A Big Data Weapon for the Mobile Ad Challenge

The march of mobile is following the usual evolutionary path in media â€" first come the eyeballs and later comes the money. People spend an estimated 10 percent of their media attention staring at their smartphones and tablets, but mobile advertising accounts for a small fraction of 1 percent of total advertising spending.

Yet from a small base, mobile ad spending is expected to grow rapidly. IDC estimates that mobile advertising will surge from $4.5 billion last year to $12 billion by 2015.

Advertisers will follow the online audience to mobile, but will they be doing it blindly Audience measurement and targeting in mobile devices, where apps rule, is way behind the Web.

Many companies are working on overcoming this mobile ad challenge, including the two giants in mobile software technology with their own operating systems and apps stores: Google and Apple. Yet a start-up, Mobilewalla, has an intriguing, Big Data approach to the problem.

“What we’ve done is develop a scientific way to measure app audiences,” said Anindya Datta, a founder and chief executive of Mobilewalla. “It is a very hard problem.”

And an important one. The promise of Internet advertising is that the measurement and targeting of audiences is vastly superior to that in traditional advertising. But in mobile, where Apple and Android applications are the windows for so much viewing, audience tracking is far more difficult. Unlike the Web, where the code â€" so-called cookies â€" for identifying Internet addresses is in browser software, mobile apps lack a similar tracking tool.

“Spray and pray” is how ad industry experts describe most mobile campaigns. That problem could hold back the growth curve of mobile ad spending. “If you can’t target ads, the utility is greatly diminished,” said Karsten Weide, an analyst at IDC.

Mobilewalla’s data scientists and engineers, Mr. Datta said, have created a way to pull and sift through more than two billion consumer reviews on app stores, Twitter messages and Facebook public profiles and posts. It also gets information from app publishers who agree to share user information.

The Mobilewalla team is looking for signals of sentiment and correlations in the data to try to make accurate predictions about the demographic characteristics of people using certain apps, and point to correlations and clusters of users. Its technology is a variant of the “collaborative filtering” techniques used by Amazon and Netflix to make recommendations. It uses machine-learning software to find apps that share similar audiences and then humans to refine the algorithms.

For example, the ESPN app and Univision Sports app were initially in the same cluster as sports apps. But 90 percent of the ESPN users are English speakers and 78 percent of Univision’s are native Spanish speakers, Mr. Datta said. They are not the same demographic group at all. So the software is tweaked accordingly.

The kindred apps can be quirky. For example, Mr. Datta said, his company’s research shows the Bible! app (published by Logos Bible Software) has a significant audience overlap with apps called Slotsjourney (FishStick Games), Slotmania â€" Slot Machine (Playtika LTD), and Slots by Zynga (Zynga). Religious Christians, Mr. Datta said, seem to like to gamble, especially playing slots.

Another: the Drudge Report app overlaps considerably with the Domino’s Pizza app. Republicans, Mr. Datta said, apparently prefer Domino’s pizza. But the Pizza Hut app has a different audience.

“It’s how they are attacking the problem that seems novel,” said Ramesh Jain, a computer scientist at the University of California, Irvine. “They are treating this as a Big Data problem, working with huge metrics of people expressing their sentiments about apps.”

The challenge is magnified by the growth and volatility of mobile apps. Every week, as about 100 new movies and 250 new books are introduced worldwide, about 15,000 smartphone apps are added to apps stores around the globe.

The churn of what’s rising or falling in popularity is accelerated in smartphone apps. The turnover rate among the 100 most popular mobile apps, according to Mobilewalla’s calculations, is 45 percent after 30 days and 85 percent after 90 days â€" double or triple of the churn rates in television or even Web sites.

Thus, a tried-and-true method of measuring audience in television and on the Web â€" observing the viewing behavior of panels of users â€" does not work for mobile apps, Mr. Datta said. The panel approach relies on persistent popularity over weeks or longer. In mobile apps, that kind of stability does not exist.

Mobilewalla demonstrated its technology last month at the American Association of Advertising Agencies convention in New Orleans. And a major mobile-display ad network, InMobi, has been using the Mobilewalla technology for a couple of months. “We use Mobilewalla to look for apps that are trending upward, to find publishers that look promising as new business prospects, and to add more color as we look for user clusters for audience mapping,” said Shrikant Latkar, vice president of global marketing for InMobi.

Mr. Latkar noted that InMobi also uses other technology tools to target audiences, like AdTruth, a software that can recognize what mobile devices people are using. “It all enriches the feed into our data store,” he said.

Mobilewalla is based in Seattle, with teams in Singapore and India. Founded in 2010, the company has been backed by angel investors, including Rajan Anandan, the chief executive of Google in India. Earlier this year, it raised its first round of venture capital, led by Madrona Venture Group in Seattle. The angel and venture funds, a total of $7.2 million, will be used to expand the 26-person start-up as it seeks to build out its business.

Scott Jacobson, a managing director of Madrona, said a former colleague of his from his days at Amazon.com, who is now in the mobile ad business, pointed him to Mobilewalla.

“Everybody is clear about where attention is going â€" to mobile,” Mr. Jacobson said. “But mobile ad networks are very primitive compared with their Web brethren. There is a need to build an ecosystem for effective advertising for measurement and targeting. Mobilewalla is a data company with a novel way to address that hard problem.”