The increased use of algorithms and automation in digital media buying, the rise of agency “trading desks,” all conjure quite rightly images of the new financial industry, where stock trading is being driven by machines responding to markets on a nano-second scale.
“Man vs. Machine” on Madison Ave. is the new trope, and one that we ourselves will explore at the upcoming OMMA RTB show on May 16 in New York. There is a lot of hand wringing over the loss of the human factor in digital advertising. Why not automate more and to an even higher level in the process, says Ezra Doty, CEO of Think Realtime and former SVP of Global Digital Initiatives at Universal Music. Like mechanized brokers looking for that fleeting opportunity, media buying online can only find new pockets of great opportunity if it can process and interpret massive numbers of impressions in real time to find the right patterns.
Doty's Think Realtime service optimizes purchasing on RTB platforms in a highly automated way. Robots simply can do things humans cannot, he contends. “We took all of the data that has been available from RTB -- first parties and third parties -- and built a system that learns over time and considers all conditions to measure how well you are performing under these conditions for this goal,” he says. If a system like his can see 100,000 ad impressions a second along with 50,000 bid requests, only a mechanized learning and decisioning approach can find the best values. The system is looking at performance based on everything from time of day to context to geographic location and more to determine where and when the most likely conversions are taking place.
While humans are always optimizing against results, and these manually optimized campaigns will always perform better than untouched ones, Doty contends that in the torrent of data now available, it takes automation to detect and exploit opportunities that no one can anticipate and may be fleeting. For instance in one campaign recently the machine was starting to bid high for impression coming between 7 a.m. and 9 a.m. in Southern states. “This was a pocket of opportunity we never would have tested for,” says Doty. “But it turns out the weather was affecting the consumer’s need for the product, and the system detected that opportunity and capitalized on it quickly.”
In fact, he says that under automation, the system is seeking out more selectively where to place the biggest and best bets for conversion. “We find that a good quality unbiased machine learning algorithm will seek out and identify high quality content,” he says. “Our system bids very high - $8, $10 – eCPMs at times when we know that we have got the rate and the user and a good context on a page that is relevant. Out system will learn that users who have this data or who are on these environments will convert well.” The idea is to create a statistical model where the machine is bidding quite high on inventory when the probability of getting the right action is also high.
Think Realtime works with Buy.com, Advance auto Parts and EyeBuyDirect, among others. Doty claims that clients see 2X to 3X the sales volume from this automated system. The company works exclusively on Google display and focuses on click-based attribution.
Following the performance also helps debunk some of the myths of contemporary and automated media buying. The numbers show that context really does matter significantly, Doty finds. Traditional media brands should not fear the RTB machine and post-human techniques, because the algorithms and performance analysis are exposing the real value of context. “There are contexts where I am willing to pay ten times more because of that environment,” he says. “It is a huge driver of performance. There are a lot of misconceptions that you are just buying audience.”
And buyers and planners at agencies shouldn’t worry about being replaced either, Doty assures. While I have heard some in the industry plot the obsolescence by machine of whole departments that now exist at agencies, Doty is more sanguine about the resilience of high-level human thinking. In addition to determining the campaign goals on the front end that an optimization machine executes, there is a wealth of massaged back-end data that requires real brain matter to leverage. “The media buyers are going to be able to look at these insights and share examples of how their consumers are behaving and understand the consumer buying patterns in way they never could before,” he says.
While I am sure there are legions of desktop campaign optimizers who would beg to differ on payday, in Doty’s view, automating optimization will be good for them in the end, too. There will be more jobs and perhaps more job satisfaction from higher-level problem-solving. “It will be much more interesting work than optimizing a campaign and staring at a spreadsheet all day. That can be pretty mind-numbing.”