As recession-racked companies search for ways to cut costs, some are rediscovering automated innovation. In the early 2000s, auto-innovation was trumpeted as the Next Big Thing. Instead of relying on engineers and designers, HAL-like computers would create goods on their own by exhaustively combining bits and pieces of previously successful products.
Hewlett-Packard (HPQ) and Pfizer (PFE) were among the early practitioners. Dubbing its software "genetic programming," HP set up GP Lab to experiment with computer code that could analyze the "genes" of earlier inventions and point to evolutionary advances. Pfizer equipped some 1,500 scientists with software that could identify chemicals that might turn into drugs.
Individual researchers also tinkered with such self- propelled programs. In 2005 auto-innovation software created by John Koza, a programmer from Los Altos, Calif., earned the first U.S. patent ever awarded to a nonhuman.
But these initial efforts in auto-innovation didn't pay off. The patent won by Koza's program, a microchip that controls a machine's operations, never turned into a commercial product. At Pfizer, auto-innovation was blamed for leading researchers down dead ends and bloating the budget. And HP found that while GP Lab suggested some dazzling new designs for data-storage products, it failed to turn up any that customers were likely to buy.
As quickly as auto-innovation swept through research and development departments, it was gone. HP killed GP Lab in 2004, four years after it was first unveiled. "We found that it worked to make theoretically good designs," says Jaap Suermondt, director of HP's Business Optimization Lab, who oversaw the project. "But it became a hammer in search of a nail."
Now that companies are under intense pressure to get more out of every dollar, automated innovation is making a comeback. Rather than being used to create products, however, it's turning into an efficiency tool to improve business processes.
Dozens of software companies are using algorithms once intended for product development to help corporations pinpoint ways to reduce spending. San Diego's Natural Selection, a 16-year-old company that created the program used by Pfizer to try to auto-invent drugs, is helping clients streamline delivery routes and retrofit facilities. Among its recent customers: General Electric (GE) and the U.S. Air Force. Some of the original corporate participants in auto-innovation are back at it, too, including Pfizer and HP.
HP's adventures (or misadventures) in particular show how ideas that bombed at first can become valuable when given a second chance. "Successful innovations are often built on the backs of failed ones," notes Scott D. Anthony, president of business consultancy Innosight and author of a just-published book, The Silver Lining: An Innovation Playbook for Uncertain Times. "It makes sense to make it a regular practice to go back and see what pieces of rejected ideas might offer important tools if they can be applied in new ways."
Like researchers at 3M (MMM) and Google (GOOG), staff scientists at HP Labs are urged to spend a chunk of their workweek on self-initiated projects. Evan Kirshenbaum, a computer programmer who has worked at HP since 1989 and holds more than 20 programming patents, began in early 1998 using his spare time on auto innovation writing code to combine and recombine snippets of ideas to discover new ones.
Here's how it works: An engineer takes information, such as design elements of laptops that sold well, then plugs this into the program. The program analyzes the data and figures out how to mix and match these items, like building blocks, to come up with something new.
Kirshenbaum, 44, who has a bachelor's degree in linguistics and a master's in computer science from Stanford University, worked anonymously on his side project for three years; his cubicle at HP's Page Mill Road campus in Palo Alto, Calif., is not far from the preserved former offices of company founders Bill Hewlett and David Packard. In 2000 he presented some of his research at a software engineering conference.
REVIVED IN NO TIME
One rung up the corporate ladder, Suermondt took notice in 2001 and decided to test the software on an existing product line. Kirshenbaum's algorithm, Suermondt thought, could help HP design data-storage systems customized to each corporate client and possibly lead to a new product. Kirshenbaum, excited about the software's prospects, wrote on his Web site that he was working on a "really neat genetic programming system." But market research soon showed clients couldn't care less. "Our existing customers wanted simpler-to-use, and really suboptimal, designs," Suermondt says.
Although GP Lab had cost almost nothing—Kirshenbaum's support staff consisted of a few interns—Suermondt concluded it didn't have an immediate payoff and pulled the plug. Kirshenbaum didn't entirely give up, however. Believing that auto-innovation would have a second life, he carefully documented everything about the project and stored the source code and the instructions on how to use it on several hard drives. His meticulous recordkeeping turned out to be fortuitous.
In 2007, even before the recession hit, HP Chief Executive Mark Hurd ordered cuts in operations spending. As sales and net income dropped in 2008, he told managers to find ways to slash their budgets further. So Suermondt began looking for new cost-cutting tools. He realized that auto-innovation software could combine financial information scoured from databases and come up with new ways to predict sales and manage the supply chain.
He recruited Kirshenbaum to bring GP Lab back to life. Because it was so well documented, Suermondt recalls, "it was up and running in days."
Today, HP is using auto-innovation to forecast manufacturing and shipping needs based on predicted sales growth and outfit its worldwide offices and factories as cheaply as possible. The program sorts through such information as, say, the number of computers HP has sold in each market and current spending on PCs in specific industries such as oil and gas. It then combines the data in thousands of ways to determine when and where to buy and ship supplies such as hard drives or products such as notebooks. "We can plug in 100 variables, and the software can consider nearly every possible combination," says Suermondt. "You couldn't do that in a human lifetime, but the algorithm can in pretty much real time."
HP hasn't estimated how much money the program will save. Meantime, the $118.4 billion company is spending little more than Kirshenbaum's salary on the revived lab.
Improving a corporate supply chain isn't nearly as sexy as inventing an iPhone killer. But Stephen Corbett, a partner in McKinsey & Co.'s operations practice, says that in the recessionary economy of 2009, saving money is just as valuable as generating revenue—and companies often can slash supply costs by 20%. "Finding new ways to identify those opportunities," he adds, "seems a logical place for companies in general to extend and focus on these days."
Plus, the system might pay off in other ways. Suermondt says management wants to commercialize the software in 5 to 10 years. Auto-innovation is a key element of a larger HP Labs initiative called Analytics for Operations, which is one of HP's "big bets," or inventions that it considers most likely to spawn a new market.
Kirshenbaum feels vindicated. "I always believed that it was a promising technique and a good system," he says. "And I fully expected that it would prove useful down the road."
Business Exchange: Read, save, and add content on BW's new Web 2.0 topic network
Invention Kit for the Home
What's next in automated innovation? Invention software similar to HP's GP Lab that's sold directly to consumers. The computer program could enable individuals to "crowdsource" viable product designs themselves. So argues Robert Plotkin, a Burlington (Mass.) patent lawyer with an MIT degree in computer programming, in his new book, The Genie in the Machine: How Computer-Automated Inventing is Revolutionizing Law & Business.