Horses and Cars in Online Advertising

An old joke told in nerdy circles is that you can tell IBM didn't invent the car because IBM would have built the car with a saddle, stirrups, and reins to be backwards-compatible with the horse. (My great-uncle might have appreciated that; his first test drive of a Model T ended with him crashing the car into a house while shouting "Whoa!" with increasing urgency and emphasis.)

By contrast, Henry Ford famously quipped, "If I had asked my customers what they wanted, they would have said a faster horse." Ford focused not on stated need but on latent need. In doing so, he solved the actual problem of getting quickly from place to place, at whim, possibly accompanied by friends or baggage, and at a cost within reach of his target market.

In online advertising, there is a lot of faster-horse building and buying going on, a lot of work putting reins and saddles onto cars, and relatively little actual car building and buying. Online advertising offers vastly different opportunities and economies relative to physical media, yet it is often bought and sold in the same way.

Old media is bought in large blocks that represent significant risk to the advertiser. For example, a :30 TV spot run nationally would cost hundreds of thousands of dollars and can't be changed in flight. A media team recommending such an expense as part of a campaign would need significant research to be taken seriously.

Real Time, Real Tuning

But a national online display campaign can be monitored in real time and tuned continuously. Even the first few hundred dollars of spend can generate actionable results that suggest ways to optimize the campaign. In this situation, it becomes less important to guess beforehand how to optimally target a campaign and support that guess with research. Instead, you want fast experimentation across a diverse set of test buys ("cells"), well-defined metrics, and a method to quickly and continuously move budget away from the worst cells and toward the best.

Instead of an automated performance-seeking approach, online campaign targeting has evolved to look similar to other types of media. For example, contextual targeting satisfies buyers accustomed to running ads in topic-focused magazines. Demographic targeting and geographic targeting satisfy buyers accustomed to those forms of targeting. Behavioral targeting, once a new concept, could have evolved in a number of directions, but the reins-and-saddle approach dominated in the early years. Instead of having an ad server automatically look at available information about a user and select the best particular ad, users were put into segments (e.g., "minivan buyers") that were sold to advertisers, much as demographic segments (e.g., "money and brains") had been sold previously.

When Behavioral Meets Contexual

In the early days of behavioral targeting, the industry took pains to explain how behavioral targeting is better than contextual and other forms of targeting. Online, where decisions can be made in real time, there's no point in religious arguments about the one best type of information to use to optimize a campaign -- all information is potentially useful.

Consider the table below, which shows response rates across four test cells of two contextual topics and two behavioral groups. Targeting based on topic alone would suggest targeting Topic 2, since it generates a 0.6 percent response rate (four times the response rate of Topic 1). Targeting Behavioral Group 2 generates the same 0.6 percent response rate. But simultaneously targeting members of Group 2 on sites about Topic 2 generates a 1 percent response rate, an almost 60 percent lift over either contextual or behavioral targeting alone.

Contextual Topics	Behavioral Group 1 (%)	Behavioral Group 2 (%)	Average (%)
Topic 1	0.10	0.20	0.15
Topic 2	0.20	1.00	0.60
Average	0.15	0.60

An advertiser who had explained her objectives to her partners might have been lucky enough to find a partner who could think out of the box and combine multiple forms of targeting in this way. But an advertiser who had only requested behavioral targeting or contextual targeting without explaining her true objective would have missed these gains for sure.

The upshot for advertising customers is that when you work with partners (agencies, ad networks, other consultants), be sure to explain the true objectives without overly constraining how your partners might best deliver it. You just might find another Henry Ford out there.

Keep in mind this quote:

"It is only a question of time when [the car] will become universal...[It] is a sight to awaken every serious-minded horse to an uneasy consideration of his future."
--Hermann Oelrichs, a founding member of the Automobile Club of California, in March 1900