A math major with a specialty in abstract mathematical concepts like combinatorics, Ben Liang envisioned a future as a financial analyst—outside of academia, a natural career path for number crunchers and the same route his dad had taken. So after college, the 26-year-old Connecticut native entered Bank of America’s two-year training program, with the aim of going to work in the global capital markets and investment banking division.
There was just one problem: Wall Street was really boring, even for somebody who loves numbers.
“You’re basically a human computer, just chugging through data,” Liang says.
A year into his training at BofA, he started looking into other, sexier avocations in which he might flaunt his quantitative talents. That search would ultimately land him a gig as a campaign analyst at ChoiceStream, the Boston ad-technology company specializing in audience targeting, where his job of optimization engineer is figuring out how to maximize the potential of complex algorithms that seek to show ads to consumers who are most likely to buy based on their online behavior.
In short, he’s helping brands figure out who their best prospects are, making him increasingly one of the most important players in the entire advertising industry.
The number of ad campaigns based on algorithms doubled last year versus 2011. In the coming years, they are expected to account for nearly half of all campaigns, according to Forrester Research. Meanwhile, this year the world’s catalog of digital data is expected to reach some 2.7 zettabytes—an amount of information so large it would take 700 billion discs to store it all.
By itself, all that data is useless, naturally. That’s why numbers people are in such high demand—not just any numbers people, but creative quants who can keep pushing online ads to the next, more sophisticated level. They are drawn to advertising not just for the job prospects but also for the creative challenge—and they’ll take a considerable pay cut versus similar jobs in financial services.
Liang himself took a 30 percent hit when he moved from banking into advertising—and he’s hardly alone in having made the leap. His colleagues at ChoiceStream include a data analyst with a Ph.D. in neurology from MIT, another with an M.S. in biostatistics from Harvard and a software engineer with a physics degree from Cambridge. Many came from fields such as finance and biomedical statistics. Meanwhile, computational advertising has become a field of study at universities including Stanford, joining majors like computational finance and financial engineering that emerged in the ’90s. Last year, Columbia University announced the launch of its Institute for Data Sciences and Engineering.
“They might perceive [advertising] as not as dry as banking, not as bureaucratic, not as governed by regulatory issues. They’re looking for industries where they feel they might have more opportunity,” says Rita Raz, a recruiter at Analytic Recruiting. “An ad agency might be more fun, or perceived as more fun.”
Michael Benisch, an analyst at Rocket Fuel with a Ph.D. in computer science, adds that many computer experts are looking to the ad space because the sheer amount of data makes it “the ultimate test.” For quants, advertising is like the Wild West of quantitative analysis: the horizon is limitless, and the rules are always changing.
Madison Avenue and Wall Street have more in common than you may think. Referencing the real-time bidding wars for digital display ads, Christopher Steiner, author of Automate This: How Algorithms Came to Rule Our World, says, “I would say advertising is the closest thing we have to high-frequency trading on Wall Street—it’s almost the exact same thing.” Meanwhile, ChoiceStream’s Liang compares his company to a hedge fund and himself to a portfolio manager. “We use our algorithms to figure out which people are most useful to our clients, which is basically what a hedge fund does,” he explains.
When a brand or agency approaches a company like ChoiceStream, it is looking to either tap a specific market or, in certain cases, determine what its market might be. The ability to do this essentially comes down to two factors: the quality of the data set and the algorithm itself. An algorithm is, at its most basic, a set of instructions for performing some kind of calculation and achieving an ideal result. Algorithms can be used for calculation, data processing and automated reasoning. On Wall Street, such methods were first used in the ’70s, when savvy investors began using them to price stocks and bonds. In advertising, algorithms grew in importance as the amount of data on consumer behavior swelled.
How exactly does a mathematical equation tell a marketer how best to market its minivan to a suburban New Jersey mother of three? Quants compose algorithms that can take in thousands and thousands of inputs (in this case, data) and spit out hundreds of possible solutions (who to target an ad to). As Steiner writes in his book, “Algorithms can be looked at as giant decision trees composed of one binary decision after another.”
Last year, the shopping site Zappos approached ChoiceStream about targeting consumers based not only on their age, location and other typical demographic data, but also what the weather was like. ChoiceStream designed a display ad that linked a three-day, location-based weather forecast with climate-appropriate merchandise that also took into account factors like age and gender. If a forecast called for showers, for example, an ad might display rain boots or slickers. In the bidding wars for online ad space, algorithms also determined how much Zappos might be willing to pay for ad space. Choice-Stream built the ad in three days.
The project was a test for Zappos, and of ChoiceStream’s own capabilities. The cost of a conversion—in this case, how much money must be invested on average to get a new customer to buy something on the site—declined significantly. Not surprisingly, Zappos decided to continue with the campaign. “The conversion rate was strong enough to realize a nice healthy return on investment,” says Lisa Archambault, manager of display marketing at Zappos. “[The ad] was much more specific to the consumer, which is really our goal.”
Meanwhile, Rocket Fuel’s data-modeling campaign last year for auto brand BMW was such a success that it was credited with helping boost North American second-quarter sales by 40 percent. Similarly, a campaign for tire manufacturer Bridgestone’s retail sites using Rocket Fuel’s Audience Booster and Insights Booster products contributed to a 45 percent bump in store sales after specific markets were targeted. “[Rocket Fuel] has an ability to drill down to something that’s very niche,” says Steve Parker Jr., co-founder of the digital ad agency Levelwing, which brought Rocket Fuel onto the Bridgestone account.
While brands like BMW and Zappos claim success from data-rich, algorithm-driven online ads, there’s no shortage of skepticism out there. VC firms have poured millions into the ad-tech sector, as have scores of companies with dubious claims of super-precise targeting. Algorithms work for Google, but still at issue are the real prospects of delivering the perfect ad to the right person at the right time.
And yet, individuals like Liang and Benisch are tapped to do just that, drilling down to unearth the perfect ad strategy. They are charged with figuring out what to input, monitoring the number crunching and analyzing what the computers ultimately spit out, as well as which algorithms to employ.
Liang likens his role to the robotic vacuum cleaner Roomba: When the process hits a wall, his job is to set it in the right direction. Often that means spotting irregularities in what an algorithm produces, and figuring out whether a formula might need to be tweaked for more accuracy.
Sometimes results can be counterintuitive. In one instance, a fast-food chain hired ChoiceStream to market a new coffee product. It turned out the ad scored particularly well not with caffeine addicts but with Jeep owners. Analysis determined that, indeed, owners of that nameplate were also ideal prospects for the coffee. As a result, the chain launched a campaign targeted to outdoorsy types—the very people who might own a Jeep.
Players in the space like ChoiceStream and Rocket Fuel point not only to their superior data sets but also to their highly effective algorithms. “The secret for success is the algorithm—your ability to unlock the power of the data,” says ChoiceStream COO Eric Bosco.
From coast to coast, each company promises an algorithm that will do slightly different things. ChoiceStream was founded in 2000 but only broke into audience targeting in 2011 with its product Crunch. The company boasts that Crunch uses “genetic programming,” meaning that its algorithms borrow from those used in behavioral genetics.
Rocket Fuel, meanwhile, markets itself as “advertising that learns” or “artificial intelligence,” though that is really the essence of what all data models do. EXelate, another company specializing in data modeling, sells its analysis directly to brands and ad agencies, while companies like Rocket Fuel and ChoiceStream are more involved in the overall advertising process. ChoiceStream has an in-house creative team to handle chores such as the design of ad-based polls or entire campaigns like the Zappos weather module.
Their formulas are closely guarded secrets, of course. Generally speaking, they are learning algorithms with the capability of analyzing data and recognizing patterns. (See formula in the graphic below.)
Benisch, who has been involved in computer programming since he was 10, holds a Ph.D. in computer science from Carnegie Mellon and started his own hedge fund in 2009 while still a student, points to the rise in sophistication of algorithms, which today can handle “billions of complex decision points per day, each within a few milliseconds.”
The 2012 presidential election brought the powers of the empirical into sharp focus. Statistician Nate Silver of The New York Times’ FiveThirtyEight blog toppled pundits and their predictions based on theory, history and intuition by correctly predicting the outcome of the electoral vote, including in nine swing states. It was a turning point in making the case that statistical analysis holds a better chance at determining the outcome of an election than, say, Chris Matthews. Obama emerged victorious, and so did empiricists.
Kevin Lyons, svp of analytics at eXelate, notes that Silver’s political calculus is essentially the same math used to figure out which consumers to target with an online ad. In marketing, however, the statistical and the intuitive are not at odds; rather, they have become complementary.
Eric Bosco of ChoiceStream notes that the traditional creative side of advertising sets the stage for fancy math to do its work. It narrows the field of possibilities, so algorithms can be more effective. For example, he points to one insurance company whose marketing campaign focused on younger consumers and sought to make the brand “hip” among the demo. The message of the campaign informed ChoiceStream’s analysis.
As Bosco says, “We take that initial intuition and convert it into something powerful.”
Perhaps powerful enough to remake advertising as we know it.
PHOTO: BEN SHAUL; ALGORITHM WRITER: JENNIFER O’DONNELL