As part of its first $100 mn technology gamble outside the US, the world’s largest infotech firm is trying to build a parallel Web for voice. If it cracks India’s huge market, the world is waiting.
The call sounds unremarkable enough at first. The number dialed is a regulation 10 digits, tinny music bursts out in welcome, and a menu materializes: press 1 for questions, 2 for announcements, 3 for archives. Thus far, no different from any other automated voice response call.
The first question after pressing 1 is a hint of something new. A farmer named Jaikishan has called in. He sets out his worries about Bt cotton and the effects of the delta endotoxin the Bt bacterium is designed to produce. Jaikishan has precisely one minute to state his concerns. Later, dialing into the same number, he can see if an agriculture expert or even another farmer has answered his questions.
Other farmers can browse through these questions and answers; they can scroll through the list of announcements and upload their own. In a more elaborate version, users can create for themselves a “voice site” with a unique URL-type number and link it to other voice sites.
In the future, they may even be able to search for specific information and navigate back and forth between voice sites. All of which begins to expand the plain-vanilla voice response system into something more delicious: an alternative, Internet-like Web.
This pilot, still running for farmers in 20 villages in Gujarat, is the second to test-drive Spoken Web, IBM India Pvt. Ltd’s plan to create a parallel Web driven by voice and accessed by phone. Mint first wrote about this ambitious enterprise, then called VoiGEN, early last year, soon after its first pilot in Andhra Pradesh in 2008. That pilot ran for eight months and was open to everybody, not just farmers.
Entirely spurred by word of mouth, around 7,000 people used Spoken Web in those eight months, making at least 110,000 calls. Parents recorded babies’ voices so relatives could dial in to listen; some sang songs; electricians and plumbers publicized their services. Most memorably, remembers Amit A. Nanavati, a senior researcher at IBM, when one man put up a matrimonial advertisement for himself, “somebody responded, warning everybody that the guy was a crook”.
A question of scale
That success was, in a way, a proof-of-concept. To hear IBM tell it, a Spoken Web seems now to be almost inevitable, and discussions for commercial uses of such a Web have already begun. “In India, Internet penetration is really low, but mobile phone penetration is growing very rapidly,” says Sougata Mukherjea, a senior manager working on Spoken Web. The Internet reaches fewer than 7% of Indians; mobile phones reach 55.4%. “It’s just a question of using those scales.”
Like a benevolent virus, that word “scale” infects nearly every innovation unfolding in IBM’s research labs in India. It’s a recognition of how to use a daunting challenge—in this case, the sheer numbers involved in any enterprise in India—to build a robustness that can then function with ease anywhere in the world. “You look at a Bharti Airtel’s customer base, say. That sort of scale has never been seen before,” says Manish Gupta, director of IBM Research-India. “That’s why innovation will happen here first, and then go elsewhere.”
Gupta is an IBM research veteran. He joined the company’s lab in New York after completing his PhD in 1992, and he just never left. Most crucially, he worked for five years on the Blue Gene team, tasked with building what was then the fastest supercomputer in the world. That was when he collaborated closely with research engineers in IBM India. “Some of the work they did on performance benchmarks simply blew us away.”
Gupta returned to India two years ago, and he took over as director of IBM’s Indian labs in February. Under his purview are the facility in New Delhi, operating since 1998, a five-year-old centre in Bangalore, and a collaboration with the Indian School of Business (ISB) in Hyderabad, established last year. At these centres, innovation is all there is. “Just last week,” Gupta says, “I invited every employee to submit the most audacious ideas they had and I promised to fund the most adventurous one.”
The open invitation is inspired by an annual practice at IBM’s Almaden research lab in California, and is symbolic of the company’s methodical approach to innovation.
Even audacity is carefully calibrated. Projects go into a “portfolio of horizons”—some near term, lasting a year or less; some middle term; and a few long term, lasting five years or more. Some of the latter are termed, in IBM lexicon, “Big Bets”—projects with an investment of the order of $100 million over five years.
“They have a low probability of success,” says Gupta, “but a high payoff if they do succeed.”
The Mobile Web framework, of which Spoken Web is a part, has become the first Big Bet to be led by an IBM research lab outside the US. It made sense to drive a mobile telephony project from outside the West, says Gopal Pingali, who now heads the collaboration with ISB but worked on Mobile Web when it started in 2008. “Besides, we’d already pioneered Spoken Web by then, so this fit right in.”
Not every innovative project in IBM India is as adventurous as Spoken Web—as unconnected, to explain Gupta’s euphemism, to existing lines of business. But they all latch on, in some way, to the dizzying scale of doing business in the country. There are products that streamline the information inundating call centres and the CVs inundating recruiters.
SNAzzy, an analysis tool, maps patterns in cellular networks, telling telecom companies about the behaviour of their millions of customers, which proves invaluable in designing subscriber packages or offers.
Clean, another tool, works across databases to regularize information—something that is especially pertinent to India where, as Gupta says, “the word ‘marg’ means the same thing as the word ‘road’.” Previous such tools, developed abroad, performed poorly when they applied their 2,000-odd rules to Indian information.
Clean patiently prioritizes those rules and applies them hierarchically for far better results; IBM will merely reveal that accuracy levels went “from fairly low to fairly high”.
These products, however, have obvious applications and, therefore, a ready pool of customers; a project like Spoken Web is far more nebulous. Its adoption will not just depend on how well IBM engineers it, but also on how willing people are to populate it with voice content and use it effectively. IBM has no control over that.
Gupta mentions that if Spoken Web finds traction in India, it could translate well into South America or Africa, where the statistics of Internet and mobile phone penetration are comparable. But to find that traction, Spoken Web will have to become more flexible, including exactly the sort of functions that IBM’s scientists are struggling to build: a voice-controlled browser, for instance, or a robust search mechanism.
It isn’t easy, Mukherjea and Nanavati admit; search, in particular, is proving to be a difficult problem to crack. But Spoken Web is, for IBM, a rare project that puts its scientists into direct contact with large swathes of lay users, and that in itself is proving to be substantial inspiration. “We got dozens of messages from the users of Spoken Web in the Andhra Pradesh pilot, saying: ‘Thanks for this. We’ve never seen anything like it’,” Nanavati remembers. “I’d get goose bumps just listening to them.”
- Hindustan Times, Mint
July 2, 2010