Skip to main content

UW-Madison alum masterminds strategy of Internet ad placement

January 29, 2014 By David Tenenbaum

Increasingly, advertising is the currency of the Internet. Those rectangles that compete to catch your eye fund many of those free services that dominate the Web.

Abhinav Gupta

Abhinav Gupta

A few years ago, banner ads were haphazardly painted across websites. Now, that “spray and pray” technique has been replaced by a data-driven, computerized marketplace that auctions off advertising space in the eye blink that elapses between the click that requests a page and the time it is delivered to the user’s screen.

That moment is just long enough for firms like Rocket Fuel, co-founded by University of Wisconsin–Madison graduate Abhinav Gupta, to evaluate a mountain of data and decide what to bid for a particular advertising spot.

Rocket Fuel went public in September, with a market valuation of about $1.6 billion. The firm has about 600 employees, with headquarters in Redwood City, Calif., and offices in San Francisco, New York, Chicago, Los Angeles, Paris, London and elsewhere.

Rocket Fuel uses a “big data” strategy to find the optimum placement for ads, says Gupta, but it can start from limited knowledge. “We always have some information about the world. Some is very general: for example, people who visit websites about babies and kids are likely to purchase online.”

That information becomes the starting point for ad placement, Gupta says. Within one-tenth of a second, Rocket Fuel and some of its competitors take what they know about the website, the time of day, the user’s location and web-use history, and decide which of their advertising customers would be interested in paying for that spot, and how much.

Gupta stresses Rocket Fuel knows little about the people who have Rocket Fuel cookies on their hard drives. “We don’t have access to any personal identifying information, no information about email address, nothing. The only thing we know is that we believe this anonymous user, identified by a random number in a cookie, is more likely to respond to a particular campaign at this point in time.”

If Rocket Fuel wins the auction, it places the ad. Then, using techniques Gupta learned in the UW–Madison computer sciences department and honed at Oracle and Yahoo, the company’s software goes through a process called machine learning.

“At the start, we may not have much idea about the market, but in a few days or weeks, the system starts learning by itself about what works for this campaign,” Gupta says. “This is our competitive advantage, leveraging machine learning and artificial intelligence, on petabytes of data.”

One petabyte is one million gigabytes.

“Working with big data is what I learned in school. Wisconsin has an amazing computer science department, and I worked with an amazing group of professors and colleagues to learn about databases and big data. Wisconsin was researching on big data when it was still small.”

Abhinav Gupta

The exchange of data continues after the ad appears. Advertisers evaluate success of a marketing campaign based on the number of positive outcomes (such as a purchase) for every dollar spent. Hence, advertisers share that data with Rocket Fuel, Gupta says. “If we learn from the confirmation page that the user has bought something, that informs our software: ‘Hey, good job, go find more users like this one!’ That’s machine learning: Look at each positive outcome, and do more of whatever produced that result.”

Gupta, Rocket Fuel’s vice-president for engineering, focuses on the core software that is used in all campaigns.

“Working with big data is what I learned in school,” he says. “Wisconsin has an amazing computer science department, and I worked with an amazing group of professors and colleagues to learn about databases and big data. Wisconsin was researching on big data when it was still small .”

But the field has changed since Gupta got his M.S. in 1997. “At the time, big data was a few thousand gigabytes, now it’s 1,000 times bigger, but I have been able to apply what I learned.”

“Abhinav was super smart, super sharp,” says Jignesh Patel, a professor of computer science at UW–Madison, who was a fellow student in computer science. “He has a very good head, and that uncommon common sense. He knows not only how to solve a particular problem, but also how to make it fit into the broader system. It’s a very delicate balance to find the right technology, and to pick the right people, and motivate them. He’s a true entrepreneur.”

Big data and data-driven decisions are rapidly gaining prominence in fields like medicine, physics, finance and manufacturing, Patel says.

“In the old days, if an enterprise wanted to bring out a new product, they would trust a human to make that decision,” Patel says. “The decision would be based on data, and a lot of intuition. That’s changing fast, and Abhinav is at the center of the revolution in big data.”