Wanamaker and Response Prediction

John Wanamaker, a Philadelphia department-store magnate in the late 1800s, famously quipped, “Half the money I spend on advertising is wasted; the trouble is I don’t know which half!”

These days, an online advertiser likely wouldn’t agree with the quote. For online campaigns we can measure conversions and other metrics for mid- and post-campaign analysis. It’s not hard to look retrospectively at a campaign to know what worked and what didn’t. But if Wanamaker were alive today and CMO of a large advertiser, he might say, “Half the money I spend on online display advertising is wasted — the trouble is, I only know which half in hindsight.”

Hindsight to Foresight

The way an ad network learns and adapts from events like clicks and conversions actually tells you a lot about that ad network. Let’s say you pick an ad network for an upcoming CPM (define) campaign with performance goals, and you want to target an audience likely to respond to accessories for a game console. You might ask, “What does your network learn while it’s running my campaign, and how will it use those lessons to adapt the way my campaign is served to meet or exceed my goals?” There are really only three answers to the question.

Answer: “Honestly, we’re going to run your campaign exactly the same on day 30 as we did on day one.”

A number of ad networks employ a fixed-campaign strategy based on their guesses about what’s likely to work. There’s really no way for them to adapt their guesswork for any single campaign. One ad network’s gamer segment might be defined as anyone who visited one of the four game sites in our network in the last 60 days. Users will gradually move in and out of the segment as they visit one of the sites and time passes, but the segment definition will stay the same.

What it means for you: When you begin a campaign with this kind of ad network, do the diligence of asking exactly how users qualify for behavioral targeting segments. When your campaign is in flight, it’s really up to you to keep an eye on its performance, and your only real option is to dial the budget up or down.

Answer: “We’ve got a couple guys that look at reports every few days and tweak your campaign.”

This kind of network might notice that it’s more effective to tighten the audience to a 15-day window versus a 60-day one for your campaign. Or that your ads work better when run on certain categories of sites. It all depends on the skill of the analyst running reports on your behalf, the data, the reporting infrastructure at the ad network, and the flexibility of its targeting and behavioral segmentation.

What it means for you: Ask for details on the reports and the kinds of tweaks the ad network will make, how often your campaign will get attention, and how long it will take for performance to level off. You’ll need to explain performance goals clearly so they aim at the right target. And, frankly, complaining will help you get attention.

Answer: “Your campaign is continuously optimized by computers and algorithms.”

A lot of networks say this, and it covers a wide range of actual functionality. Ideally, this kind of ad network has a mathematical model that predicts the likelihood of a particular impression yielding a response to your ads and uses it to select just the best impressions for your campaign. It might learn that super-passionate gamers respond to your ad whenever they see it, casual gamers only respond on weekends on major portals, and there’s a gray area in between that the network can optimize.

What it means for you: It’s even more important to clearly define campaign performance goals, since the network will be optimizing toward the goal round the clock. Ask what kind of optimization the algorithms might make by themselves, as some networks talk about algorithms but only really do creative optimization. Ask to see ROC (define) curves on the network’s past performance; these describe how effective the network is at selecting the best impressions for campaigns.

The chart below shows a network that has found the old 80/20 rule (define) applying in online advertising: out of 100 million impressions analyzed for a campaign, selecting only the best 20 million would yield 80 percent as many responses, giving four times the response rate of a campaign run without the models.

ROC chart for online advertising
click to enlarge

Brave New World

In his 1920s book “Scientific Advertising,” Claude Hopkins extols the virtues of direct mail advertising (the measurable medium of its day), where solid metrics and data make “false theories melt away like snowflakes in the sun.” In 2009, the bottom-line advice is that whenever you can, it’s a win to dial up the sunshine, test more snowflakes, and let some computers help you out!

Join ClickZ Expert Julie Batten for a free Webinar: Should You Outsource Your Online Marketing Services?, Thursday, April 16, 2009, at 1 pm (EDT). Learn why outsourcing your online marketing activities — including search marketing — can help you save money and achieve better results.

Related reading