Advertising via Algorithm

Much of the advertising seen on both desktop and mobile Web sites is not the result of savvy ad buyers picking up the phone and calling publishers, but instead is procured through programmatic advertising, a process by which ads are automatically served up based on a bidding system driven by algorithms. 

According to 2015 data from ZenithOptimedia, global programmatic ad spending grew to $38 billion in 2015, up from $5 billion in 2012, and is projected to account for 60% of all digital display ads this year.

Programmatic advertising is based on gathering and analyzing Internet user data contained in browser cookies, which track a user’s Internet browsing activity, including purchases, interests, sites visited, and media consumed. This data is analyzed and used to create a digital profile of the person, which is then grouped into larger sample sizes of 10,000 or more, to create a statistically relevant profile based on any number of variables, such as gender, age, geographic location, income, and so forth.

Marketers can create algorithms that automatically purchase ads across the Web based on these criteria, instead of having to pay for a specific number of ads, as was the practice with traditional ads.  Furthermore, while a campaign is ongoing, these algorithms can evaluate what is working best, in terms of geographic segmentation, daytimes, audience segments, and publishers, to help marketers narrow their target so they are paying only for highly effective ads.

Targeting algorithms, which find the most relevant potential users of a product or service, are somewhat more mature, since browser-tracking data can clearly illustrate the types of products and services that interest a particular user. However, attribution algorithms for managing the process of assigning credit for the specific element of an ad ( such as color, size, image A vs image B, etc.), can be far more difficult to create and manage.

"In order to build strong targeting, you need to build a really strong attribution model because otherwise, you’re assigning incorrectly credit to stuff that really didn’t make a difference," says Jay Friedman, COO of Goodway Group, a nationwide firm that has worked on digital campaigns with Fortune 500 companies such as Subaru, McDonalds, and General Motors. "That’s the part that’s hard, but Goodway is pretty much over the hump."

Indeed, due to pervasive use of the Internet both at home and on mobile devices, cookies are jam-packed with data that can be used in ad targeting, Friedman explains, noting that some marketers are starting to develop algorithms that can address these myriad characteristics.

"The median is 58; 58 user characteristics attached to any cookie or device ID," Friedman says. "Now if you have 10,000 people who have converted or bought, you can then start to look at where the similarities are among those 10,000 users in all of the data you didn’t buy, but comes along with it."

As Friedman explains, by examining the data contained in each cookie across a large set of users, it is possible to uncover other, unrelated insights that can be then used to inform another ad buy. For example, if a marketer had a set of data consisting of females who purchased makeup and also frequented salons to get their hair done, perhaps a close, automated scan of these users would uncover that this segment over-represents in the purchase of, say, Snickers candy bars. By developing an algorithm that can quickly identify new data patterns, and then targeting new ads based on this data, marketers can make the most of any data they encounter.

Algorithms are being used to do more than just serve up relevant ads online. Ori Stitelman, vice president of data science at Dstillery, a New York-based provider of programmatic advertising and creative services, notes a lot work is being done to link the virtual ad world with the physical world.

"More recently, we’re doing digital intelligence to understand what’s happening in the physical world," Stitelman says, explaining that by using the data captured on smartphones with location-based services turned on, and then "cross-walking" that data with their browsing data from the Web, marketers can now glean more information on consumer preferences and activities across the Web and around the neighborhood, leading to more relevant and better-targeted ads in both the digital and physical worlds.

Still, marketers are the ones deciding the type, and extent,  of algorithmic modeling to use.

"The tech exists to avoid serving ads to you for the thing you just bought on Amazon, regardless of which device you’re using," says Tom Hespos, co-founder and chief media officer for Underscore Marketing LLC, based in New York. However, he adds, "Marketers tend to think of anyone who has ever bought anything from them as ‘their’ customer, and they will often avoid turning down any opportunity to put a message in front of an existing customer."

Ad frequency is another issue that should be addressed by marketers, since there is technology available to manage the process.

"At what point is showing another ad to a user not worth it anymore?" Friedman asks. "Our algorithm just cuts off users once they’ve hit a certain amount, because we’ve deemed it not worth it anymore. But I think that it annoys consumers; ‘you’ve shown me the ad 10 times, and I’m not going to buy.’  It’s beneficial for both marketers and consumers to mind the frequency."

Keith Kirkpatrick is principal of 4K Research & Consulting, LLC, based in Lynbrook, NY.