — and banking on future sales. It has 200 employees (including the Russian programmers), 10 sales offices, and a huge bill for phone calls to St. Petersburg. Sales per employee in 2000 came to only $50,000. Even with projected 2001 revenues of $20 million, sales per employee would be just $100,000.

And Relativity's world is a moving target. Even as the company perfects version 6.0 — which can translate old programs (or even newer programs, such as Visual Basic) into J2EE (the latest version of Java), XML, and Intel's Itanium platform — another crew is working on version 7.0, which will serve the next generation of technology platforms.

Relativity is trying to act on lessons that Wadhwa and Erlikh learned during their days at First Boston and at Seer Technologies. Lesson one: Automate as much as possible. Any change made to a version of RescueWare becomes immediately accessible to all of Relativity's programmers. And every night, a computer automatically tests the software to ensure that there are no new glitches.

Another important lesson: Don't rush growth. For all of his enthusiasm about his company's technical achievements, Wadhwa understands the challenge of building a business. "Had I not been through the Seer crisis, Relativity would be a public company right now, and we'd be sweating," he says. "We're not an overnight success story, and we don't want to be. We've had nine years of struggle, and we're still working at it."

Rekha Balu (rbalu@fastcompany.com) is a Fast Company senior writer. She still has files on her office computer in WordPerfect 2.4. Learn more about Relativity Technologies on the Web (www.relativity.com).

Add New Comment

0 Comments